Machine learning approaches for predicting protein-ligand binding sites from sequence data

Proteins, composed of amino acids, are crucial for a wide range of biological functions. Proteins have various interaction sites, one of which is the protein-ligand binding site, essential for molecular interactions and biochemical reactions. These sites enable proteins to bind with other molecules, facilitating key biological functions. Accurate prediction of these binding sites is pivotal in computational drug discovery, helping to identify therapeutic targets and facilitate treatment development. Machine learning has made significant contributions to this field by improving the prediction of protein-ligand interactions. This paper reviews studies that use machine learning to predict protein-ligand binding sites from sequence data, focusing on recent advancements. The review examines various embedding methods and machine learning architectures, addressing current challenges and the ongoing debates in the field. Additionally, research gaps in the existing literature are highlighted, and potential future directions for advancing the field are discussed. This study provides a thorough overview of sequence-based approaches for predicting protein-ligand binding sites, offering insights into the current state of research and future possibilities.

BalancerGNN: Balancer Graph Neural Networks for imbalanced datasets: A case study on fraud detection

Fraud detection for imbalanced datasets is challenging due to machine learning models inclination to learn the majority class. Imbalance in fraud detection datasets affects how graphs are built, an important step in many Graph Neural Networks (GNNs). In this paper, we introduce our BalancerGNN framework to tackle with imbalanced datasets and show its effectiveness on fraud detection. Our framework has three major components: (i) node construction with feature representations, (ii) graph construction using balanced neighbor sampling, and (iii) GNN training using balanced training batches leveraging a custom loss function with multiple components. For node construction, we have introduced (i) Graph-based Variable Clustering (GVC) to optimize feature selection and remove redundancies by analyzing multi-collinearity and (ii) Encoder-Decoder based Dimensionality Reduction (EDDR) using transformer-based techniques to reduce feature dimensions while keeping important information intact about textual embeddings. Our experiments on Medicare, Equifax, IEEE, and auto insurance fraud datasets highlight the importance of node construction with features representations. BalancerGNN trained with balanced batches consistently outperforms other methods, showing strong abilities in identifying fraud cases, with sensitivity rates ranging from 72.87% to 81.23% across datasets while balancing specificity. Additionally, BalancerGNN achieves impressive accuracy rates, ranging from 73.99% to 94.28%. These outcomes underscore the crucial role of graph representation and neighbor sampling techniques in optimizing BalancerGNN for fraud detection models in real-world applications.

VC dimension of Graph Neural Networks with Pfaffian activation functions

Graph Neural Networks (GNNs) have emerged in recent years as a powerful tool to learn tasks across a wide range of graph domains in a data-driven fashion. Based on a message passing mechanism, GNNs have gained increasing popularity due to their intuitive formulation, closely linked to the Weisfeiler-Lehman (WL) test for graph isomorphism, to which they were demonstrated to be equivalent (Morris et al., 2019 and Xu et al., 2019). From a theoretical point of view, GNNs have been shown to be universal approximators, and their generalization capability - related to the Vapnik Chervonekis (VC) dimension (Scarselli et al., 2018) - has recently been investigated for GNNs with piecewise polynomial activation functions (Morris et al., 2023). The aim of our work is to extend this analysis on the VC dimension of GNNs to other commonly used activation functions, such as the sigmoid and hyperbolic tangent, using the framework of Pfaffian function theory. Bounds are provided with respect to the architecture parameters (depth, number of neurons, input size) as well as with respect to the number of colors resulting from the 1-WL test applied on the graph domain. The theoretical analysis is supported by a preliminary experimental study.

Weighted Echo State Graph Neural Networks Based on Robust and Epitaxial Film Memristors

Hardware system customized toward the demands of graph neural network learning would promote efficiency and strong temporal processing for graph-structured data. However, most amorphous/polycrystalline oxides-based memristors commonly have unstable conductance regulation due to random growth of conductive filaments. And graph neural networks based on robust and epitaxial film memristors can especially improve energy efficiency due to their high endurance and ultra-low power consumption. Here, robust and epitaxial Gd: HfO2-based film memristors are reported and construct a weighted echo state graph neural network (WESGNN). Benefiting from the optimized epitaxial films, the high switching speed (20 ns), low energy consumption (2.07 fJ), multi-value storage (4 bits), and high endurance (109) outperform most memristors. Notably, thanks to the appropriately dispersed conductance distribution (standard deviation = 7.68 nS), the WESGNN finely regulates the relative weights of input nodes and recursive matrix to realize state-of-the-art performance using the MUTAG and COLLAB datasets for graph classification tasks. Overall, robust and epitaxial film memristors offer nanoscale scalability, high reliability, and low energy consumption, making them energy-efficient hardware solutions for graph learning applications.

Redundancy is Not What You Need: An Embedding Fusion Graph Auto-Encoder for Self-Supervised Graph Representation Learning

Attribute graphs are a crucial data structure for graph communities. However, the presence of redundancy and noise in the attribute graph can impair the aggregation effect of integrating two different heterogeneous distributions of attribute and structural features, resulting in inconsistent and distorted data that ultimately compromises the accuracy and reliability of attribute graph learning. For instance, redundant or irrelevant attributes can result in overfitting, while noisy attributes can lead to underfitting. Similarly, redundant or noisy structural features can affect the accuracy of graph representations, making it challenging to distinguish between different nodes or communities. To address these issues, we propose the embedded fusion graph auto-encoder framework for self-supervised learning (SSL), which leverages multitask learning to fuse node features across different tasks to reduce redundancy. The embedding fusion graph auto-encoder (EFGAE) framework comprises two phases: pretraining (PT) and downstream task learning (DTL). During the PT phase, EFGAE uses a graph auto-encoder (GAE) based on adversarial contrastive learning to learn structural and attribute embeddings separately and then fuses these embeddings to obtain a representation of the entire graph. During the DTL phase, we introduce an adaptive graph convolutional network (AGCN), which is applied to graph neural network (GNN) classifiers to enhance recognition for downstream tasks. The experimental results demonstrate that our approach outperforms state-of-the-art (SOTA) techniques in terms of accuracy, generalization ability, and robustness.

Transformer graph variational autoencoder for generative molecular design

In the field of drug discovery, the generation of new molecules with desirable properties remains a critical challenge. Traditional methods often rely on simplified molecular input line entry system representations for molecular input data, which can limit the diversity and novelty of generated molecules. To address this, we present the transformer graph variational autoencoder (TGVAE), an innovative AI model that employs molecular graphs as input data, thus capturing the complex structural relationships within molecules more effectively than string models. To enhance molecular generation capabilities, TGVAE combines a transformer, graph neural network (GNN), and VAE. Additionally, we address common issues like over-smoothing in training GNNs and posterior collapse in VAEs to ensure robust training and improve the generation of chemically valid and diverse molecular structures. Our results demonstrate that TGVAE outperforms existing approaches, generating a larger collection of diverse molecules and discovering structures that were previously unexplored. This advancement not only brings more possibilities for drug discovery but also sets a new level for the use of AI in molecular generation.

Graph Neural Network Learning on the Pediatric Structural Connectome

PURPOSE

Sex classification is a major benchmark of previous work in learning on the structural connectome, a naturally occurring brain graph that has proven useful for studying cognitive function and impairment. While graph neural networks (GNNs), specifically graph convolutional networks (GCNs), have gained popularity lately for their effectiveness in learning on graph data, achieving strong performance in adult sex classification tasks, their application to pediatric populations remains unexplored. We seek to characterize the capacity for GNN models to learn connectomic patterns on pediatric data through an exploration of training techniques and architectural design choices.

METHODS

Two datasets comprising an adult BRIGHT dataset (N = 147 Hodgkin's lymphoma survivors and N = 162 age similar controls) and a pediatric Human Connectome Project in Development (HCP-D) dataset (N = 135 healthy subjects) were utilized. Two GNN models (GCN simple and GCN residual), a deep neural network (multi-layer perceptron), and two standard machine learning models (random forest and support vector machine) were trained. Architecture exploration experiments were conducted to evaluate the impact of network depth, pooling techniques, and skip connections on the ability of GNN models to capture connectomic patterns. Models were assessed across a range of metrics including accuracy, AUC score, and adversarial robustness.

RESULTS

GNNs outperformed other models across both populations. Notably, adult GNN models achieved 85.1% accuracy in sex classification on unseen adult participants, consistent with prior studies. The extension of the adult models to the pediatric dataset and training on the smaller pediatric dataset were sub-optimal in their performance. Using adult data to augment pediatric models, the best GNN achieved comparable accuracy across unseen pediatric (83.0%) and adult (81.3%) participants. Adversarial sensitivity experiments showed that the simple GCN remained the most robust to perturbations, followed by the multi-layer perceptron and the residual GCN.

CONCLUSIONS

These findings underscore the potential of GNNs in advancing our understanding of sex-specific neurological development and disorders and highlight the importance of data augmentation in overcoming challenges associated with small pediatric datasets. Further, they highlight relevant tradeoffs in the design landscape of connectomic GNNs. For example, while the simpler GNN model tested exhibits marginally worse accuracy and AUC scores in comparison to the more complex residual GNN, it demonstrates a higher degree of adversarial robustness.

hERGAT: predicting hERG blockers using graph attention mechanism through atom- and molecule-level interaction analyses

The human ether-a-go-go-related gene (hERG) channel plays a critical role in the electrical activity of the heart, and its blockers can cause serious cardiotoxic effects. Thus, screening for hERG channel blockers is a crucial step in the drug development process. Many in silico models have been developed to predict hERG blockers, which can efficiently save time and resources. However, previous methods have found it hard to achieve high performance and to interpret the predictive results. To overcome these challenges, we have proposed hERGAT, a graph neural network model with an attention mechanism, to consider compound interactions on atomic and molecular levels. In the atom-level interaction analysis, we applied a graph attention mechanism (GAT) that integrates information from neighboring nodes and their extended connections. The hERGAT employs a gated recurrent unit (GRU) with the GAT to learn information between more distant atoms. To confirm this, we performed clustering analysis and visualized a correlation heatmap, verifying the interactions between distant atoms were considered during the training process. In the molecule-level interaction analysis, the attention mechanism enables the target node to focus on the most relevant information, highlighting the molecular substructures that play crucial roles in predicting hERG blockers. Through a literature review, we confirmed that highlighted substructures have a significant role in determining the chemical and biological characteristics related to hERG activity. Furthermore, we integrated physicochemical properties into our hERGAT model to improve the performance. Our model achieved an area under the receiver operating characteristic of 0.907 and an area under the precision-recall of 0.904, demonstrating its effectiveness in modeling hERG activity and offering a reliable framework for optimizing drug safety in early development stages.Scientific contribution:hERGAT is a deep learning model for predicting hERG blockers by combining GAT and GRU, enabling it to capture complex interactions at atomic and molecular levels. We improve the model's interpretability by analyzing the highlighted molecular substructures, providing valuable insights into their roles in determining hERG activity. The model achieves high predictive performance, confirming its potential as a preliminary tool for early cardiotoxicity assessment and enhancing the reliability of the results.

Identification of hypertension gene expression biomarkers based on the DeepGCFS algorithm

Hypertension is a critical risk factor and cause of mortality in cardiovascular diseases, and it remains a global public health issue. Therefore, understanding its mechanisms is essential for treating and preventing hypertension. Gene expression data is an important source for obtaining hypertension biomarkers. However, this data has a small sample size and high feature dimensionality, posing challenges to biomarker identification. We propose a novel deep graph clustering feature selection (DeepGCFS) algorithm to identify hypertension gene biomarkers with more biological significance. This algorithm utilizes a graph network to represent the interaction information between genes, builds a GNN model, designs a loss function based on link prediction and self-supervised learning ideas for training, and allows each gene node to obtain a feature vector representing global information. The algorithm then uses hybrid clustering methods for gene module detection. Finally, it combines integrated feature selection methods to determine the gene biomarkers. The experiment revealed that all the ten identified hypertension biomarkers were significantly differentiated, and it was found that the classification performance of AUC can reach 97.50%, which is better than other literature methods. Six genes (PTGS2, TBXA2R, ZNF101, KCNJ2, MSRA, and CMTM5) have been reported to be associated with hypertension. By using GSE113439 as the validation dataset, the AUC value of classification performance was to be 95.45%, and seven of the genes (LYSMD3, TBXA2R, KLC3, GPR171, PTGS2, MSRA, and CMTM5) were to be significantly different. In addition, this algorithm's performance of gene feature vector clustering was better than other comparative methods. Therefore, the proposed algorithm has significant advantages in selecting potential hypertension biomarkers.

Transformer-based short-term traffic forecasting model considering traffic spatiotemporal correlation

Traffic forecasting is crucial for a variety of applications, including route optimization, signal management, and travel time estimation. However, many existing prediction models struggle to accurately capture the spatiotemporal patterns in traffic data due to its inherent nonlinearity, high dimensionality, and complex dependencies. To address these challenges, a short-term traffic forecasting model, Trafficformer, is proposed based on the Transformer framework. The model first uses a multilayer perceptron to extract features from historical traffic data, then enhances spatial interactions through Transformer-based encoding. By incorporating road network topology, a spatial mask filters out noise and irrelevant interactions, improving prediction accuracy. Finally, traffic speed is predicted using another multilayer perceptron. In the experiments, Trafficformer is evaluated on the Seattle Loop Detector dataset. It is compared with six baseline methods, with Mean Absolute Error, Mean Absolute Percentage Error, and Root Mean Square Error used as metrics. The results show that Trafficformer not only has higher prediction accuracy, but also can effectively identify key sections, and has great potential in intelligent traffic control optimization and refined traffic resource allocation.

A comprehensive survey of scoring functions for protein docking models

BACKGROUND

While protein-protein docking is fundamental to our understanding of how proteins interact, scoring protein-protein complex conformations is a critical component of successful docking programs. Without accurate and efficient scoring functions to differentiate between native and non-native binding complexes, the accuracy of current docking tools cannot be guaranteed. Although many innovative scoring functions have been proposed, a good scoring function for docking remains elusive. Deep learning models offer alternatives to using explicit empirical or mathematical functions for scoring protein-protein complexes.

RESULTS

In this study, we perform a comprehensive survey of the state-of-the-art scoring functions by considering the most popular and highly performant approaches, both classical and deep learning-based, for scoring protein-protein complexes. The methods were also compared based on their runtime as it directly impacts their use in large-scale docking applications.

CONCLUSIONS

We evaluate the strengths and weaknesses of classical and deep learning-based approaches across seven public and popular datasets to aid researchers in understanding the progress made in this field.

Deep learning in microbiome analysis: a comprehensive review of neural network models

Microbiome research, the study of microbial communities in diverse environments, has seen significant advances due to the integration of deep learning (DL) methods. These computational techniques have become essential for addressing the inherent complexity and high-dimensionality of microbiome data, which consist of different types of omics datasets. Deep learning algorithms have shown remarkable capabilities in pattern recognition, feature extraction, and predictive modeling, enabling researchers to uncover hidden relationships within microbial ecosystems. By automating the detection of functional genes, microbial interactions, and host-microbiome dynamics, DL methods offer unprecedented precision in understanding microbiome composition and its impact on health, disease, and the environment. However, despite their potential, deep learning approaches face significant challenges in microbiome research. Additionally, the biological variability in microbiome datasets requires tailored approaches to ensure robust and generalizable outcomes. As microbiome research continues to generate vast and complex datasets, addressing these challenges will be crucial for advancing microbiological insights and translating them into practical applications with DL. This review provides an overview of different deep learning models in microbiome research, discussing their strengths, practical uses, and implications for future studies. We examine how these models are being applied to solve key problems and highlight potential pathways to overcome current limitations, emphasizing the transformative impact DL could have on the field moving forward.

Machine Learning-Aided Inverse Design and Discovery of Novel Polymeric Materials for Membrane Separation

Polymeric membranes have been widely used for liquid and gas separation in various industrial applications over the past few decades because of their exceptional versatility and high tunability. Traditional trial-and-error methods for material synthesis are inadequate to meet the growing demands for high-performance membranes. Machine learning (ML) has demonstrated huge potential to accelerate design and discovery of membrane materials. In this review, we cover strengths and weaknesses of the traditional methods, followed by a discussion on the emergence of ML for developing advanced polymeric membranes. We describe methodologies for data collection, data preparation, the commonly used ML models, and the explainable artificial intelligence (XAI) tools implemented in membrane research. Furthermore, we explain the experimental and computational validation steps to verify the results provided by these ML models. Subsequently, we showcase successful case studies of polymeric membranes and emphasize inverse design methodology within a ML-driven structured framework. Finally, we conclude by highlighting the recent progress, challenges, and future research directions to advance ML research for next generation polymeric membranes. With this review, we aim to provide a comprehensive guideline to researchers, scientists, and engineers assisting in the implementation of ML to membrane research and to accelerate the membrane design and material discovery process.

DeepDrug as an expert guided and AI driven drug repurposing methodology for selecting the lead combination of drugs for Alzheimer's disease

Alzheimer's Disease (AD) significantly aggravates human dignity and quality of life. While newly approved amyloid immunotherapy has been reported, effective AD drugs remain to be identified. Here, we propose a novel AI-driven drug-repurposing method, DeepDrug, to identify a lead combination of approved drugs to treat AD patients. DeepDrug advances drug-repurposing methodology in four aspects. Firstly, it incorporates expert knowledge to extend candidate targets to include long genes, immunological and aging pathways, and somatic mutation markers that are associated with AD. Secondly, it incorporates a signed directed heterogeneous biomedical graph encompassing a rich set of nodes and edges, and node/edge weighting to capture crucial pathways associated with AD. Thirdly, it encodes the weighted biomedical graph through a Graph Neural Network into a new embedding space to capture the granular relationships across different nodes. Fourthly, it systematically selects the high-order drug combinations via diminishing return-based thresholds. A five-drug lead combination, consisting of Tofacitinib, Niraparib, Baricitinib, Empagliflozin, and Doxercalciferol, has been selected from the top drug candidates based on DeepDrug scores to achieve the maximum synergistic effect. These five drugs target neuroinflammation, mitochondrial dysfunction, and glucose metabolism, which are all related to AD pathology. DeepDrug offers a novel AI-and-big-data, expert-guided mechanism for new drug combination discovery and drug-repurposing across AD and other neuro-degenerative diseases, with immediate clinical applications.

Multi-Peptide: Multimodality Leveraged Language-Graph Learning of Peptide Properties

Peptides are crucial in biological processes and therapeutic applications. Given their importance, advancing our ability to predict peptide properties is essential. In this study, we introduce Multi-Peptide, an innovative approach that combines transformer-based language models with graph neural networks (GNNs) to predict peptide properties. We integrate PeptideBERT, a transformer model specifically designed for peptide property prediction, with a GNN encoder to capture both sequence-based and structural features. By employing a contrastive loss framework, Multi-Peptide aligns embeddings from both modalities into a shared latent space, thereby enhancing the transformer model's predictive accuracy. Evaluations on hemolysis and nonfouling data sets demonstrate Multi-Peptide's robustness, achieving state-of-the-art 88.057% accuracy in hemolysis prediction. This study highlights the potential of multimodal learning in bioinformatics, paving the way for accurate and reliable predictions in peptide-based research and applications.

MalHAPGNN: An Enhanced Call Graph-Based Malware Detection Framework Using Hierarchical Attention Pooling Graph Neural Network

While deep learning techniques have been extensively employed in malware detection, there is a notable challenge in effectively embedding malware features. Current neural network methods primarily capture superficial characteristics, lacking in-depth semantic exploration of functions and failing to preserve structural information at the file level. Motivated by the aforementioned challenges, this paper introduces MalHAPGNN, a novel framework for malware detection that leverages a hierarchical attention pooling graph neural network based on enhanced call graphs. Firstly, to ensure semantic richness, a Bidirectional Encoder Representations from Transformers-based (BERT) attribute-enhanced function embedding method is proposed for the extraction of node attributes in the function call graph. Subsequently, this work designs a hierarchical graph neural network that integrates attention mechanisms and pooling operations, complemented by function node sampling and structural learning strategies. This framework delivers a comprehensive profile of malicious code across semantic, syntactic, and structural dimensions. Extensive experiments conducted on the Kaggle and VirusShare datasets have demonstrated that the proposed framework outperforms other graph neural network (GNN)-based malware detection methods.

Self-Driving Microscopes: AI Meets Super-Resolution Microscopy

The integration of Machine Learning (ML) with super-resolution microscopy represents a transformative advancement in biomedical research. Recent advances in ML, particularly deep learning (DL), have significantly enhanced image processing tasks, such as denoising and reconstruction. This review explores the growing potential of automation in super-resolution microscopy, focusing on how DL can enable autonomous imaging tasks. Overcoming the challenges of automation, particularly in adapting to dynamic biological processes and minimizing manual intervention, is crucial for the future of microscopy. Whilst still in its infancy, automation in super-resolution can revolutionize drug discovery and disease phenotyping leading to similar breakthroughs as have been recognized in this year's Nobel Prizes for Physics and Chemistry.

Integrating deformable CNN and attention mechanism into multi-scale graph neural network for few-shot image classification

Graph neural networks have excellent performance and powerful representation capabilities, and have been widely used to handle Few-shot image classification problems. The feature extraction module of graph neural networks has always been designed as a fixed convolutional neural network (CNN), but due to the intrinsic properties of convolution operations, its receiving domain is limited. This method has limitations in capturing global feature information and easily ignores key feature information of the image. In order to extract comprehensive and critical feature information, a new CA-MFE algorithm is proposed. The algorithm first utilizes different convolution kernels in CNN to extract multi-scale local feature information, and then based on the global feature extraction ability of attention mechanism, parallel processing of channel and spatial attention mechanism is used to extract multidimensional global feature information. This paper provides a comprehensive performance evaluation of the new model on both mini-ImageNet and tiered ImageNet datasets. Compared with the benchmark model, the classification accuracy has increased by 1.07% and 1.33% respectively; In the 5-way 5-shot task, the classification accuracy of the mini-ImageNet dataset was improved by 11.41%, 7.42%, and 5.38% compared to GNN, TPN, and dynamic models, respectively. The experimental results show that compared with the benchmark model and several representative Few-shot classification algorithm models, the new CA-MFE model has significant superior performance in processing few-shot classification data.

An Adaptive Spatio-Temporal Traffic Flow Prediction Using Self-Attention and Multi-Graph Networks

Traffic flow prediction is a pivotal element in Intelligent Transportation Systems (ITSs) that provides significant opportunities for real-world applications. Capturing complex and dynamic spatio-temporal patterns within traffic data remains a significant challenge for traffic flow prediction. Different approaches to effectively modeling complex spatio-temporal correlations within traffic data have been proposed. These approaches often rely on a single model to capture temporal dependencies, which neglects the varying influences of different time periods on traffic flow. Additionally, these models frequently utilize either static or dynamic graphs to represent spatial dependencies, which limits their ability to address complex and overlapping spatial relationships. Moreover, some approaches struggle to fully capture spatio-temporal variations, leading to the exclusion of critical information and ultimately resulting in suboptimal prediction performance. Thus, this paper introduces the Adaptive Spatio-Temporal Attention-Based Multi-Model (ASTAM), an architecture designed to capture spatio-temporal dependencies within traffic data. The ASTAM employs multi-temporal gated convolution with multi-scale temporal input segments to model complex non-linear temporal correlations. It utilizes static and dynamic parallel multi-graphs to facilitate the modeling of complex spatial dependencies. Furthermore, this model incorporates a spatio-temporal self-attention mechanism to adaptively capture the dynamic and long-term spatio-temporal variations in traffic flow. Experiments conducted on four real-world datasets reveal that the proposed architecture outperformed 13 baseline approaches, achieving average reductions of 5.0% in MAE, 13.28% in RMSE, and 6.46% in MAPE across four datasets.

Alzheimer's disease recognition using graph neural network by leveraging image-text similarity from vision language model

Alzheimer's disease (AD), a progressive neurodegenerative condition, notably impacts cognitive functions and daily activity. One method of detecting dementia involves a task where participants describe a given picture, and extensive research has been conducted using the participants' speech and transcribed text. However, very few studies have explored the modality of the image itself. In this work, we propose a method that predicts dementia automatically by representing the relationship between images and texts as a graph. First, we transcribe the participants' speech into text using an automatic speech recognition system. Then, we employ a vision language model to represent the relationship between the parts of the image and the corresponding descriptive sentences as a bipartite graph. Finally, we use a graph convolutional network (GCN), considering each subject as an individual graph, to classify AD patients through a graph-level classification task. In experiments conducted on the ADReSSo Challenge datasets, our model surpassed the existing state-of-the-art performance by achieving an accuracy of 88.73%. Additionally, ablation studies that removed the relationship between images and texts demonstrated the critical role of graphs in improving performance. Furthermore, by utilizing the sentence representations learned through the GCN, we identified the sentences and keywords critical for AD classification.

Graph Geometric Algebra networks for graph representation learning

Graph neural networks (GNNs) have emerged as a prominent approach for capturing graph topology and modeling vertex-to-vertex relationships. They have been widely used in pattern recognition tasks including node and graph label prediction. However, when dealing with graphs from non-Euclidean domains, the relationships, and interdependencies between objects become more complex. Existing GNNs face limitations in handling a large number of model parameters in such complex graphs. To address this, we propose the integration of Geometric Algebra into graph neural networks, enabling the generalization of GNNs within the geometric space to learn geometric embeddings for nodes and graphs. Our proposed Graph Geometric Algebra Network (GGAN) enhances correlations among nodes by leveraging relations within the Geometric Algebra space. This approach reduces model complexity and improves the learning of graph representations. Through extensive experiments on various benchmark datasets, we demonstrate that our models, utilizing the properties of Geometric Algebra operations, outperform state-of-the-art methods in graph classification and semi-supervised node classification tasks. Our theoretical findings are empirically validated, confirming that our model achieves state-of-the-art performance.

Hybrid physics-based and data-driven modeling of vascular bifurcation pressure differences

Reduced-order models allow for the simulation of blood flow in patient-specific vasculatures. They offer a significant reduction in computational cost and wait time compared to traditional computational fluid dynamics models. Unfortunately, due to the simplifications made in their formulations, reduced-order models can suffer from significantly reduced accuracy. One common simplifying assumption is that of continuity of static or total pressure over vascular bifurcations. In many cases, this assumption has been shown to introduce significant errors in pressure predictions. We propose a model to account for this pressure difference, with the ultimate goal of increasing the accuracy of cardiovascular reduced-order models. Our model successfully uses a structure common in existing reduced-order models in conjunction with machine-learning techniques to predict the pressure difference over a vascular bifurcation. We analyze the performance of our model on steady and transient flows, testing it on three bifurcation cohorts representing three different bifurcation geometric types. We find that our model makes significantly more accurate predictions than other models for approximating bifurcation pressure losses commonly used in the reduced-order cardiovascular modeling community. We also compare the efficacy of different machine-learning techniques and observe that a neural network performs most robustly. Additionally, we consider two different model modalities: one in which the model is fit using both steady and transient flows, and one in which it is optimized for performance in transient flows. We discuss the trade-off between the physical interpretability associated with the first option and the improved accuracy in transient flows associated with the latter option. We also demonstrate the model's ability to generalize by testing it on a combined dataset containing two different bifurcation types. This work marks a step towards improving the accuracy of cardiovascular reduced-order models, thereby increasing their utility for cardiovascular flow modeling.

Enhancing Motor Imagery Classification with Residual Graph Convolutional Networks and Multi-Feature Fusion

Stroke, an abrupt cerebrovascular ailment resulting in brain tissue damage, has prompted the adoption of motor imagery (MI)-based brain-computer interface (BCI) systems in stroke rehabilitation. However, analyzing electroencephalogram (EEG) signals from stroke patients poses challenges. To address the issues of low accuracy and efficiency in EEG classification, particularly involving MI, the study proposes a residual graph convolutional network (M-ResGCN) framework based on the modified S-transform (MST), and introduces the self-attention mechanism into residual graph convolutional network (ResGCN). This study uses MST to extract EEG time-frequency domain features, derives spatial EEG features by calculating the absolute Pearson correlation coefficient (aPcc) between channels, and devises a method to construct the adjacency matrix of the brain network using aPcc to measure the strength of the connection between channels. Experimental results involving 16 stroke patients and 16 healthy subjects demonstrate significant improvements in classification quality and robustness across tests and subjects. The highest classification accuracy reached 94.91% and a Kappa coefficient of 0.8918. The average accuracy and F1 scores from 10 times 10-fold cross-validation are 94.38% and 94.36%, respectively. By validating the feasibility and applicability of brain networks constructed using the aPcc in EEG signal analysis and feature encoding, it was established that the aPcc effectively reflects overall brain activity. The proposed method presents a novel approach to exploring channel relationships in MI-EEG and improving classification performance. It holds promise for real-time applications in MI-based BCI systems.

Generalization limits of Graph Neural Networks in identity effects learning

Graph Neural Networks (GNNs) have emerged as a powerful tool for data-driven learning on various graph domains. They are usually based on a message-passing mechanism and have gained increasing popularity for their intuitive formulation, which is closely linked to the Weisfeiler-Lehman (WL) test for graph isomorphism to which they have been proven equivalent in terms of expressive power. In this work, we establish new generalization properties and fundamental limits of GNNs in the context of learning so-called identity effects, i.e., the task of determining whether an object is composed of two identical components or not. Our study is motivated by the need to understand the capabilities of GNNs when performing simple cognitive tasks, with potential applications in computational linguistics and chemistry. We analyze two case studies: (i) two-letters words, for which we show that GNNs trained via stochastic gradient descent are unable to generalize to unseen letters when utilizing orthogonal encodings like one-hot representations; (ii) dicyclic graphs, i.e., graphs composed of two cycles, for which we present positive existence results leveraging the connection between GNNs and the WL test. Our theoretical analysis is supported by an extensive numerical study.

Application of artificial intelligence to quantitative structure-retention relationship calculations in chromatography

Quantitative structure-retention relationship (QSRR) is an important tool in chromatography. QSRR examines the correlation between molecular structures and their retention behaviors during chromatographic separation. This approach involves developing models for predicting the retention time (RT) of analytes, thereby accelerating method development and facilitating compound identification. In addition, QSRR can be used to study compound retention mechanisms and support drug screening efforts. This review provides a comprehensive analysis of QSRR workflows and applications, with a special focus on the role of artificial intelligence-an area not thoroughly explored in previous reviews. Moreover, we discuss current limitations in RT prediction and propose promising solutions. Overall, this review offers a fresh perspective on future QSRR research, encouraging the development of innovative strategies that enable the diverse applications of QSRR models in chromatographic analysis.

GDMol: Generative Double-Masking Self-Supervised Learning for Molecular Property Prediction

BACKGROUND

Effective molecular feature representation is crucial for drug property prediction. Recent years have seen increased attention on graph neural networks (GNNs) that are pre-trained using self-supervised learning techniques, aiming to overcome the scarcity of labeled data in molecular property prediction. Traditional GNNs in self-supervised molecular property prediction typically perform a single masking operation on the nodes and edges of the input molecular graph, masking only local information and insufficient for thorough self-supervised training.

METHOD

Hence, we propose a model for molecular property prediction based on generative double-masking self-supervised learning, termed as GDMol. This integrates generative learning into the self-supervised learning framework for latent representation, and applies a second round of masking to these latent representations, enabling the model to better capture global information and semantic knowledge of the molecules for a richer, more informative representation, thereby achieving more accurate and robust molecular property prediction.

RESULTS

Our experiments on 5 datasets demonstrated superior performance of GDMol in predicting molecular properties across different domains. Moreover, we used the masking operation to traverse through the gradient changes of each node, the magnitude and sign of which reflect the positive and negative contribution respectively of the local structure in the molecule to the prediction outcome. This in-depth interpretative analysis not only enhances the model's interpretability, but also provides more targeted insights and direction for optimizing drug molecules.

CONCLUSIONS

In summary, this research offers novel insights on improving molecular property prediction tasks, and paves the way for further research on the application of generative learning and self-supervised learning in the field of chemistry.

Advanced statistical inference of myocardial stiffness: A time series Gaussian process approach of emulating cardiac mechanics for real-time clinical decision support

Cardiac mechanics modelling promises to revolutionize personalized health care; however, inferring patient-specific biophysical parameters, which are critical for understanding myocardial functions and performance, poses substantial methodological challenges. Our work is primarily motivated to determine the passive stiffness of the myocardium from the measurement of the left ventricle (LV) volume at various time points, which is crucial for diagnosing cardiac physiological conditions. Although there have been significant advancements in cardiac mechanics modelling, the tasks of inference and uncertainty quantification of myocardial stiffness remain challenging, with high computational costs preventing real-time decision support. We adapt Gaussian processes to construct a statistical surrogate model for emulating LV cavity volume during diastolic filling to overcome this challenge. As the LV volumes, obtained at different time points in diastole, constitute a time series, we apply the Kronecker product trick to decompose the complex covariance matrix of the whole system into two separate covariance matrices, one for time and the other for biophysical parameters. To proceed towards personalized health care, we further integrate patient-specific LV geometries into the Gaussian process emulator using principal component analysis (PCA). Utilizing a deep learning neural network for extracting time-series left ventricle volumes from magnetic resonance images, Bayesian inference is applied to determine the posterior probability distribution of critical cardiac mechanics parameters. Tests on real-patient data illustrate the potential for real-time estimation of myocardial properties for clinical decision-making. These advancements constitute a crucial step towards clinical impact, offering valuable insights into posterior uncertainty quantification for complex cardiac mechanics models.

Adaptive multi-graph contrastive learning for bundle recommendation

Recently, recommending bundles - sets of items that complement each other - instead of individual items to users has drawn much attention in both academia and industry. Models based on Graph Neural Networks (GNNs) for bundle recommendation have achieved great success in capturing users' preferences by modeling pairwise correlations among users, bundles, and items via information propagation on graphs. However, a notable limitation lies in their insufficient focus on explicitly modeling intricate ternary relationships. Additionally, the loose combination of node embeddings from different graphs tends to introduce noise, as it fails to consider disparities among the graphs. To this end, we propose a novel approach called Adaptive Multi-Graph Contrastive Learning for Bundle Recommendation (AMCBR). Specifically, AMCBR models ternary interactions by constructing multiple graphs, including a bundle preference graph based on direct user-bundle interactions, a collaborative neighborhoods graph featuring user-level and bundle-level subgraphs, and an item-level preference hypergraph capturing indirect user-bundle relationships through items. Then, (hyper)graph convolution is applied to each (hyper)graph to encode diverse potential preferences into node embeddings. To enhance the model's robustness, an adaptive aggregation module is employed to assign varying weights to node embeddings from different graphs during the fusion process, which enriches the semantic and comprehensive information in the embeddings while mitigating potential noise. Finally, a contrastive learning strategy is proposed to jointly optimize the model, strengthening collaborative links between individual graphs. Extensive experiments on three real datasets demonstrate that AMCBR can outperform the state-of-the-art baselines on the Top-K recommendations.

HSTrans: Homogeneous substructures transformer for predicting frequencies of drug-side effects

Identifying the frequencies of drug-side effects is crucial for assessing drug risk-benefit. However, accurately determining these frequencies remains challenging due to the limitations of time and scale in clinical randomized controlled trials. As a result, several computational methods have been proposed to address these issues. Nonetheless, two primary problems still persist. Firstly, most of these methods face challenges in generating accurate predictions for novel drugs, as they heavily depend on the interaction graph between drugs and side effects (SEs) within their modeling framework. Secondly, some previous methods often simply concatenate the features of drugs and SEs, which fails to effectively capture their underlying association. In this work, we present HSTrans, a novel approach that treats drugs and SEs as sets of substructures, leveraging a transformer encoder for unified substructure embedding and incorporating an interaction module for association capture. Specifically, HSTrans extracts drug substructures through a specialized algorithm and identifies effective substructures for each SE by employing an indicator that measures the importance of each substructure and SE. Additionally, HSTrans applies convolutional neural network (CNN) in the interaction module to capture complex relationships between drugs and SEs. Experimental results on datasets from Galeano et al.'s study demonstrate that the proposed method outperforms other state-of-the-art approaches. The demo codes for HSTrans are available at https://github.com/Dtdtxuky/HSTrans/tree/master.

AI Methods for Antimicrobial Peptides: Progress and Challenges

Antimicrobial peptides (AMPs) are promising candidates to combat multidrug-resistant pathogens. However, the high cost of extensive wet-lab screening has made AI methods for identifying and designing AMPs increasingly important, with machine learning (ML) techniques playing a crucial role. AI approaches have recently revolutionised this field by accelerating the discovery of new peptides with anti-infective activity, particularly in preclinical mouse models. Initially, classical ML approaches dominated the field, but recently there has been a shift towards deep learning (DL) models. Despite significant contributions, existing reviews have not thoroughly explored the potential of large language models (LLMs), graph neural networks (GNNs) and structure-guided AMP discovery and design. This review aims to fill that gap by providing a comprehensive overview of the latest advancements, challenges and opportunities in using AI methods, with a particular emphasis on LLMs, GNNs and structure-guided design. We discuss the limitations of current approaches and highlight the most relevant topics to address in the coming years for AMP discovery and design.

Predicting the effects of drugs and unveiling their mechanisms of action using an interpretable pharmacodynamic mechanism knowledge graph (IPM-KG)

BACKGROUND

Multiple studies have aimed to consolidate drug-related data and predict drug effects. However, most of these studies have focused on integrating diverse data through correlation rather than representing them based on the pharmacodynamic mechanism of action (MOA). It is thus crucial to obtain interpretability to validate prediction results. In this study, we propose a novel framework to construct knowledge graphs that represent pharmacodynamic MOA, predict drug effects, and derive conceivable mechanistic pathways.

METHODS AND RESULTS

We constructed an interpretable pharmacodynamic mechanism knowledge graph (IPM-KG) by integrating various existing databases and combining them with the approach of this study to automatically fill in the missing data. This yielded a knowledge graph comprising 1455 drugs and 2547 diseases. Additionally, a graph neural network (GNN)-based approach was used to predict therapeutic medication and indication, which outperformed previous approaches that relied on correlation-based knowledge graphs lacking pharmacodynamic MOA representations. Furthermore, we proposed and assessed a method to interpret pharmacodynamic MOA using gene perturbation data. This feasibility study demonstrated the successful derivation of an accurate mechanism in approximately 50 % of cases. Additionally, it facilitated the identification of candidate drugs, which are currently unapproved but exhibit potential for drug repositioning, and their mechanisms of action.

CONCLUSIONS

This framework not only enables the derivation of highly accurate "drug-indication" predictions but also offers a basic mechanistic understanding, thereby facilitating future drug repositioning efforts.

Mitigating off-target effects of small RNAs: conventional approaches, network theory and artificial intelligence

Three types of highly promising small RNA therapeutics, namely, small interfering RNAs (siRNAs), microRNAs (miRNAs) and the RNA subtype of antisense oligonucleotides (ASOs), offer advantages over small-molecule drugs. These small RNAs can target any gene product, opening up new avenues of effective and safe therapeutic approaches for a wide range of diseases. In preclinical research, synthetic small RNAs play an essential role in the investigation of physiological and pathological pathways as silencers of specific genes, facilitating discovery and validation of drug targets in different conditions. Off-target effects of small RNAs, however, could make it difficult to interpret experimental results in the preclinical phase and may contribute to adverse events of small RNA therapeutics. Out of the two major types of off-target effects we focused on the hybridization-dependent, especially on the miRNA-like off-target effects. Our main aim was to discuss several approaches, including sequence design, chemical modifications and target prediction, to reduce hybridization-dependent off-target effects that should be considered even at the early development phase of small RNA therapy. Because there is no standard way of predicting hybridization-dependent off-target effects, this review provides an overview of all major state-of-the-art computational methods and proposes new approaches, such as the possible inclusion of network theory and artificial intelligence (AI) in the prediction workflows. Case studies and a concise survey of experimental methods for validating in silico predictions are also presented. These methods could contribute to interpret experimental results, to minimize off-target effects and hopefully to avoid off-target-related adverse events of small RNA therapeutics. LINKED ARTICLES: This article is part of a themed issue Non-coding RNA Therapeutics. To view the other articles in this section visit http://onlinelibrary.wiley.com/doi/10.1111/bph.v182.2/issuetoc.

Decoding Gestures in Electromyography: Spatiotemporal Graph Neural Networks for Generalizable and Interpretable Classification

In recent years, significant strides in deep learning have propelled the advancement of electromyography (EMG)-based upper-limb gesture recognition systems, yielding notable successes across a spectrum of domains, including rehabilitation, orthopedics, robotics, and human-computer interaction. Despite these achievements, prevailing methodologies often overlook the intrinsic physical configurations and interconnectivity of multi-channel sensory inputs, resulting in a failure to adequately capture relational information embedded within the connections of deployed EMG sensor network topology. This oversight poses a significant challenge, impeding the extraction of crucial features from collaborative multi-channel EMG inputs and subsequently constraining model performance, generalizability, and interpretability. To address these limitations, we introduce novel graph structures meticulously crafted to encapsulate the spatial proximity of distributed EMG sensors and the temporal adjacency of EMG signals. Harnessing these tailored graph structures, we present Graph Convolution Network (GCN)-based classification models adept at effectively extracting and aggregating key features associated with various gestures. Our methodology exhibits remarkable efficacy, achieving state-of-the-art performance across five publicly available datasets, thus underscoring its prowess in gesture recognition tasks. Furthermore, our approach provides interpretable insights into muscular activation patterns, thereby reaffirming the practical effectiveness of our GCN model. Moreover, we show the effectiveness of our graph-based input structure and GCN-based classifier in maintaining high accuracy even with reduced sensor configurations, suggesting their potential for seamless integration into AI-powered rehabilitation strategies utilizing EMG-based gesture classification systems.

Distinct network patterns emerge from Cartesian and XOR epistasis models: a comparative network science analysis

BACKGROUND

Epistasis, the phenomenon where the effect of one gene (or variant) is masked or modified by one or more other genes, significantly contributes to the phenotypic variance of complex traits. Traditionally, epistasis has been modeled using the Cartesian epistatic model, a multiplicative approach based on standard statistical regression. However, a recent study investigating epistasis in obesity-related traits has identified potential limitations of the Cartesian epistatic model, revealing that it likely only detects a fraction of the genetic interactions occurring in natural systems. In contrast, the exclusive-or (XOR) epistatic model has shown promise in detecting a broader range of epistatic interactions and revealing more biologically relevant functions associated with interacting variants. To investigate whether the XOR epistatic model also forms distinct network structures compared to the Cartesian model, we applied network science to examine genetic interactions underlying body mass index (BMI) in rats (Rattus norvegicus).

RESULTS

Our comparative analysis of XOR and Cartesian epistatic models in rats reveals distinct topological characteristics. The XOR model exhibits enhanced sensitivity to epistatic interactions between the network communities found in the Cartesian epistatic network, facilitating the identification of novel trait-related biological functions via community-based enrichment analysis. Additionally, the XOR network features triangle network motifs, indicative of higher-order epistatic interactions. This research also evaluates the impact of linkage disequilibrium (LD)-based edge pruning on network-based epistasis analysis, finding that LD-based edge pruning may lead to increased network fragmentation, which may hinder the effectiveness of network analysis for the investigation of epistasis. We confirmed through network permutation analysis that most XOR and Cartesian epistatic networks derived from the data display distinct structural properties compared to randomly shuffled networks.

CONCLUSIONS

Collectively, these findings highlight the XOR model's ability to uncover meaningful biological associations and higher-order epistasis derived from lower-order network topologies. The introduction of community-based enrichment analysis and motif-based epistatic discovery emphasize network science as a critical approach for advancing epistasis research and understanding complex genetic architectures.

Interface-aware molecular generative framework for protein-protein interaction modulators

Protein-protein interactions (PPIs) play a crucial role in numerous biochemical and biological processes. Although several structure-based molecular generative models have been developed, PPI interfaces and compounds targeting PPIs exhibit distinct physicochemical properties compared to traditional binding pockets and small-molecule drugs. As a result, generating compounds that effectively target PPIs, particularly by considering PPI complexes or interface hotspot residues, remains a significant challenge. In this work, we constructed a comprehensive dataset of PPI interfaces with active and inactive compound pairs. Based on this, we propose a novel molecular generative framework tailored to PPI interfaces, named GENiPPI. Our evaluation demonstrates that GENiPPI captures the implicit relationships between the PPI interfaces and the active molecules, and can generate novel compounds that target these interfaces. Moreover, GENiPPI can generate structurally diverse novel compounds with limited PPI interface modulators. To the best of our knowledge, this is the first exploration of a structure-based molecular generative model focused on PPI interfaces, which could facilitate the design of PPI modulators. The PPI interface-based molecular generative model enriches the existing landscape of structure-based (pocket/interface) molecular generative model. SCIENTIFIC CONTRIBUTION: This study introduces GENiPPI, a protein-protein interaction (PPI) interface-aware molecular generative framework. The framework first employs Graph Attention Networks to capture atomic-level interaction features at the protein complex interface. Subsequently, Convolutional Neural Networks extract compound representations in voxel and electron density spaces. These features are integrated into a Conditional Wasserstein Generative Adversarial Network, which trains the model to generate compound representations targeting PPI interfaces. GENiPPI effectively captures the relationship between PPI interfaces and active/inactive compounds. Furthermore, in fewshot molecular generation, GENiPPI successfully generates compounds comparable to known disruptors. GENiPPI provides an efficient tool for structure-based design of PPI modulators.

Kolmogorov-Arnold Network Made Learning Physics Laws Simple

In recent years, contrastive learning has gained widespread adoption in machine learning applications to physical systems primarily due to its distinctive cross-modal capabilities and scalability. Building on the foundation of Kolmogorov-Arnold Networks (KANs) [Liu, Z. et al. Kan: Kolmogorov-arnold networks. arXiv 2024, 2404.19756], we introduce a novel contrastive learning framework, Kolmogorov-Arnold Contrastive Crystal Property Pretraining (KCCP), which integrates the principles of CLIP and KAN to establish robust correlations between crystal structures and their physical properties. During the training process, we conducted a comparative analysis between Multilayer Perceptron (MLP) and KAN, revealing that KAN significantly outperforms MLP in both accuracy and convergence speed for this task. By extending the capabilities of contrastive learning to the realm of physical systems, KCCP offers a promising approach for constructing cross-data structural and cross-modal physical models, representing an area of considerable potential.

How to build the virtual cell with artificial intelligence: Priorities and opportunities

Cells are essential to understanding health and disease, yet traditional models fall short of modeling and simulating their function and behavior. Advances in AI and omics offer groundbreaking opportunities to create an AI virtual cell (AIVC), a multi-scale, multi-modal large-neural-network-based model that can represent and simulate the behavior of molecules, cells, and tissues across diverse states. This Perspective provides a vision on their design and how collaborative efforts to build AIVCs will transform biological research by allowing high-fidelity simulations, accelerating discoveries, and guiding experimental studies, offering new opportunities for understanding cellular functions and fostering interdisciplinary collaborations in open science.

Recent Applications of Artificial Intelligence in Discovery of New Antibacterial Agents

Antimicrobial resistance (AMR) represents today a major challenge for global public health, compromising the effectiveness of treatments against a multitude of bacterial infections. In recent decades, artificial intelligence (AI) has emerged as a promising technology for the identification and development of new antibacterial agents. This review focuses on AI methodologies applied to discover new antibacterial candidates. Case studies that identified small molecules and peptides showing antimicrobial activity and demonstrating efficiency against pathogenic resistant bacteria by employing AI are summarized. We also discuss the challenges and opportunities offered by AI, highlighting the importance of AI progress for the identification of new promising antibacterial drug candidates to combat the AMR.

Investigation of in silico studies for cytochrome P450 isoforms specificity

Cytochrome P450 (CYP450) enzymes comprise a highly diverse superfamily of heme-thiolate proteins that responsible for catalyzing over 90 % of enzymatic reactions associated with xenobiotic metabolism in humans. Accurately predicting whether chemicals are substrates or inhibitors of different CYP450 isoforms can aid in pre-selecting hit compounds for the drug discovery process, chemical toxicology studies, and patients treatment planning. In this work, we investigated in silico studies on CYP450s specificity over past twenty years, categorizing these studies into structure-based and ligand-based approaches. Subsequently, we utilized 100 of the most frequently prescribed drugs to test eleven machine learning-based prediction models which were published between 2015 and 2024. We analyzed various aspects of the evaluated models, such as their datasets, algorithms, and performance. This will give readers with a comprehensive overview of these prediction models and help them choose the most suitable one to do prediction. We also provide our insights for future research trend in both structure-based and ligand-based approaches in this field.

Multi-source Selective Graph Domain Adaptation Network for cross-subject EEG emotion recognition

Affective brain-computer interface is an important part of realizing emotional human-computer interaction. However, existing objective individual differences among subjects significantly hinder the application of electroencephalography (EEG) emotion recognition. Existing methods still lack the complete extraction of subject-invariant representations for EEG and the ability to fuse valuable information from multiple subjects to facilitate the emotion recognition of the target subject. To address the above challenges, we propose a Multi-source Selective Graph Domain Adaptation Network (MSGDAN), which can better utilize data from different source subjects and perform more robust emotion recognition on the target subject. The proposed network extracts and selects the individual information specific to each subject, where public information refers to subject-invariant components from multi-source subjects. Moreover, the graph domain adaptation network captures both functional connectivity and regional states of the brain via a dynamic graph network and then integrates graph domain adaptation to ensure the invariance of both functional connectivity and regional states. To evaluate our method, we conduct cross-subject emotion recognition experiments on the SEED, SEED-IV, and DEAP datasets. The results demonstrate that the MSGDAN has superior classification performance.

Redundancy-Free Self-Supervised Relational Learning for Graph Clustering

Graph clustering, which learns the node representations for effective cluster assignments, is a fundamental yet challenging task in data analysis and has received considerable attention accompanied by graph neural networks (GNNs) in recent years. However, most existing methods overlook the inherent relational information among the nonindependent and nonidentically distributed nodes in a graph. Due to the lack of exploration of relational attributes, the semantic information of the graph-structured data fails to be fully exploited which leads to poor clustering performance. In this article, we propose a novel self-supervised deep graph clustering method named relational redundancy-free graph clustering (R2FGC) to tackle the problem. It extracts the attribute- and structure-level relational information from both global and local views based on an autoencoder (AE) and a graph AE (GAE). To obtain effective representations of the semantic information, we preserve the consistent relationship among augmented nodes, whereas the redundant relationship is further reduced for learning discriminative embeddings. In addition, a simple yet valid strategy is used to alleviate the oversmoothing issue. Extensive experiments are performed on widely used benchmark datasets to validate the superiority of our R2FGC over state-of-the-art baselines. Our codes are available at https://github.com/yisiyu95/R2FGC.

Gra-CRC-miRTar: The pre-trained nucleotide-to-graph neural networks to identify potential miRNA targets in colorectal cancer

Colorectal cancer (CRC) is the third most diagnosed cancer and the second deadliest cancer worldwide representing a major public health problem. In recent years, increasing evidence has shown that microRNA (miRNA) can control the expression of targeted human messenger RNA (mRNA) by reducing their abundance or translation, acting as oncogenes or tumor suppressors in various cancers, including CRC. Due to the significant up-regulation of oncogenic miRNAs in CRC, elucidating the underlying mechanism and identifying dysregulated miRNA targets may provide a basis for improving current therapeutic interventions. In this paper, we proposed Gra-CRC-miRTar, a pre-trained nucleotide-to-graph neural network framework, for identifying potential miRNA targets in CRC. Different from previous studies, we constructed two pre-trained models to encode RNA sequences and transformed them into de Bruijn graphs. We employed different graph neural networks to learn the latent representations. The embeddings generated from de Bruijn graphs were then fed into a Multilayer Perceptron (MLP) to perform the prediction tasks. Our extensive experiments show that Gra-CRC-miRTar achieves better performance than other deep learning algorithms and existing predictors. In addition, our analyses also successfully revealed 172 out of 201 functional interactions through experimentally validated miRNA-mRNA pairs in CRC. Collectively, our effort provides an accurate and efficient framework to identify potential miRNA targets in CRC, which can also be used to reveal miRNA target interactions in other malignancies, facilitating the development of novel therapeutics. The Gra-CRC-miRTar web server can be found at: http://gra-crc-mirtar.com/.

DCSGMDA: A dual-channel convolutional model based on stacked deep learning collaborative gradient decomposition for predicting miRNA-disease associations

Numerous studies have shown that microRNAs (miRNAs) play a key role in human diseases as critical biomarkers. Its abnormal expression is often accompanied by the emergence of specific diseases. Therefore, studying the relationship between miRNAs and diseases can deepen the insights of their pathogenesis, grasp the process of disease onset and development, and promote drug research of specific diseases. However, many undiscovered relationships between miRNAs and diseases remain, significantly limiting research on miRNA-disease correlations. To explore more potential correlations, we propose a dual-channel convolutional model based on stacked deep learning collaborative gradient decomposition for predicting miRNA-disease associations (DCSGMDA). Firstly, we constructed similarity networks for miRNAs and diseases, as well as an association relationship network. Secondly, potential features were fully mined using stacked deep learning and gradient decomposition networks, along with dual-channel convolutional neural networks. Finally, correlations were scored by a multilayer perceptron. We performed 5-fold and 10-fold cross-validation experiments on DCSGMDA using two datasets based on the Human MicroRNA Disease Database (HMDD). Additionally, parametric, ablation, and comparative experiments, along with case studies, were conducted. The experimental results demonstrate that DCSGMDA performs well in predicting miRNA-disease associations.

Mind Reasoning Manners: Enhancing Type Perception for Generalized Zero-Shot Logical Reasoning Over Text

Logical reasoning task involves diverse types of complex reasoning over text, based on the form of multiple-choice question answering (MCQA). Given the context, question and a set of options as the input, previous methods achieve superior performances on the full-data setting. However, the current benchmark dataset has the ideal assumption that the reasoning type distribution on the train split is close to the test split, which is inconsistent with many real application scenarios. To address it, there remain two problems to be studied: 1) how is the zero-shot capability of the models (train on seen types and test on unseen types)? and 2) how to enhance the perception of reasoning types for the models? For problem 1, we propose a new benchmark for generalized zero-shot logical reasoning, named ZsLR. It includes six splits based on the three type sampling strategies. For problem 2, a type-aware model TaCo is proposed. It utilizes the heuristic input reconstruction and builds a text graph with a global node. Incorporating graph reasoning and contrastive learning, TaCo can improve the type perception in the global representation. Extensive experiments on both the zero-shot and full-data settings prove the superiority of TaCo over the state-of-the-art (SOTA) methods. Also, we experiment and verify the generalization capability of TaCo on other logical reasoning dataset.

Detail-sensitive 3D-UNet for pulmonary airway segmentation from CT images

The segmentation of airway from computed tomography (CT) images plays a vital role in pulmonary disease diagnosis, evaluation, surgical planning, and treatment. Nevertheless, it is still challenging for current methods to handle distal thin and low-contrast airways, leading to mis-segmentation issues. This paper proposes a detail-sensitive 3D-UNet (DS-3D-UNet) that incorporates two new modules into 3D-UNet to segment airways accurately from CT images. The feature recalibration module is designed to give more attention to the foreground airway features through a new attention mechanism. The detail extractor module aims to restore multi-scale detailed features by fusion of features at different levels. Extensive experiments were conducted on the ATM'22 challenge dataset composed of 300 CT scans with airway annotations to evaluate its performance. Quantitative comparisons prove that the proposed model achieves the best performance in terms of Dice similarity coefficient (92.6%) and Intersection over Union (86.3%), outperforming other state-of-the-art methods. Qualitative comparisons further exhibit the superior performance of our method in segmenting thin and confused distal bronchi. The proposed model could provide important references for the diagnosis and treatment of pulmonary diseases, holding promising prospects in the field of digital medicine. Codes are available at https://github.com/nighlevil/DS-3D-UNet/tree/master .

A systematic review of networks for prognostic prediction of health outcomes and diagnostic prediction of health conditions within Electronic Health Records

BACKGROUND AND OBJECTIVE

Using graph theory, Electronic Health Records (EHRs) can be represented graphically to exploit the relational dependencies of the multiple information formats to improve Machine Learning (ML) prediction models. In this systematic qualitative review, we explore the question: How are graphs used on EHRs, to predict diagnosis and health outcomes?

METHODOLOGY

The search strategy identified studies that used patient-level graph representations of EHRs to utilise ML to predict health outcomes and diagnoses. We conducted our search on MEDLINE, Web of Science and Scopus.

RESULTS

832 studies were identified by the search strategy, of which 27 studies were selected for data extraction. Following data extraction, 18 studies used ML with patient-level graph-based representations of EHRs to predict health outcomes and diagnoses. Models ranged from traditional ML to neural network-based models. MIMIC-III was the most used dataset (n = 6, where n is the number of occurrences), followed by National Health Insurance Research Database (NHIRD) (n = 4) and eICU Collaborative Research Database (eICU) (n = 4). The most predicted health outcomes were mortality (n = 9; 21%), hospital readmission (n = 9; 21%), and treatment success (n = 4; 9%). Model performances ranged across outcomes, mortality prediction (Area Under the Receiver Operating Characteristic (AUROC): 72.1 - 91.6; Area Under Precision-Recall Curve (AUPRC): 34.8 - 81.3) and readmission prediction (AUROC: 63.7 - 85.8; AUPRC 39.86 - 84.7). Only one paper had a low Risk of Bias (RoB) that applied to our research question (4%).

CONCLUSION

Graph-based representations using EHRs, for individual health outcomes and diagnoses requires further research before we can see the results applied clinically. The use of graph representations appears to improve EHR representation and predictive performance compared to baseline ML methods in multiple fields of medicine.

AnnoGCD: a generalized category discovery framework for automatic cell type annotation

The identification of cell types in single-cell RNA sequencing (scRNA-seq) data is a critical task in understanding complex biological systems. Traditional supervised machine learning methods rely on large, well-labeled datasets, which are often impractical to obtain in open-world scenarios due to budget constraints and incomplete information. To address these challenges, we propose a novel computational framework, named AnnoGCD, building on Generalized Category Discovery (GCD) and Anomaly Detection (AD) for automatic cell type annotation. Our semi-supervised method combines labeled and unlabeled data to accurately classify known cell types and to discover novel ones, even in imbalanced datasets. AnnoGCD includes a semi-supervised block to first classify known cell types, followed by an unsupervised block aimed at identifying and clustering novel cell types. We evaluated our approach on five human scRNA-seq datasets and a mouse model atlas, demonstrating superior performance in both known and novel cell type identification compared to existing methods. Our model also exhibited robustness in datasets with significant class imbalance. The results suggest that AnnoGCD is a powerful tool for the automatic annotation of cell types in scRNA-seq data, providing a scalable solution for biological research and clinical applications. Our code and the datasets used for evaluations are publicly available on GitHub: https://github.com/cecca46/AnnoGCD/.

GPTFF: A high-accuracy out-of-the-box universal AI force field for arbitrary inorganic materials

This study introduces a novel artificial intelligence (AI) force field, namely a graph-based pre-trained transformer force field (GPTFF), which can simulate arbitrary inorganic systems with good precision and generalizability. Harnessing a large trove of the data and the attention mechanism of transformer algorithms, the model can accurately predict energy, atomic force, and stress with mean absolute error (MAE) values of 32 meV/atom, 71 meV/Å, and 0.365 GPa, respectively. The dataset used to train the model includes 37.8 million single-point energies, 11.7 billion force pairs, and 340.2 million stresses. We also demonstrated that the GPTFF can be universally used to simulate various physical systems, such as crystal structure optimization, phase transition simulations, and mass transport. The model is publicly released with this paper, enabling anyone to use it immediately without needing to train it.

MERGE: A Modal Equilibrium Relational Graph Framework for Multi-Modal Knowledge Graph Completion

The multi-modal knowledge graph completion (MMKGC) task aims to automatically mine the missing factual knowledge from the existing multi-modal knowledge graphs (MMKGs), which is crucial in advancing cross-modal learning and reasoning. However, few methods consider the adverse effects caused by different missing modal information in the model learning process. To address the above challenges, we innovatively propose a Modal Equilibrium Relational Graph framEwork, called MERGE. By constructing three modal-specific directed relational graph attention networks, MERGE can implicitly represent missing modal information for entities by aggregating the modal embeddings from neighboring nodes. Subsequently, a fusion approach based on low-rank tensor decomposition is adopted to align multiple modal features in both the explicit structural level and the implicit semantic level, utilizing the structural information inherent in the original knowledge graphs, which enhances the interpretability of the fused features. Furthermore, we introduce a novel interpolation re-ranking strategy to adjust the importance of modalities during inference while preserving the semantic integrity of each modality. The proposed framework has been validated on four publicly available datasets, and the experimental results have demonstrated the effectiveness and robustness of our method in the MMKGC task.