Rossmann-toolbox: a deep learning-based protocol for the prediction and design of cofactor specificity in Rossmann fold proteins

TopEC: prediction of Enzyme Commission classes by 3D graph neural networks and localized 3D protein descriptor

Tools available for inferring enzyme function from general sequence, fold, or evolutionary information are generally successful. However, they can lead to misclassification if a deviation in local structural features influences the function. Here, we present TopEC, a 3D graph neural network based on a localized 3D descriptor to learn chemical reactions of enzymes from enzyme structures and predict Enzyme Commission (EC) classes. Using message-passing frameworks, we include distance and angle information to significantly improve the predictive performance for EC classification (F-score: 0.72) compared to regular 2D graph neural networks. We trained networks without fold bias that can classify enzyme structures for a vast functional space (>800 ECs). Our model is robust to uncertainties in binding site locations and similar functions in distinct binding sites. We observe that TopEC networks learn from an interplay between biochemical features and local shape-dependent features. TopEC is available as a repository on GitHub: https://github.com/IBG4-CBCLab/TopEC and https://doi.org/10.25838/d5p-66 .

A structure-based framework for selective inhibitor design and optimization

Structure-based drug design aims to create active compounds with favorable properties by analyzing target structures. Recently, deep generative models have facilitated structure-specific molecular generation. However, many methods are limited by inadequate pharmaceutical data, resulting in suboptimal molecular properties and unstable conformations. Additionally, these approaches often overlook binding pocket interactions and struggle with selective inhibitor design. To address these challenges, we developed a framework called Coarse-grained and Multi-dimensional Data-driven molecular generation (CMD-GEN). CMD-GEN bridges ligand-protein complexes with drug-like molecules by utilizing coarse-grained pharmacophore points sampled from diffusion model, enriching training data. Through a hierarchical architecture, it decomposes three-dimensional molecule generation within the pocket into pharmacophore point sampling, chemical structure generation, and conformation alignment, mitigating instability issues. CMD-GEN outperforms other methods in benchmark tests and controls drug-likeness effectively. Furthermore, CMD-GEN excels in cases across three synthetic lethal targets, and wet-lab validation with PARP1/2 inhibitors confirms its potential in selective inhibitor design.

Recent advances in AI-driven protein-ligand interaction predictions

Structure-based drug discovery is a fundamental approach in modern drug development, leveraging computational models to predict protein-ligand interactions. AI-driven methodologies are significantly improving key aspects of the field, including ligand binding site prediction, protein-ligand binding pose estimation, scoring function development, and virtual screening. In this review, we summarize the recent AI-driven advances in various protein-ligand interaction prediction tasks. Traditional docking methods based on empirical scoring functions often lack accuracy, whereas AI models, including graph neural networks, mixture density networks, transformers, and diffusion models, have enhanced predictive performance. Ligand binding site prediction has been refined using geometric deep learning and sequence-based embeddings, aiding in the identification of potential druggable target sites. Binding pose prediction has evolved with sampling-based and regression-based models, as well as protein-ligand co-generation frameworks. AI-powered scoring functions now integrate physical constraints and deep learning techniques to improve binding affinity estimation, leading to more robust virtual screening strategies. Despite these advances, generalization across diverse protein-ligand pairs remains a challenge. As AI technologies continue to evolve, they are expected to revolutionize molecular docking and affinity prediction, increasing both the accuracy and efficiency of structure-based drug discovery.

Characterization of the oxygen-tolerant formate dehydrogenase from Clostridium carboxidivorans

Fixation of CO2 into the organic compound formate by formate dehydrogenases (FDHs) is regarded as the oldest autotrophic process on Earth. It has been proposed that an FDH-dependent CO2 fixation module could support CO2 assimilation even in photoautotrophic organisms. In the present study, we characterized FDH from Clostridium carboxidivorans (ccFDH) due to its ability to reduce CO2 under aerobic conditions. During the production of recombinant ccFDH, in which the selenocysteine codon was replaced by Cys, we were able to replace the W with Mo as the transition metal in the ccFDH metal cofactor, resulting in a two-fold increase of 6 μmol formate min-1 in enzyme activity. Then, we generated ccFDH variants in which the strict NADH preference of the enzyme was changed to NADPH, as this reducing agent is produced in high amounts during the photosynthetic light process. Finally, we showed that the native ccFDH can also directly use ferredoxin as a reducing agent, which is produced by the photosynthetic light reactions at photosystem I. These data collectively suggest that ccFDH and, particularly, its optimized variants can be regarded as suitable enzymes to couple formate production to photosynthesis in photoautotroph organisms, which could potentially support CO2 assimilation via the Calvin-Benson-Bassham (CBB) cycle and minimize CO2 losses due to photorespiration.

Deep learning for NAD/NADP cofactor prediction and engineering using transformer attention analysis in enzymes

Understanding and manipulating the cofactor preferences of NAD(P)-dependent oxidoreductases, the most widely distributed enzyme group in nature, is increasingly crucial in bioengineering. However, large-scale identification of the cofactor preferences and the design of mutants to switch cofactor specificity remain as complex tasks. Here, we introduce DISCODE (Deep learning-based Iterative pipeline to analyze Specificity of COfactors and to Design Enzyme), a novel transformer-based deep learning model to predict NAD(P) cofactor preferences. For model training, a total of 7,132 NAD(P)-dependent enzyme sequences were collected. Leveraging whole-length sequence information, DISCODE classifies the cofactor preferences of NAD(P)-dependent oxidoreductase protein sequences without structural or taxonomic limitation. The model showed 97.4% and 97.3% of accuracy and F1 score, respectively. A notable feature of DISCODE is the interpretability of its transformer layers. Analysis of attention layers in the model enables identification of several residues that showed significantly higher attention weights. They were well aligned with structurally important residues that closely interact with NAD(P), facilitating the identification of key residues for determining cofactor specificities. These key residues showed high consistency with verified cofactor switching mutants. Integrated into an enzyme design pipeline, DISCODE coupled with attention analysis, enables a fully automated approach to redesign cofactor specificity.

The INSIGHT platform: Enhancing NAD(P)-dependent specificity prediction for co-factor specificity engineering

Enzyme specificity towards cofactors like NAD(P)H is crucial for applications in bioremediation and eco-friendly chemical synthesis. Despite their role in converting pollutants and creating sustainable products, predicting enzyme specificity faces challenges due to sparse data and inadequate models. To bridge this gap, we developed the cutting-edge INSIGHT platform to enhance the prediction of coenzyme specificity in NAD(P)-dependent enzymes. INSIGHT integrates extensive data from principal bioinformatics resources, concentrating on both NADH and NADPH specificities, and utilizes advanced protein language models to refine the predictions. This integration not only strengthens computational predictions but also meets the practical demands of high-throughput screening and optimization. Experimental validation confirms INSIGHT's effectiveness, boosting our ability to engineer enzymes for efficient, sustainable industrial and environmental processes. This work advances the practical use of computational tools in enzyme research, addressing industrial needs and offering scalable solutions for environmental challenges.

Unraveling the evolutionary origin of the P5CS gene: a story of gene fusion and horizontal transfer

The accumulation of proline in response to the most diverse types of stress is a widespread defense mechanism. In prokaryotes, fungi, and certain unicellular eukaryotes (green algae), the first two reactions of proline biosynthesis occur through two distinct enzymes, γ-glutamyl kinase (GK E.C. 2.7.2.11) and γ-glutamyl phosphate reductase (GPR E.C. 1.2.1.41), encoded by two different genes, ProB and ProA, respectively. Plants, animals, and a few unicellular eukaryotes carry out these reactions through a single bifunctional enzyme, the Δ1-pyrroline-5-carboxylate synthase (P5CS), which has the GK and GPR domains fused. To better understand the origin and diversification of the P5CS gene, we use a robust phylogenetic approach with a broad sampling of the P5CS, ProB and ProA genes, including species from all three domains of life. Our results suggest that the collected P5CS genes have arisen from a single fusion event between the ProA and ProB gene paralogs. A peculiar fusion event occurred in an ancestral eukaryotic lineage and was spread to other lineages through horizontal gene transfer. As for the diversification of this gene family, the phylogeny of the P5CS gene in plants shows that there have been multiple independent processes of duplication and loss of this gene, with the duplications being related to old polyploidy events.

Graph to Activation Energy Models Easily Reach Irreducible Errors but Show Limited Transferability

Activation energy characterization of competing reactions is a costly but crucial step for understanding the kinetic relevance of distinct reaction pathways, product yields, and myriad other properties of reacting systems. The standard methodology for activation energy characterization has historically been a transition state search using the highest level of theory that can be afforded. However, recently, several groups have popularized the idea of predicting activation energies directly based on nothing more than the reactant and product graphs, a sufficiently complex neural network, and a broad enough data set. Here, we have revisited this task using the recently developed Reaction Graph Depth 1 (RGD1) transition state data set and several newly developed graph attention architectures. All of these new architectures achieve similar state-of-the-art results of ∼4 kcal/mol mean absolute error on withheld testing sets of reactions but poor performance on external testing sets composed of reactions with differing mechanisms, reaction molecularity, or reactant size distribution. Limited transferability is also shown to be shared by other contemporary graph to activation energy architectures through a series of case studies. We conclude that an array of standard graph architectures can already achieve results comparable to the irreducible error of available reaction data sets but that out-of-distribution performance remains poor.

A new paradigm for applying deep learning to protein-ligand interaction prediction

Protein-ligand interaction prediction presents a significant challenge in drug design. Numerous machine learning and deep learning (DL) models have been developed to accurately identify docking poses of ligands and active compounds against specific targets. However, current models often suffer from inadequate accuracy or lack practical physical significance in their scoring systems. In this research paper, we introduce IGModel, a novel approach that utilizes the geometric information of protein-ligand complexes as input for predicting the root mean square deviation of docking poses and the binding strength (pKd, the negative value of the logarithm of binding affinity) within the same prediction framework. This ensures that the output scores carry intuitive meaning. We extensively evaluate the performance of IGModel on various docking power test sets, including the CASF-2016 benchmark, PDBbind-CrossDocked-Core and DISCO set, consistently achieving state-of-the-art accuracies. Furthermore, we assess IGModel's generalizability and robustness by evaluating it on unbiased test sets and sets containing target structures generated by AlphaFold2. The exceptional performance of IGModel on these sets demonstrates its efficacy. Additionally, we visualize the latent space of protein-ligand interactions encoded by IGModel and conduct interpretability analysis, providing valuable insights. This study presents a novel framework for DL-based prediction of protein-ligand interactions, contributing to the advancement of this field. The IGModel is available at GitHub repository https://github.com/zchwang/IGModel.

Standigm ASK™: knowledge graph and artificial intelligence platform applied to target discovery in idiopathic pulmonary fibrosis

Standigm ASK™ revolutionizes healthcare by addressing the critical challenge of identifying pivotal target genes in disease mechanisms-a fundamental aspect of drug development success. Standigm ASK™ integrates a unique combination of a heterogeneous knowledge graph (KG) database and an attention-based neural network model, providing interpretable subgraph evidence. Empowering users through an interactive interface, Standigm ASK™ facilitates the exploration of predicted results. Applying Standigm ASK™ to idiopathic pulmonary fibrosis (IPF), a complex lung disease, we focused on genes (AMFR, MDFIC and NR5A2) identified through KG evidence. In vitro experiments demonstrated their relevance, as TGFβ treatment induced gene expression changes associated with epithelial-mesenchymal transition characteristics. Gene knockdown reversed these changes, identifying AMFR, MDFIC and NR5A2 as potential therapeutic targets for IPF. In summary, Standigm ASK™ emerges as an innovative KG and artificial intelligence platform driving insights in drug target discovery, exemplified by the identification and validation of therapeutic targets for IPF.

EEG emotion recognition using improved graph neural network with channel selection

BACKGROUND AND OBJECTIVE

Emotion classification tasks based on electroencephalography (EEG) are an essential part of artificial intelligence, with promising applications in healthcare areas such as autism research and emotion detection in pregnant women. However, the complex data acquisition environment provides a variable number of EEG channels, which interferes with the model to simulate the process of information transfer in the human brain. Therefore, this paper proposes an improved graph convolution model with dynamic channel selection.

METHODS

The proposed model combines the advantages of 1D convolution and graph convolution to capture the intra- and inter-channel EEG features, respectively. We add functional connectivity in the graph structure that helps to simulate the relationship between brain regions further. In addition, an adjustable scale of channel selection can be performed based on the attention distribution in the graph structure.

RESULTS

We conducted various experiments on the DEAP-Twente, DEAP-Geneva, and SEED datasets and achieved average accuracies of 90.74%, 91%, and 90.22%, respectively, which exceeded most existing models. Meanwhile, with only 20% of the EEG channels retained, the models achieved average accuracies of 82.78%, 84%, and 83.93% on the above three datasets, respectively.

CONCLUSIONS

The experimental results show that the proposed model can achieve effective emotion classification in complex dataset environments. Also, the proposed channel selection method is informative for reducing the cost of affective computing.

Learning deep representations of enzyme thermal adaptation

Temperature is a fundamental environmental factor that shapes the evolution of organisms. Learning thermal determinants of protein sequences in evolution thus has profound significance for basic biology, drug discovery, and protein engineering. Here, we use a data set of over 3 million BRENDA enzymes labeled with optimal growth temperatures (OGTs) of their source organisms to train a deep neural network model (DeepET). The protein-temperature representations learned by DeepET provide a temperature-related statistical summary of protein sequences and capture structural properties that affect thermal stability. For prediction of enzyme optimal catalytic temperatures and protein melting temperatures via a transfer learning approach, our DeepET model outperforms classical regression models trained on rationally designed features and other deep-learning-based representations. DeepET thus holds promise for understanding enzyme thermal adaptation and guiding the engineering of thermostable enzymes.

Localpdb- a Python package to manage protein structures and their annotations

Motivation

The wealth of protein structures collected in the Protein Data Bank enabled large-scale studies of their function and evolution. Such studies, however, require the generation of customized datasets combining the structural data with miscellaneous accessory resources providing functional, taxonomic and other annotations. Unfortunately, the functionality of currently available tools for the creation of such datasets is limited and their usage frequently requires laborious surveying of various data sources and resolving inconsistencies between their versions.

Results

To address this problem, we developed localpdb, a versatile Python library for the management of protein structures and their annotations. The library features a flexible plugin system enabling seamless unification of the structural data with diverse auxiliary resources, full version control and powerful functionality of creating highly customized datasets. The localpdb can be used in a wide range of bioinformatic tasks, in particular those involving large-scale protein structural analyses and machine learning.

Availability and implementation

localpdb is freely available at https://github.com/labstructbioinf/localpdb. Documentation along with the usage examples can be accessed at https://labstructbioinf.github.io/localpdb/.