Learning continuous and data-driven molecular descriptors by translating equivalent chemical representations

Data-driven, explainable machine learning model for predicting volatile organic compounds' standard vaporization enthalpy

The accurate prediction of standard vaporization enthalpy (ΔvapHm°) for volatile organic compounds (VOCs) is of paramount importance in environmental chemistry, industrial applications and regulatory compliance. To overcome traditional experimental methods for predicting ΔvapHm° of VOCs, machine learning (ML) models enable a high-throughput, cost-effective property estimation. But despite a rising momentum, existing ML algorithms still present limitations in prediction accuracy and broad chemical applications. In this work, we present a data driven, explainable supervised ML model to predict ΔvapHm° of VOCs. The model was built on an established experimental database of 2410 unique molecules and 223 VOCs categorized by chemical groups. Using supervised ML regression algorithms, the Random Forest successfully predicted VOCs' ΔvapHm° with a mean absolute error of 3.02 kJ mol-1 and a 95% test score. The model was successfully validated through the prediction of ΔvapHm° for a known database of VOCs and through molecular group hold-out tests. Through chemical feature importance analysis, this explainable model revealed that VOC polarizability, connectivity indexes and electrotopological state are key for the model's prediction accuracy. We thus present a replicable and explainable model, which can be further expanded towards the prediction of other thermodynamic properties of VOCs.

Enhancing Molecular Property Prediction through Task-Oriented Transfer Learning: Integrating Universal Structural Insights and Domain-Specific Knowledge

Precisely predicting molecular properties is crucial in drug discovery, but the scarcity of labeled data poses a challenge for applying deep learning methods. While large-scale self-supervised pretraining has proven an effective solution, it often neglects domain-specific knowledge. To tackle this issue, we introduce Task-Oriented Multilevel Learning based on BERT (TOML-BERT), a dual-level pretraining framework that considers both structural patterns and domain knowledge of molecules. TOML-BERT achieved state-of-the-art prediction performance on 10 pharmaceutical datasets. It has the capability to mine contextual information within molecular structures and extract domain knowledge from massive pseudo-labeled data. The dual-level pretraining accomplished significant positive transfer, with its two components making complementary contributions. Interpretive analysis elucidated that the effectiveness of the dual-level pretraining lies in the prior learning of a task-related molecular representation. Overall, TOML-BERT demonstrates the potential of combining multiple pretraining tasks to extract task-oriented knowledge, advancing molecular property prediction in drug discovery.

Evolutionary Multiobjective Molecule Optimization in an Implicit Chemical Space

Optimization techniques play a pivotal role in advancing drug development, serving as the foundation of numerous generative methods tailored to efficiently design optimized molecules derived from existing lead compounds. However, existing methods often encounter difficulties in generating diverse, novel, and high-property molecules that simultaneously optimize multiple drug properties. To overcome this bottleneck, we propose a multiobjective molecule optimization framework (MOMO). MOMO employs a specially designed Pareto-based multiproperty evaluation strategy at the molecular sequence level to guide the evolutionary search in an implicit chemical space. A comparative analysis of MOMO with five state-of-the-art methods across two benchmark multiproperty molecule optimization tasks reveals that MOMO markedly outperforms them in terms of diversity, novelty, and optimized properties. The practical applicability of MOMO in drug discovery has also been validated on four challenging tasks in the real-world discovery problem. These results suggest that MOMO can provide a useful tool to facilitate molecule optimization problems with multiple properties.

Integrating transformers and many-objective optimization for drug design

BACKGROUND

Drug design is a challenging and important task that requires the generation of novel and effective molecules that can bind to specific protein targets. Artificial intelligence algorithms have recently showed promising potential to expedite the drug design process. However, existing methods adopt multi-objective approaches which limits the number of objectives.

RESULTS

In this paper, we expand this thread of research from the many-objective perspective, by proposing a novel framework that integrates a latent Transformer-based model for molecular generation, with a drug design system that incorporates absorption, distribution, metabolism, excretion, and toxicity prediction, molecular docking, and many-objective metaheuristics. We compared the performance of two latent Transformer models (ReLSO and FragNet) on a molecular generation task and show that ReLSO outperforms FragNet in terms of reconstruction and latent space organization. We then explored six different many-objective metaheuristics based on evolutionary algorithms and particle swarm optimization on a drug design task involving potential drug candidates to human lysophosphatidic acid receptor 1, a cancer-related protein target.

CONCLUSION

We show that multi-objective evolutionary algorithm based on dominance and decomposition performs the best in terms of finding molecules that satisfy many objectives, such as high binding affinity and low toxicity, and high drug-likeness. Our framework demonstrates the potential of combining Transformers and many-objective computational intelligence for drug design.

AMPred-CNN: Ames mutagenicity prediction model based on convolutional neural networks

Mutagenicity assessment plays a pivotal role in the safety evaluation of chemicals, pharmaceuticals, and environmental compounds. In recent years, the development of robust computational models for predicting chemical mutagenicity has gained significant attention, driven by the need for efficient and cost-effective toxicity assessments. In this paper, we proposed AMPred-CNN, an innovative Ames mutagenicity prediction model based on Convolutional Neural Networks (CNNs), uniquely employing molecular structures as images to leverage CNNs' powerful feature extraction capabilities. The study employs the widely used benchmark mutagenicity dataset from Hansen et al. for model development and evaluation. Comparative analyses with traditional ML models on different molecular features reveal substantial performance enhancements. AMPred-CNN outshines these models, demonstrating superior accuracy, AUC, F1 score, MCC, sensitivity, and specificity on the test set. Notably, AMPred-CNN is further benchmarked against seven recent ML and DL models, consistently showcasing superior performance with an impressive AUC of 0.954. Our study highlights the effectiveness of CNNs in advancing mutagenicity prediction, paving the way for broader applications in toxicology and drug development.

Machine learning models to predict ligand binding affinity for the orexin 1 receptor

The orexin 1 receptor (OX1R) is a G-protein coupled receptor that regulates a variety of physiological processes through interactions with the neuropeptides orexin A and B. Selective OX1R antagonists exhibit therapeutic effects in preclinical models of several behavioral disorders, including drug seeking and overeating. However, currently there are no selective OX1R antagonists approved for clinical use, fueling demand for novel compounds that act at this target. In this study, we meticulously curated a dataset comprising over 1300 OX1R ligands using a stringent filter and criteria cascade. Subsequently, we developed highly predictive quantitative structure-activity relationship (QSAR) models employing the optimized hyper-parameters for the random forest machine learning algorithm and twelve 2D molecular descriptors selected by recursive feature elimination with a 5-fold cross-validation process. The predictive capacity of the QSAR model was further assessed using an external test set and enrichment study, confirming its high predictivity. The practical applicability of our final QSAR model was demonstrated through virtual screening of the DrugBank database. This revealed two FDA-approved drugs (isavuconazole and cabozantinib) as potential OX1R ligands, confirmed by radiolabeled OX1R binding assays. To our best knowledge, this study represents the first report of highly predictive QSAR models on a large comprehensive dataset of diverse OX1R ligands, which should prove useful for the discovery and design of new compounds targeting this receptor.

Advances in machine learning with chemical language models in molecular property and reaction outcome predictions

Molecular properties and reactions form the foundation of chemical space. Over the years, innumerable molecules have been synthesized, a smaller fraction of them found immediate applications, while a larger proportion served as a testimony to creative and empirical nature of the domain of chemical science. With increasing emphasis on sustainable practices, it is desirable that a target set of molecules are synthesized preferably through a fewer empirical attempts instead of a larger library, to realize an active candidate. In this front, predictive endeavors using machine learning (ML) models built on available data acquire high timely significance. Prediction of molecular property and reaction outcome remain one of the burgeoning applications of ML in chemical science. Among several methods of encoding molecular samples for ML models, the ones that employ language like representations are gaining steady popularity. Such representations would additionally help adopt well-developed natural language processing (NLP) models for chemical applications. Given this advantageous background, herein we describe several successful chemical applications of NLP focusing on molecular property and reaction outcome predictions. From relatively simpler recurrent neural networks (RNNs) to complex models like transformers, different network architecture have been leveraged for tasks such as de novo drug design, catalyst generation, forward and retro-synthesis predictions. The chemical language model (CLM) provides promising avenues toward a broad range of applications in a time and cost-effective manner. While we showcase an optimistic outlook of CLMs, attention is also placed on the persisting challenges in reaction domain, which would optimistically be addressed by advanced algorithms tailored to chemical language and with increased availability of high-quality datasets.

Predicting drug-Protein interaction with deep learning framework for molecular graphs and sequences: Potential candidates against SAR-CoV-2

The severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) caused the COVID-19 disease, which represents a new life-threatening disaster. Regarding viral infection, many therapeutics have been investigated to alleviate the epidemiology such as vaccines and receptor decoys. However, the continuous mutating coronavirus, especially the variants of Delta and Omicron, are tended to invalidate the therapeutic biological product. Thus, it is necessary to develop molecular entities as broad-spectrum antiviral drugs. Coronavirus replication is controlled by the viral 3-chymotrypsin-like cysteine protease (3CLpro) enzyme, which is required for the virus's life cycle. In the cases of severe acute respiratory syndrome coronavirus (SARS-CoV) and middle east respiratory syndrome coronavirus (MERS-CoV), 3CLpro has been shown to be a promising therapeutic development target. Here we proposed an attention-based deep learning framework for molecular graphs and sequences, training from the BindingDB 3CLpro dataset (114,555 compounds). After construction of such model, we conducted large-scale screening the in vivo/vitro dataset (276,003 compounds) from Zinc Database and visualize the candidate compounds with attention score. geometric-based affinity prediction was employed for validation. Finally, we established a 3CLpro-specific deep learning framework, namely GraphDPI-3CL (AUROC: 0.958) achieved superior performance beyond the existing state of the art model and discovered 10 molecules with a high binding affinity of 3CLpro and superior binding mode.

G-K BertDTA: A graph representation learning and semantic embedding-based framework for drug-target affinity prediction

Developing new drugs is costly, time-consuming, and risky. Drug-target affinity (DTA), indicating the binding capability between drugs and target proteins, is a crucial indicator for drug development. Accurately predicting interaction strength between new drug-target pairs by analyzing previous experiments aids in screening potential drug molecules, repurposing them, and developing safe and effective medicines. Existing computational models for DTA prediction rely on strings or single-graph neural networks, lacking consideration of protein structure and molecular semantic information, leading to limited accuracy. Our experiments demonstrate that string-based methods may overlook protein conformations, causing a high root mean square error (RMSE) of 3.584 in affinity due to a lack of spatial context. Single graph networks also underperform on topology features, with a 6% lower confidence interval (CI) for activity classification. Absent semantic information also limits generalization across diverse compounds, resulting in 18% increment in RMSE and 5% in misclassifications within quantifications study, restricting potential drug discovery. To address these limitations, we propose G-K BertDTA, a novel framework for accurate DTA prediction incorporating protein features, molecular semantic features, and molecular structural information. In this proposed model, we represent drugs as graphs, with a GIN employed to learn the molecular topological information. For the extraction of protein structural features, we utilize a DenseNet architecture. A knowledge-based BERT semantic model is incorporated to obtain rich pre-trained semantic embeddings, thereby enhancing the feature information. We extensively evaluated our proposed approach on the publicly available benchmark datasets (i.e., KIBA and Davis), and experimental results demonstrate the promising performance of our method, which consistently outperforms previous state-of-the-art approaches. Code is available at https://github.com/AmbitYuki/G-K-BertDTA.

In Silico Prediction of Oral Acute Rodent Toxicity Using Consensus Machine Learning

Acute oral toxicity (AOT) is required for the classification and labeling of chemicals according to the global harmonized system (GHS). Acute oral toxicity studies are optimized to minimize the use of animals. However, with the advent of the three Rs principles and machine learning in toxicology, alternative in silico methods became a reasonable alternative approach for addressing the AOT of new chemical matter. Here, we describe the compilation of AOT data from a commercial database and the development of a consensus classification model after evaluating different combinations of molecular representations and machine learning algorithms. The model shows significantly better performance compared to publicly available AOT models. Its performance was evaluated on an external validation data set, which was compiled from the literature, and an applicability domain was deduced.

Advancing Adverse Drug Reaction Prediction with Deep Chemical Language Model for Drug Safety Evaluation

The accurate prediction of adverse drug reactions (ADRs) is essential for comprehensive drug safety evaluation. Pre-trained deep chemical language models have emerged as powerful tools capable of automatically learning molecular structural features from large-scale datasets, showing promising capabilities for the downstream prediction of molecular properties. However, the performance of pre-trained chemical language models in predicting ADRs, especially idiosyncratic ADRs induced by marketed drugs, remains largely unexplored. In this study, we propose MoLFormer-XL, a pre-trained model for encoding molecular features from canonical SMILES, in conjunction with a CNN-based model to predict drug-induced QT interval prolongation (DIQT), drug-induced teratogenicity (DIT), and drug-induced rhabdomyolysis (DIR). Our results demonstrate that the proposed model outperforms conventional models applied in previous studies for predicting DIQT, DIT, and DIR. Notably, an analysis of the learned linear attention maps highlights amines, alcohol, ethers, and aromatic halogen compounds as strongly associated with the three types of ADRs. These findings hold promise for enhancing drug discovery pipelines and reducing the drug attrition rate due to safety concerns.

MELLODDY: Cross-pharma Federated Learning at Unprecedented Scale Unlocks Benefits in QSAR without Compromising Proprietary Information

Federated multipartner machine learning has been touted as an appealing and efficient method to increase the effective training data volume and thereby the predictivity of models, particularly when the generation of training data is resource-intensive. In the landmark MELLODDY project, indeed, each of ten pharmaceutical companies realized aggregated improvements on its own classification or regression models through federated learning. To this end, they leveraged a novel implementation extending multitask learning across partners, on a platform audited for privacy and security. The experiments involved an unprecedented cross-pharma data set of 2.6+ billion confidential experimental activity data points, documenting 21+ million physical small molecules and 40+ thousand assays in on-target and secondary pharmacodynamics and pharmacokinetics. Appropriate complementary metrics were developed to evaluate the predictive performance in the federated setting. In addition to predictive performance increases in labeled space, the results point toward an extended applicability domain in federated learning. Increases in collective training data volume, including by means of auxiliary data resulting from single concentration high-throughput and imaging assays, continued to boost predictive performance, albeit with a saturating return. Markedly higher improvements were observed for the pharmacokinetics and safety panel assay-based task subsets.

De Novo Generation of Chemical Structures of Inhibitor and Activator Candidates for Therapeutic Target Proteins by a Transformer-Based Variational Autoencoder and Bayesian Optimization

Deep generative models for molecular generation have been gaining much attention as structure generators to accelerate drug discovery. However, most previously developed methods are chemistry-centric approaches, and comprehensive biological responses in the cell have not been taken into account. In this study, we propose a novel computational method, TRIOMPHE-BOA (transcriptome-based inference and generation of molecules with desired phenotypes using the Bayesian optimization algorithm), to generate new chemical structures of inhibitor or activator candidates for therapeutic target proteins by integrating chemically and genetically perturbed transcriptome profiles. In the algorithm, the substructures of multiple molecules that were selected based on the transcriptome analysis are fused in the design of new chemical structures by exploring the latent space of a Transformer-based variational autoencoder using Bayesian optimization. Our results demonstrate the usefulness of the proposed method in terms of having high reproducibility of existing ligands for 10 therapeutic target proteins when compared with previous methods. Moreover, this method can be applied to proteins without detailed 3D structures or known ligands and is expected to become a powerful tool for more efficient hit identification.

Explainable Supervised Machine Learning Model To Predict Solvation Gibbs Energy

Many challenges persist in developing accurate computational models for predicting solvation free energy (ΔGsol). Despite recent developments in Machine Learning (ML) methodologies that outperformed traditional quantum mechanical models, several issues remain concerning explanatory insights for broad chemical predictions with an acceptable speed-accuracy trade-off. To overcome this, we present a novel supervised ML model to predict the ΔGsol for an array of solvent-solute pairs. Using two different ensemble regressor algorithms, we made fast and accurate property predictions using open-source chemical features, encoding complex electronic, structural, and surface area descriptors for every solvent and solute. By integrating molecular properties and chemical interaction features, we have analyzed individual descriptor importance and optimized our model though explanatory information form feature groups. On aqueous and organic solvent databases, ML models revealed the predictive relevance of solutes with increasing polar surface area and decreasing polarizability, yielding better results than state-of-the-art benchmark Neural Network methods (without complex quantum mechanical or molecular dynamic simulations). Both algorithms successfully outperformed previous ΔGsol predictions methods, with a maximum absolute error of 0.22 ± 0.02 kcal mol-1, further validated in an external benchmark database and with solvent hold-out tests. With these explanatory and statistical insights, they allow a thoughtful application of this method for predicting other thermodynamic properties, stressing the relevance of ML modeling for further complex computational chemistry problems.

HyperPCM: Robust Task-Conditioned Modeling of Drug-Target Interactions

A central problem in drug discovery is to identify the interactions between drug-like compounds and protein targets. Over the past few decades, various quantitative structure-activity relationship (QSAR) and proteo-chemometric (PCM) approaches have been developed to model and predict these interactions. While QSAR approaches solely utilize representations of the drug compound, PCM methods incorporate both representations of the protein target and the drug compound, enabling them to achieve above-chance predictive accuracy on previously unseen protein targets. Both QSAR and PCM approaches have recently been improved by machine learning and deep neural networks, that allow the development of drug-target interaction prediction models from measurement data. However, deep neural networks typically require large amounts of training data and cannot robustly adapt to new tasks, such as predicting interaction for unseen protein targets at inference time. In this work, we propose to use HyperNetworks to efficiently transfer information between tasks during inference and thus to accurately predict drug-target interactions on unseen protein targets. Our HyperPCM method reaches state-of-the-art performance compared to previous methods on multiple well-known benchmarks, including Davis, DUD-E, and a ChEMBL derived data set, and particularly excels at zero-shot inference involving unseen protein targets. Our method, as well as reproducible data preparation, is available at https://github.com/ml-jku/hyper-dti.

Deep Generative Models in De Novo Drug Molecule Generation

The discovery of new drugs has important implications for human health. Traditional methods for drug discovery rely on experiments to optimize the structure of lead molecules, which are time-consuming and high-cost. Recently, artificial intelligence has exhibited promising and efficient performance for drug-like molecule generation. In particular, deep generative models achieve great success in de novo generation of drug-like molecules with desired properties, showing massive potential for novel drug discovery. In this study, we review the recent progress of molecule generation using deep generative models, mainly focusing on molecule representations, public databases, data processing tools, and advanced artificial intelligence based molecule generation frameworks. In particular, we present a comprehensive comparison of state-of-the-art deep generative models for molecule generation and a summary of commonly used molecular design strategies. We identify research gaps and challenges of molecule generation such as the need for better databases, missing 3D information in molecular representation, and the lack of high-precision evaluation metrics. We suggest future directions for molecular generation and drug discovery.

OptADMET: a web-based tool for substructure modifications to improve ADMET properties of lead compounds

Lead optimization is a crucial step in the drug discovery process, which aims to design potential drug candidates from biologically active hits. During lead optimization, active hits undergo modifications to improve their absorption, distribution, metabolism, excretion and toxicity (ADMET) profiles. Medicinal chemists face key questions regarding which compound(s) should be synthesized next and how to balance multiple ADMET properties. Reliable transformation rules from multiple experimental analyses are critical to improve this decision-making process. We developed OptADMET ( https://cadd.nscc-tj.cn/deploy/optadmet/ ), an integrated web-based platform that provides chemical transformation rules for 32 ADMET properties and leverages prior experimental data for lead optimization. The multiproperty transformation rule database contains a total of 41,779 validated transformation rules generated from the analysis of 177,191 reliable experimental datasets. Additionally, 146,450 rules were generated by analyzing 239,194 molecular data predictions. OptADMET provides the ADMET profiles of all optimized molecules from the queried molecule and enables the prediction of desirable substructure transformations and subsequent validation of drug candidates. OptADMET is based on matched molecular pairs analysis derived from synthetic chemistry, thus providing improved practicality over other methods. OptADMET is designed for use by both experimental and computational scientists.

Machine learning study of the extended drug-target interaction network informed by pain related voltage-gated sodium channels

ABSTRACT

Pain is a significant global health issue, and the current treatment options for pain management have limitations in terms of effectiveness, side effects, and potential for addiction. There is a pressing need for improved pain treatments and the development of new drugs. Voltage-gated sodium channels, particularly Nav1.3, Nav1.7, Nav1.8, and Nav1.9, play a crucial role in neuronal excitability and are predominantly expressed in the peripheral nervous system. Targeting these channels may provide a means to treat pain while minimizing central and cardiac adverse effects. In this study, we construct protein-protein interaction (PPI) networks based on pain-related sodium channels and develop a corresponding drug-target interaction network to identify potential lead compounds for pain management. To ensure reliable machine learning predictions, we carefully select 111 inhibitor data sets from a pool of more than 1000 targets in the PPI network. We employ 3 distinct machine learning algorithms combined with advanced natural language processing (NLP)-based embeddings, specifically pretrained transformer and autoencoder representations. Through a systematic screening process, we evaluate the side effects and repurposing potential of more than 150,000 drug candidates targeting Nav1.7 and Nav1.8 sodium channels. In addition, we assess the ADMET (absorption, distribution, metabolism, excretion, and toxicity) properties of these candidates to identify leads with near-optimal characteristics. Our strategy provides an innovative platform for the pharmacological development of pain treatments, offering the potential for improved efficacy and reduced side effects.

GLDM: hit molecule generation with constrained graph latent diffusion model

Discovering hit molecules with desired biological activity in a directed manner is a promising but profound task in computer-aided drug discovery. Inspired by recent generative AI approaches, particularly Diffusion Models (DM), we propose Graph Latent Diffusion Model (GLDM)-a latent DM that preserves both the effectiveness of autoencoders of compressing complex chemical data and the DM's capabilities of generating novel molecules. Specifically, we first develop an autoencoder to encode the molecular data into low-dimensional latent representations and then train the DM on the latent space to generate molecules inducing targeted biological activity defined by gene expression profiles. Manipulating DM in the latent space rather than the input space avoids complicated operations to map molecule decomposition and reconstruction to diffusion processes, and thus improves training efficiency. Experiments show that GLDM not only achieves outstanding performances on molecular generation benchmarks, but also generates samples with optimal chemical properties and potentials to induce desired biological activity.

Bidirectional generation of structure and properties through a single molecular foundation model

Recent successes of foundation models in artificial intelligence have prompted the emergence of large-scale chemical pre-trained models. Despite the growing interest in large molecular pre-trained models that provide informative representations for downstream tasks, attempts for multimodal pre-training approaches on the molecule domain were limited. To address this, here we present a multimodal molecular pre-trained model that incorporates the modalities of structure and biochemical properties, drawing inspiration from recent advances in multimodal learning techniques. Our proposed model pipeline of data handling and training objectives aligns the structure/property features in a common embedding space, which enables the model to regard bidirectional information between the molecules' structure and properties. These contributions emerge synergistic knowledge, allowing us to tackle both multimodal and unimodal downstream tasks through a single model. Through extensive experiments, we demonstrate that our model has the capabilities to solve various meaningful chemical challenges, including conditional molecule generation, property prediction, molecule classification, and reaction prediction.

The openOCHEM consensus model is the best-performing open-source predictive model in the First EUOS/SLAS joint compound solubility challenge

The EUOS/SLAS challenge aimed to facilitate the development of reliable algorithms to predict the aqueous solubility of small molecules using experimental data from 100 K compounds. In total, hundred teams took part in the challenge to predict low, medium and highly soluble compounds as measured by the nephelometry assay. This article describes the winning model, which was developed using the publicly available Online CHEmical database and Modeling environment (OCHEM) available on the website https://ochem.eu/article/27. We describe in detail the assumptions and steps used to select methods, descriptors and strategy which contributed to the winning solution. In particular we show that consensus based on 28 models calculated using descriptor-based and representation learning methods allowed us to obtain the best score, which was higher than those based on individual approaches or consensus models developed using each individual approach. A combination of diverse models allowed us to decrease both bias and variance of individual models and to calculate the highest score. The model based on Transformer CNN contributed the best individual score thus highlighting the power of Natural Language Processing (NLP) methods. The inclusion of information about aleatoric uncertainty would be important to better understand and use the challenge data by the contestants.

COATI: Multimodal Contrastive Pretraining for Representing and Traversing Chemical Space

Creating a successful small molecule drug is a challenging multiparameter optimization problem in an effectively infinite space of possible molecules. Generative models have emerged as powerful tools for traversing data manifolds composed of images, sounds, and text and offer an opportunity to dramatically improve the drug discovery and design process. To create generative optimization methods that are more useful than brute-force molecular generation and filtering via virtual screening, we propose that four integrated features are necessary: large, quantitative data sets of molecular structure and activity, an invertible vector representation of realistic accessible molecules, smooth and differentiable regressors that quantify uncertainty, and algorithms to simultaneously optimize properties of interest. Over the course of 12 months, Terray Therapeutics has collected a data set of 2 billion quantitative binding measurements of small molecules to therapeutic targets, which directly motivates multiparameter generative optimization of molecules conditioned on these data. To this end, we present contrastive optimization for accelerated therapeutic inference (COATI), a pretrained, multimodal encoder-decoder model of druglike chemical space. COATI is constructed without any human biasing of features, using contrastive learning from text and 3D representations of molecules to allow for downstream use with structural models. We demonstrate that COATI possesses many of the desired properties of universal molecular embedding: fixed-dimension, invertibility, autoencoding, accurate regression, and low computation cost. Finally, we present a novel metadynamics algorithm for generative optimization using a small subset of our proprietary data collected for a model protein, carbonic anhydrase, designing molecules that satisfy the multiparameter optimization task of potency, solubility, and drug likeness. This work sets the stage for fully integrated generative molecular design and optimization for small molecules.

Difficulty in chirality recognition for Transformer architectures learning chemical structures from string representations

Recent years have seen rapid development of descriptor generation based on representation learning of extremely diverse molecules, especially those that apply natural language processing (NLP) models to SMILES, a literal representation of molecular structure. However, little research has been done on how these models understand chemical structure. To address this black box, we investigated the relationship between the learning progress of SMILES and chemical structure using a representative NLP model, the Transformer. We show that while the Transformer learns partial structures of molecules quickly, it requires extended training to understand overall structures. Consistently, the accuracy of molecular property predictions using descriptors generated from models at different learning steps was similar from the beginning to the end of training. Furthermore, we found that the Transformer requires particularly long training to learn chirality and sometimes stagnates with low performance due to misunderstanding of enantiomers. These findings are expected to deepen the understanding of NLP models in chemistry.

Quantification of chemicals in non-targeted analysis without analytical standards - Understanding the mechanism of electrospray ionization and making predictions

The constant creation and release of new chemicals to the environment is forming an ever-widening gap between available analytical standards and known chemicals. Developing non-targeted analysis (NTA) methods that have the ability to detect a broad spectrum of compounds is critical for research and analysis of emerging contaminants. There is a need to develop methods that make it possible to identify compound structures from their MS and MS/MS information and quantify them without analytical standards. Method refinements that utilize machine learning algorithms and chemical descriptors to estimate the instrument response of particular compounds have made progress in recent years. This narrative review seeks to summarize the current state of the field of non-targeted analysis (NTA) toward quantification of unknowns without the use of analytical standards. Despite the limited accumulation of validation studies on real samples, the ongoing enhancement in data processing and refinement of machine learning tools could lead to more comprehensive chemical coverage of NTA and validated quantitative NTA methods, thus boosting confidence in their usage and enhancing the utility of quantitative NTA.

A deep neural network: mechanistic hybrid model to predict pharmacokinetics in rat

An important aspect in the development of small molecules as drugs or agrochemicals is their systemic availability after intravenous and oral administration. The prediction of the systemic availability from the chemical structure of a potential candidate is highly desirable, as it allows to focus the drug or agrochemical development on compounds with a favorable kinetic profile. However, such predictions are challenging as the availability is the result of the complex interplay between molecular properties, biology and physiology and training data is rare. In this work we improve the hybrid model developed earlier (Schneckener in J Chem Inf Model 59:4893-4905, 2019). We reduce the median fold change error for the total oral exposure from 2.85 to 2.35 and for intravenous administration from 1.95 to 1.62. This is achieved by training on a larger data set, improving the neural network architecture as well as the parametrization of mechanistic model. Further, we extend our approach to predict additional endpoints and to handle different covariates, like sex and dosage form. In contrast to a pure machine learning model, our model is able to predict new end points on which it has not been trained. We demonstrate this feature by predicting the exposure over the first 24 h, while the model has only been trained on the total exposure.

Multiscale topology in interactomic network: from transcriptome to antiaddiction drug repurposing

The escalating drug addiction crisis in the United States underscores the urgent need for innovative therapeutic strategies. This study embarked on an innovative and rigorous strategy to unearth potential drug repurposing candidates for opioid and cocaine addiction treatment, bridging the gap between transcriptomic data analysis and drug discovery. We initiated our approach by conducting differential gene expression analysis on addiction-related transcriptomic data to identify key genes. We propose a novel topological differentiation to identify key genes from a protein-protein interaction network derived from DEGs. This method utilizes persistent Laplacians to accurately single out pivotal nodes within the network, conducting this analysis in a multiscale manner to ensure high reliability. Through rigorous literature validation, pathway analysis and data-availability scrutiny, we identified three pivotal molecular targets, mTOR, mGluR5 and NMDAR, for drug repurposing from DrugBank. We crafted machine learning models employing two natural language processing (NLP)-based embeddings and a traditional 2D fingerprint, which demonstrated robust predictive ability in gauging binding affinities of DrugBank compounds to selected targets. Furthermore, we elucidated the interactions of promising drugs with the targets and evaluated their drug-likeness. This study delineates a multi-faceted and comprehensive analytical framework, amalgamating bioinformatics, topological data analysis and machine learning, for drug repurposing in addiction treatment, setting the stage for subsequent experimental validation. The versatility of the methods we developed allows for applications across a range of diseases and transcriptomic datasets.

ChemMORT: an automatic ADMET optimization platform using deep learning and multi-objective particle swarm optimization

Drug discovery and development constitute a laborious and costly undertaking. The success of a drug hinges not only good efficacy but also acceptable absorption, distribution, metabolism, elimination, and toxicity (ADMET) properties. Overall, up to 50% of drug development failures have been contributed from undesirable ADMET profiles. As a multiple parameter objective, the optimization of the ADMET properties is extremely challenging owing to the vast chemical space and limited human expert knowledge. In this study, a freely available platform called Chemical Molecular Optimization, Representation and Translation (ChemMORT) is developed for the optimization of multiple ADMET endpoints without the loss of potency (https://cadd.nscc-tj.cn/deploy/chemmort/). ChemMORT contains three modules: Simplified Molecular Input Line Entry System (SMILES) Encoder, Descriptor Decoder and Molecular Optimizer. The SMILES Encoder can generate the molecular representation with a 512-dimensional vector, and the Descriptor Decoder is able to translate the above representation to the corresponding molecular structure with high accuracy. Based on reversible molecular representation and particle swarm optimization strategy, the Molecular Optimizer can be used to effectively optimize undesirable ADMET properties without the loss of bioactivity, which essentially accomplishes the design of inverse QSAR. The constrained multi-objective optimization of the poly (ADP-ribose) polymerase-1 inhibitor is provided as the case to explore the utility of ChemMORT.

Artificial Intelligence and Machine Learning Methods to Evaluate Cardiotoxicity following the Adverse Outcome Pathway Frameworks

Cardiovascular disease is a leading global cause of mortality. The potential cardiotoxic effects of chemicals from different classes, such as environmental contaminants, pesticides, and drugs can significantly contribute to effects on health. The same chemical can induce cardiotoxicity in different ways, following various Adverse Outcome Pathways (AOPs). In addition, the potential synergistic effects between chemicals further complicate the issue. In silico methods have become essential for tackling the problem from different perspectives, reducing the need for traditional in vivo testing, and saving valuable resources in terms of time and money. Artificial intelligence (AI) and machine learning (ML) are among today's advanced approaches for evaluating chemical hazards. They can serve, for instance, as a first-tier component of Integrated Approaches to Testing and Assessment (IATA). This study employed ML and AI to assess interactions between chemicals and specific biological targets within the AOP networks for cardiotoxicity, starting with molecular initiating events (MIEs) and progressing through key events (KEs). We explored methods to encode chemical information in a suitable way for ML and AI. We started with commonly used approaches in Quantitative Structure-Activity Relationship (QSAR) methods, such as molecular descriptors and different types of fingerprint. We then increased the complexity of encoders, incorporating graph-based methods, auto-encoders, and character embeddings employed in neural language processing. We also developed a multimodal neural network architecture, capable of considering the complementary nature of different chemical representations simultaneously. The potential of this approach, compared to more conventional architectures designed to handle a single encoder, becomes apparent when the amount of data increases.

Bayesian Optimization in Drug Discovery

Drug discovery deals with the search for initial hits and their optimization toward a targeted clinical profile. Throughout the discovery pipeline, the candidate profile will evolve, but the optimization will mainly stay a trial-and-error approach. Tons of in silico methods have been developed to improve and fasten this pipeline. Bayesian optimization (BO) is a well-known method for the determination of the global optimum of a function. In the last decade, BO has gained popularity in the early drug design phase. This chapter starts with the concept of black box optimization applied to drug design and presents some approaches to tackle it. Then it focuses on BO and explains its principle and all the algorithmic building blocks needed to implement it. This explanation aims to be accessible to people involved in drug discovery projects. A strong emphasis is made on the solutions to deal with the specific constraints of drug discovery. Finally, a large set of practical applications of BO is highlighted.

Design and Diversity Analysis of Chemical Libraries in Drug Discovery

Chemical libraries and compound data sets are among the main inputs to start the drug discovery process at universities, research institutes, and the pharmaceutical industry. The approach used in the design of compound libraries, the chemical information they possess, and the representation of structures, play a fundamental role in the development of studies: chemoinformatics, food informatics, in silico pharmacokinetics, computational toxicology, bioinformatics, and molecular modeling to generate computational hits that will continue the optimization process of drug candidates. The prospects for growth in drug discovery and development processes in chemical, biotechnological, and pharmaceutical companies began a few years ago by integrating computational tools with artificial intelligence methodologies. It is anticipated that it will increase the number of drugs approved by regulatory agencies shortly.

A community effort in SARS-CoV-2 drug discovery

The COVID-19 pandemic continues to pose a substantial threat to human lives and is likely to do so for years to come. Despite the availability of vaccines, searching for efficient small-molecule drugs that are widely available, including in low- and middle-income countries, is an ongoing challenge. In this work, we report the results of an open science community effort, the "Billion molecules against COVID-19 challenge", to identify small-molecule inhibitors against SARS-CoV-2 or relevant human receptors. Participating teams used a wide variety of computational methods to screen a minimum of 1 billion virtual molecules against 6 protein targets. Overall, 31 teams participated, and they suggested a total of 639,024 molecules, which were subsequently ranked to find 'consensus compounds'. The organizing team coordinated with various contract research organizations (CROs) and collaborating institutions to synthesize and test 878 compounds for biological activity against proteases (Nsp5, Nsp3, TMPRSS2), nucleocapsid N, RdRP (only the Nsp12 domain), and (alpha) spike protein S. Overall, 27 compounds with weak inhibition/binding were experimentally identified by binding-, cleavage-, and/or viral suppression assays and are presented here. Open science approaches such as the one presented here contribute to the knowledge base of future drug discovery efforts in finding better SARS-CoV-2 treatments.

Transformer-Based Representation of Organic Molecules for Potential Modeling of Physicochemical Properties

In this work, we study the use of three configurations of an autoencoder neural network to process organic substances with the aim of generating meaningful molecular descriptors that can be employed to develop property prediction models. A total of 18,322,500 compounds represented as SMILES strings were used to train the model, demonstrating that a latent space of 24 units is able to adequately reconstruct the data. After AE training, an analysis of the latent space properties in terms of compound similarity was carried out, indicating that this space possesses desired properties for the potential development of models for forecasting physical properties of organic compounds. As a final step, a QSPR model was developed to predict the boiling point of chemical substances based on the AE descriptors. 5276 substances were used for the regression task, and the predictive ability was compared with models available in the literature evaluated on the same database. The final AE model has an overall error of 1.40% (1.39% with augmented SMILES) in the prediction of the boiling temperature, while other models have errors between 2.0 and 3.2%. This shows that the SMILES representation is comparable and even outperforms the state-of-the-art representations widely used in the literature.

Deep Learning-Enabled MS/MS Spectrum Prediction Facilitates Automated Identification Of Novel Psychoactive Substances

The market for illicit drugs has been reshaped by the emergence of more than 1100 new psychoactive substances (NPS) over the past decade, posing a major challenge to the forensic and toxicological laboratories tasked with detecting and identifying them. Tandem mass spectrometry (MS/MS) is the primary method used to screen for NPS within seized materials or biological samples. The most contemporary workflows necessitate labor-intensive and expensive MS/MS reference standards, which may not be available for recently emerged NPS on the illicit market. Here, we present NPS-MS, a deep learning method capable of accurately predicting the MS/MS spectra of known and hypothesized NPS from their chemical structures alone. NPS-MS is trained by transfer learning from a generic MS/MS prediction model on a large data set of MS/MS spectra. We show that this approach enables a more accurate identification of NPS from experimentally acquired MS/MS spectra than any existing method. We demonstrate the application of NPS-MS to identify a novel derivative of phencyclidine (PCP) within an unknown powder seized in Denmark without the use of any reference standards. We anticipate that NPS-MS will allow forensic laboratories to identify more rapidly both known and newly emerging NPS. NPS-MS is available as a web server at https://nps-ms.ca/, which provides MS/MS spectra prediction capabilities for given NPS compounds. Additionally, it offers MS/MS spectra identification against a vast database comprising approximately 8.7 million predicted NPS compounds from DarkNPS and 24.5 million predicted ESI-QToF-MS/MS spectra for these compounds.

ChatGPT in Drug Discovery: A Case Study on Anticocaine Addiction Drug Development with Chatbots

The birth of ChatGPT, a cutting-edge language model-based chatbot developed by OpenAI, ushered in a new era in AI. However, due to potential pitfalls, its role in rigorous scientific research is not clear yet. This paper vividly showcases its innovative application within the field of drug discovery. Focused specifically on developing anticocaine addiction drugs, the study employs GPT-4 as a virtual guide, offering strategic and methodological insights to researchers working on generative models for drug candidates. The primary objective is to generate optimal drug-like molecules with desired properties. By leveraging the capabilities of ChatGPT, the study introduces a novel approach to the drug discovery process. This symbiotic partnership between AI and researchers transforms how drug development is approached. Chatbots become facilitators, steering researchers toward innovative methodologies and productive paths for creating effective drug candidates. This research sheds light on the collaborative synergy between human expertise and AI assistance, wherein ChatGPT's cognitive abilities enhance the design and development of pharmaceutical solutions. This paper not only explores the integration of advanced AI in drug discovery but also reimagines the landscape by advocating for AI-powered chatbots as trailblazers in revolutionizing therapeutic innovation.

BatmanNet: bi-branch masked graph transformer autoencoder for molecular representation

Although substantial efforts have been made using graph neural networks (GNNs) for artificial intelligence (AI)-driven drug discovery, effective molecular representation learning remains an open challenge, especially in the case of insufficient labeled molecules. Recent studies suggest that big GNN models pre-trained by self-supervised learning on unlabeled datasets enable better transfer performance in downstream molecular property prediction tasks. However, the approaches in these studies require multiple complex self-supervised tasks and large-scale datasets , which are time-consuming, computationally expensive and difficult to pre-train end-to-end. Here, we design a simple yet effective self-supervised strategy to simultaneously learn local and global information about molecules, and further propose a novel bi-branch masked graph transformer autoencoder (BatmanNet) to learn molecular representations. BatmanNet features two tailored complementary and asymmetric graph autoencoders to reconstruct the missing nodes and edges, respectively, from a masked molecular graph. With this design, BatmanNet can effectively capture the underlying structure and semantic information of molecules, thus improving the performance of molecular representation. BatmanNet achieves state-of-the-art results for multiple drug discovery tasks, including molecular properties prediction, drug-drug interaction and drug-target interaction, on 13 benchmark datasets, demonstrating its great potential and superiority in molecular representation learning.

Molecular geometric deep learning

Molecular representation learning plays an important role in molecular property prediction. Existing molecular property prediction models rely on the de facto standard of covalent-bond-based molecular graphs for representing molecular topology at the atomic level and totally ignore the non-covalent interactions within the molecule. In this study, we propose a molecular geometric deep learning model to predict the properties of molecules that aims to comprehensively consider the information of covalent and non-covalent interactions of molecules. The essential idea is to incorporate a more general molecular representation into geometric deep learning (GDL) models. We systematically test molecular GDL (Mol-GDL) on fourteen commonly used benchmark datasets. The results show that Mol-GDL can achieve a better performance than state-of-the-art (SOTA) methods. Extensive tests have demonstrated the important role of non-covalent interactions in molecular property prediction and the effectiveness of Mol-GDL models.

Deep Learning with Geometry-Enhanced Molecular Representation for Augmentation of Large-Scale Docking-Based Virtual Screening

Structure-based virtual screening has been a crucial tool in drug discovery for decades. However, as the chemical space expands, the existing structure-based virtual screening techniques based on molecular docking and scoring struggle to handle billion-entry ultralarge libraries due to the high computational cost. To address this challenge, people have resorted to machine learning techniques to enhance structure-based virtual screening for efficiently exploring the vast chemical space. In those cases, compounds are usually treated as sequential strings or two-dimensional topology graphs, limiting their ability to incorporate three-dimensional structural information for downstream tasks. We herein propose a novel deep learning protocol, GEM-Screen, which utilizes the geometry-enhanced molecular representation of the compounds docking to a specific target and is trained on docking scores of a small fraction of a library through an active learning strategy to approximate the docking outcome for yet nontraining entries. This protocol is applied to virtual screening campaigns against the AmpC and D4 targets, demonstrating that GEM-Screen enriches more than 90% of the hit scaffolds for AmpC in the top 4% of model predictions and more than 80% of the hit scaffolds for D4 in the same top-ranking size of library. GEM-Screen can be used in conjunction with traditional docking programs for docking of only the top-ranked compounds to avoid the exhaustive docking of the whole library, thus allowing for discovering top-scoring compounds from billion-entry libraries in a rapid yet accurate fashion.

Designing molecules with autoencoder networks

Autoencoders are versatile tools in molecular informatics. These unsupervised neural networks serve diverse tasks such as data-driven molecular representation and constructive molecular design. This Review explores their algorithmic foundations and applications in drug discovery, highlighting the most active areas of development and the contributions autoencoder networks have made in advancing this field. We also explore the challenges and prospects concerning the utilization of autoencoders and the various adaptations of this neural network architecture in molecular design.

Artificial intelligence for natural product drug discovery

Developments in computational omics technologies have provided new means to access the hidden diversity of natural products, unearthing new potential for drug discovery. In parallel, artificial intelligence approaches such as machine learning have led to exciting developments in the computational drug design field, facilitating biological activity prediction and de novo drug design for molecular targets of interest. Here, we describe current and future synergies between these developments to effectively identify drug candidates from the plethora of molecules produced by nature. We also discuss how to address key challenges in realizing the potential of these synergies, such as the need for high-quality datasets to train deep learning algorithms and appropriate strategies for algorithm validation.

ChatGPT in Drug Discovery: A Case Study on Anti-Cocaine Addiction Drug Development with Chatbots

The birth of ChatGPT, a cutting-edge language model-based chatbot developed by OpenAI, ushered in a new era in AI. However, due to potential pitfalls, its role in rigorous scientific research is not clear yet. This paper vividly showcases its innovative application within the field of drug discovery. Focused specifically on developing anti-cocaine addiction drugs, the study employs GPT-4 as a virtual guide, offering strategic and methodological insights to researchers working on generative models for drug candidates. The primary objective is to generate optimal drug-like molecules with desired properties. By leveraging the capabilities of ChatGPT, the study introduces a novel approach to the drug discovery process. This symbiotic partnership between AI and researchers transforms how drug development is approached. Chatbots become facilitators, steering researchers towards innovative methodologies and productive paths for creating effective drug candidates. This research sheds light on the collaborative synergy between human expertise and AI assistance, wherein ChatGPT's cognitive abilities enhance the design and development of potential pharmaceutical solutions. This paper not only explores the integration of advanced AI in drug discovery but also reimagines the landscape by advocating for AI-powered chatbots as trailblazers in revolutionizing therapeutic innovation.

Efficient 3D Molecular Design with an E(3) Invariant Transformer VAE

This work introduces a three-dimensional (3D) invariant graph-to-string transformer variational autoencoders (VAE) (Vagrant) for generating molecules with accurate density functional theory (DFT)-level properties. Vagrant learns to model the joint probability distribution of a 3D molecular structure and its properties by encoding molecular structures into a 3D-aware latent space. Directed navigation through this latent space implicitly optimizes the 3D structure of a molecule, and the latent embedding can be used to condition a generative transformer to predict the candidate structure as a one-dimensional (1D) sequence. Additionally, we introduce two novel sampling methods that exploit the latent characteristics of a VAE to improve performance. We show that our method outperforms comparable 3D autoregressive and diffusion methods for predicting quantum chemical property values of novel molecules in terms of both sample quality and computational efficiency.

Multiobjective Molecular Optimization for Opioid Use Disorder Treatment Using Generative Network Complex

Opioid use disorder (OUD) has emerged as a significant global public health issue, necessitating the discovery of new medications. In this study, we propose a deep generative model that combines a stochastic differential equation (SDE)-based diffusion model with a pretrained autoencoder. The molecular generator enables efficient generation of molecules that target multiple opioid receptors, including mu, kappa, and delta. Additionally, we assess the ADMET (absorption, distribution, metabolism, excretion, and toxicity) properties of the generated molecules to identify druglike compounds. We develop a molecular optimization approach to enhance the pharmacokinetic properties of some lead compounds. Advanced binding affinity predictors were built using molecular fingerprints, including autoencoder embeddings, transformer embeddings, and topological Laplacians. Our process yields druglike molecules that can be used in highly focused experimental studies to further evaluate their pharmacological effects. Our machine learning platform serves as a valuable tool for designing effective molecules to address OUD.

Equivariant Graph Neural Networks for Toxicity Prediction

Predictive modeling of toxicity is a crucial step in the drug discovery pipeline. It can help filter out molecules with a high probability of failing in the early stages of de novo drug design. Thus, several machine learning (ML) models have been developed to predict the toxicity of molecules by combining classical ML techniques or deep neural networks with well-known molecular representations such as fingerprints or 2D graphs. But the more natural, accurate representation of molecules is expected to be defined in physical 3D space like in ab initio methods. Recent studies successfully used equivariant graph neural networks (EGNNs) for representation learning based on 3D structures to predict quantum-mechanical properties of molecules. Inspired by this, we investigated the performance of EGNNs to construct reliable ML models for toxicity prediction. We used the equivariant transformer (ET) model in TorchMD-NET for this. Eleven toxicity data sets taken from MoleculeNet, TDCommons, and ToxBenchmark have been considered to evaluate the capability of ET for toxicity prediction. Our results show that ET adequately learns 3D representations of molecules that can successfully correlate with toxicity activity, achieving good accuracies on most data sets comparable to state-of-the-art models. We also test a physicochemical property, namely, the total energy of a molecule, to inform the toxicity prediction with a physical prior. However, our work suggests that these two properties can not be related. We also provide an attention weight analysis for helping to understand the toxicity prediction in 3D space and thus increase the explainability of the ML model. In summary, our findings offer promising insights considering 3D geometry information via EGNNs and provide a straightforward way to integrate molecular conformers into ML-based pipelines for predicting and investigating toxicity prediction in physical space. We expect that in the future, especially for larger, more diverse data sets, EGNNs will be an essential tool in this domain.

Similarity-based pairing improves efficiency of siamese neural networks for regression tasks and uncertainty quantification

Siamese networks, representing a novel class of neural networks, consist of two identical subnetworks sharing weights but receiving different inputs. Here we present a similarity-based pairing method for generating compound pairs to train Siamese neural networks for regression tasks. In comparison with the conventional exhaustive pairing, it reduces the algorithm complexity from O(n2) to O(n). It also results in a better prediction performance consistently on the three physicochemical datasets, using a multilayer perceptron with the circular fingerprint as a proof of concept. We further include into a Siamese neural network the transformer-based Chemformer, which extracts task-specific features from the simplified molecular-input line-entry system representation of compounds. Additionally, we propose a means to measure the prediction uncertainty by utilizing the variance in predictions from a set of reference compounds. Our results demonstrate that the high prediction accuracy correlates with the high confidence. Finally, we investigate implications of the similarity property principle in machine learning.

Practical guidelines for the use of gradient boosting for molecular property prediction

Decision tree ensembles are among the most robust, high-performing and computationally efficient machine learning approaches for quantitative structure-activity relationship (QSAR) modeling. Among them, gradient boosting has recently garnered particular attention, for its performance in data science competitions, virtual screening campaigns, and bioactivity prediction. However, different variants of gradient boosting exist, the most popular being XGBoost, LightGBM and CatBoost. Our study provides the first comprehensive comparison of these approaches for QSAR. To this end, we trained 157,590 gradient boosting models, which were evaluated on 16 datasets and 94 endpoints, comprising 1.4 million compounds in total. Our results show that XGBoost generally achieves the best predictive performance, while LightGBM requires the least training time, especially for larger datasets. In terms of feature importance, the models surprisingly rank molecular features differently, reflecting differences in regularization techniques and decision tree structures. Thus, expert knowledge must always be employed when evaluating data-driven explanations of bioactivity. Furthermore, our results show that the relevance of each hyperparameter varies greatly across datasets and that it is crucial to optimize as many hyperparameters as possible to maximize the predictive performance. In conclusion, our study provides the first set of guidelines for cheminformatics practitioners to effectively train, optimize and evaluate gradient boosting models for virtual screening and QSAR applications.

PREFER: A New Predictive Modeling Framework for Molecular Discovery

Machine-learning and deep-learning models have been extensively used in cheminformatics to predict molecular properties, to reduce the need for direct measurements, and to accelerate compound prioritization. However, different setups and frameworks and the large number of molecular representations make it difficult to properly evaluate, reproduce, and compare them. Here we present a new PREdictive modeling FramEwoRk for molecular discovery (PREFER), written in Python (version 3.7.7) and based on AutoSklearn (version 0.14.7), that allows comparison between different molecular representations and common machine-learning models. We provide an overview of the design of our framework and show exemplary use cases and results of several representation-model combinations on diverse data sets, both public and in-house. Finally, we discuss the use of PREFER on small data sets. The code of the framework is freely available on GitHub.

Employing Molecular Conformations for Ligand-Based Virtual Screening with Equivariant Graph Neural Network and Deep Multiple Instance Learning

Ligand-based virtual screening (LBVS) is a promising approach for rapid and low-cost screening of potentially bioactive molecules in the early stage of drug discovery. Compared with traditional similarity-based machine learning methods, deep learning frameworks for LBVS can more effectively extract high-order molecule structure representations from molecular fingerprints or structures. However, the 3D conformation of a molecule largely influences its bioactivity and physical properties, and has rarely been considered in previous deep learning-based LBVS methods. Moreover, the relative bioactivity benchmark dataset is still lacking. To address these issues, we introduce a novel end-to-end deep learning architecture trained from molecular conformers for LBVS. We first extracted molecule conformers from multiple public molecular bioactivity data and consolidated them into a large-scale bioactivity benchmark dataset, which totally includes millions of endpoints and molecules corresponding to 954 targets. Then, we devised a deep learning-based LBVS called EquiVS to learn molecule representations from conformers for bioactivity prediction. Specifically, graph convolutional network (GCN) and equivariant graph neural network (EGNN) are sequentially stacked to learn high-order molecule-level and conformer-level representations, followed with attention-based deep multiple-instance learning (MIL) to aggregate these representations and then predict the potential bioactivity for the query molecule on a given target. We conducted various experiments to validate the data quality of our benchmark dataset, and confirmed EquiVS achieved better performance compared with 10 traditional machine learning or deep learning-based LBVS methods. Further ablation studies demonstrate the significant contribution of molecular conformation for bioactivity prediction, as well as the reasonability and non-redundancy of deep learning architecture in EquiVS. Finally, a model interpretation case study on CDK2 shows the potential of EquiVS in optimal conformer discovery. The overall study shows that our proposed benchmark dataset and EquiVS method have promising prospects in virtual screening applications.

De novo drug design based on patient gene expression profiles via deep learning

Computational de novo drug design is a challenging issue in medicine, and it is desirable to consider all of the relevant information of the biological systems in a disease state. Here, we propose a novel computational method to generate drug candidate molecular structures from patient gene expression profiles via deep learning, which we call DRAGONET. Our model can generate new molecules that are likely to counteract disease-specific gene expression patterns in patients, which is made possible by exploring the latent space constructed by a transformer-based variational autoencoder and integrating the substructures of disease-correlated molecules. We applied DRAGONET to generate drug candidate molecules for gastric cancer, atopic dermatitis, and Alzheimer's disease, and demonstrated that the newly generated molecules were chemically similar to registered drugs for each disease. This approach is applicable to diseases with unknown therapeutic target proteins and will make a significant contribution to the field of precision medicine.

An end-to-end deep learning framework for translating mass spectra to de-novo molecules

Elucidating the structure of a chemical compound is a fundamental task in chemistry with applications in multiple domains including drug discovery, precision medicine, and biomarker discovery. The common practice for elucidating the structure of a compound is to obtain a mass spectrum and subsequently retrieve its structure from spectral databases. However, these methods fail for novel molecules that are not present in the reference database. We propose Spec2Mol, a deep learning architecture for molecular structure recommendation given mass spectra alone. Spec2Mol is inspired by the Speech2Text deep learning architectures for translating audio signals into text. Our approach is based on an encoder-decoder architecture. The encoder learns the spectra embeddings, while the decoder, pre-trained on a massive dataset of chemical structures for translating between different molecular representations, reconstructs SMILES sequences of the recommended chemical structures. We have evaluated Spec2Mol by assessing the molecular similarity between the recommended structures and the original structure. Our analysis showed that Spec2Mol is able to identify the presence of key molecular substructures from its mass spectrum, and shows on par performance, when compared to existing fragmentation tree methods particularly when test structure information is not available during training or present in the reference database.

Combatting over-specialization bias in growing chemical databases

BACKGROUND

Predicting in advance the behavior of new chemical compounds can support the design process of new products by directing the research toward the most promising candidates and ruling out others. Such predictive models can be data-driven using Machine Learning or based on researchers' experience and depend on the collection of past results. In either case: models (or researchers) can only make reliable assumptions about compounds that are similar to what they have seen before. Therefore, consequent usage of these predictive models shapes the dataset and causes a continuous specialization shrinking the applicability domain of all trained models on this dataset in the future, and increasingly harming model-based exploration of the space.

PROPOSED SOLUTION

In this paper, we propose CANCELS (CounterActiNg Compound spEciaLization biaS), a technique that helps to break the dataset specialization spiral. Aiming for a smooth distribution of the compounds in the dataset, we identify areas in the space that fall short and suggest additional experiments that help bridge the gap. Thereby, we generally improve the dataset quality in an entirely unsupervised manner and create awareness of potential flaws in the data. CANCELS does not aim to cover the entire compound space and hence retains a desirable degree of specialization to a specified research domain.

RESULTS

An extensive set of experiments on the use-case of biodegradation pathway prediction not only reveals that the bias spiral can indeed be observed but also that CANCELS produces meaningful results. Additionally, we demonstrate that mitigating the observed bias is crucial as it cannot only intervene with the continuous specialization process, but also significantly improves a predictor's performance while reducing the number of required experiments. Overall, we believe that CANCELS can support researchers in their experimentation process to not only better understand their data and potential flaws, but also to grow the dataset in a sustainable way. All code is available under github.com/KatDost/Cancels .