Gram matrix: an efficient representation of molecular conformation and learning objective for molecular pretraining.

Accurate prediction of molecular properties is fundamental in drug discovery and development, providing crucial guidance for effective drug design. A critical factor in achieving accurate molecular property prediction lies in the appropriate representation of molecular structures. Presently, prevalent deep learning-based molecular representations rely on 2D structure information as the primary molecular representation, often overlooking essential three-dimensional (3D) conformational information due to the inherent limitations of 2D structures in conveying atomic spatial relationships. In this study, we propose employing the Gram matrix as a condensed representation of 3D molecular structures and for efficient pretraining objectives. Subsequently, we leverage this matrix to construct a novel molecular representation model, Pre-GTM, which inherently encapsulates 3D information. The model accurately predicts the 3D structure of a molecule by estimating the Gram matrix. Our findings demonstrate that Pre-GTM model outperforms the baseline Graphormer model and other pretrained models in the QM9 and MoleculeNet quantitative property prediction task. The integration of the Gram matrix as a condensed representation of 3D molecular structure, incorporated into the Pre-GTM model, opens up promising avenues for its potential application across various domains of molecular research, including drug design, materials science, and chemical engineering.

KinomeMETA: meta-learning enhanced kinome-wide polypharmacology profiling.

Kinase inhibitors are crucial in cancer treatment, but drug resistance and side effects hinder the development of effective drugs. To address these challenges, it is essential to analyze the polypharmacology of kinase inhibitor and identify compound with high selectivity profile. This study presents KinomeMETA, a framework for profiling the activity of small molecule kinase inhibitors across a panel of 661 kinases. By training a meta-learner based on a graph neural network and fine-tuning it to create kinase-specific learners, KinomeMETA outperforms benchmark multi-task models and other kinase profiling models. It provides higher accuracy for understudied kinases with limited known data and broader coverage of kinase types, including important mutant kinases. Case studies on the discovery of new scaffold inhibitors for membrane-associated tyrosine- and threonine-specific cdc2-inhibitory kinase and selective inhibitors for fibroblast growth factor receptors demonstrate the role of KinomeMETA in virtual screening and kinome-wide activity profiling. Overall, KinomeMETA has the potential to accelerate kinase drug discovery by more effectively exploring the kinase polypharmacology landscape.

An inductive graph neural network model for compound-protein interaction prediction based on a homogeneous graph.

Identifying the potential compound-protein interactions (CPIs) plays an essential role in drug development. The computational approaches for CPI prediction can reduce time and costs of experimental methods and have benefited from the continuously improved graph representation learning. However, most of the network-based methods use heterogeneous graphs, which is challenging due to their complex structures and heterogeneous attributes. Therefore, in this work, we transformed the compound-protein heterogeneous graph to a homogeneous graph by integrating the ligand-based protein representations and overall similarity associations. We then proposed an Inductive Graph AggrEgator-based framework, named CPI-IGAE, for CPI prediction. CPI-IGAE learns the low-dimensional representations of compounds and proteins from the homogeneous graph in an end-to-end manner. The results show that CPI-IGAE performs better than some state-of-the-art methods. Further ablation study and visualization of embeddings reveal the advantages of the model architecture and its role in feature extraction, and some of the top ranked CPIs by CPI-IGAE have been validated by a review of recent literature. The data and source codes are available at https://github.com/wanxiaozhe/CPI-IGAE.

Drug repurposing against breast cancer by integrating drug-exposure expression profiles and drug-drug links based on graph neural network.

MOTIVATION: Breast cancer is one of the leading causes of cancer deaths among women worldwide. It is necessary to develop new breast cancer drugs because of the shortcomings of existing therapies. The traditional discovery process is time-consuming and expensive. Repositioning of clinically approved drugs has emerged as a novel approach for breast cancer therapy. However, serendipitous or experiential repurposing cannot be used as a routine method. RESULTS: In this study, we proposed a graph neural network model GraphRepur based on GraphSAGE for drug repurposing against breast cancer. GraphRepur integrated two major classes of computational methods, drug network-based and drug signature-based. The differentially expressed genes of disease, drug-exposure gene expression data and the drug-drug links information were collected. By extracting the drug signatures and topological structure information contained in the drug relationships, GraphRepur can predict new drugs for breast cancer, outperforming previous state-of-the-art approaches and some classic machine learning methods. The high-ranked drugs have indeed been reported as new uses for breast cancer treatment recently. AVAILABILITYAND IMPLEMENTATION: The source code of our model and datasets are available at: https://github.com/cckamy/GraphRepur and https://figshare.com/articles/software/GraphRepur_Breast_Cancer_Drug_Repurposing/14220050. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

TransformerCPI: improving compound-protein interaction prediction by sequence-based deep learning with self-attention mechanism and label reversal experiments.

MOTIVATION: Identifying compound-protein interaction (CPI) is a crucial task in drug discovery and chemogenomics studies, and proteins without three-dimensional structure account for a large part of potential biological targets, which requires developing methods using only protein sequence information to predict CPI. However, sequence-based CPI models may face some specific pitfalls, including using inappropriate datasets, hidden ligand bias and splitting datasets inappropriately, resulting in overestimation of their prediction performance. RESULTS: To address these issues, we here constructed new datasets specific for CPI prediction, proposed a novel transformer neural network named TransformerCPI, and introduced a more rigorous label reversal experiment to test whether a model learns true interaction features. TransformerCPI achieved much improved performance on the new experiments, and it can be deconvolved to highlight important interacting regions of protein sequences and compound atoms, which may contribute chemical biology studies with useful guidance for further ligand structural optimization. AVAILABILITY AND IMPLEMENTATION: https://github.com/lifanchen-simm/transformerCPI.

Distinctive Roles and Mechanisms of Human Neutrophil Peptides in Experimental Sepsis and Acute Respiratory Distress Syndrome.

OBJECTIVES: To examine the effects and mechanisms of human neutrophil peptides in systemic infection and noninfectious inflammatory lung injury. DESIGN: Prospective experimental study. SETTING: University hospital-based research laboratory. SUBJECTS: In vitro human cells and in vivo mouse models. INTERVENTIONS: Wild-type (Friend virus B-type) and conditional leukocyte human neutrophil peptides transgenic mice were subjected to either sepsis induced by cecal ligation and puncture or acute lung injury by intratracheal instillation of hydrochloric acid followed by mechanical ventilation. Using human neutrophil peptides as bait, the basal cell adhesion molecule (CD239) and the purinergic P2Y purinoceptor 6 receptor were identified as the putative human neutrophil peptides receptor complex in human lung epithelial cells. MEASUREMENTS AND MAIN RESULTS: In the cecal ligation and puncture sepsis model, Friend virus B-type mice exhibited higher systemic bacterial load, cytokine production, and lung injury than human neutrophil peptides transgenic mice. Conversely, an increased lung cytokine production was seen in Friend virus B-type mice, which was further enhanced in human neutrophil peptides transgenic mice in response to two-hit lung injury induced by hydrochloric acid and mechanical ventilation. The human neutrophil peptides-mediated inflammatory response was mediated through the basal cell adhesion molecule-P2Y purinoceptor 6 receptor signal pathway in human lung epithelial cells. CONCLUSIONS: Human neutrophil peptides are critical in host defense against infectious sepsis by their cationic antimicrobial properties but may exacerbate tissue injury when neutrophil-mediated inflammatory responses are excessive in noninfectious lung injury. Targeting the basal cell adhesion molecule/P2Y purinoceptor 6 signaling pathway may serve as a novel approach to attenuate the neutrophil-mediated inflammatory responses and injury while maintaining the antimicrobial function of human neutrophil peptides in critical illness.

Mechanical Stress and the Induction of Lung Fibrosis via the Midkine Signaling Pathway.

RATIONALE: Lung-protective ventilatory strategies have been widely used in patients with acute respiratory distress syndrome (ARDS), but the ARDS mortality rate remains unacceptably high and there is no proven pharmacologic therapy. OBJECTIVES: Mechanical ventilation can induce oxidative stress and lung fibrosis, which may contribute to high dependency on ventilator support and increased ARDS mortality. We hypothesized that the novel cytokine, midkine (MK), which can be up-regulated in oxidative stress, plays a key role in the pathogenesis of ARDS-associated lung fibrosis. METHODS: Blood samples were collected from 17 patients with ARDS and 10 healthy donors. Human lung epithelial cells were challenged with hydrogen chloride followed by mechanical stretch for 72 hours. Wild-type and MK gene-deficient (MK(-/-)) mice received two-hit injury of acid aspiration and mechanical ventilation, and were monitored for 14 days. MEASUREMENTS AND MAIN RESULTS: Plasma concentrations of MK were higher in patients with ARDS than in healthy volunteers. Exposure to mechanical stretch of lung epithelial cells led to an epithelial-mesenchymal transition profile associated with increased expression of angiotensin-converting enzyme, which was attenuated by silencing MK, its receptor Notch2, or NADP reduced oxidase 1. An increase in collagen deposition and hydroxyproline level and a decrease in lung tissue compliance seen in wild-type mice were largely attenuated in MK(-/-) mice. CONCLUSIONS: Mechanical stretch can induce an epithelial-mesenchymal transition phenotype mediated by the MK-Notch2-angiotensin-converting enzyme signaling pathway, contributing to lung remodeling. The MK pathway is a potential therapeutic target in the context of ARDS-associated lung fibrosis.

Application of ensemble learning for predicting GABAA receptor agonists.

BACKGROUND: Over the past few decades, agonists binding to the benzodiazepine site of the GABAA receptor have been successfully developed as clinical drugs. Different modulators (agonist, antagonist, and reverse agonist) bound to benzodiazepine sites exhibit different or even opposite pharmacological effects, however, their structures are so similar that it is difficult to distinguish them based solely on molecular skeleton. This study aims to develop classification models for predicting the agonists. METHODS: 306 agonists or non-agonists were collected from literature. Six machine learning algorithms including RF, XGBoost, AdaBoost, GBoost, SVM, and ANN algorithms were employed for model development. Using six descriptors including 1D/2D Descriptors, ECFP4, 2D-Pharmacophore, MACCS, PubChem, and Estate fingerprint to characterize chemical structures. The model interpretability was explored by SHAP method. RESULTS: The best model demonstrated an AUC value of 0.905 and an MCC value of 0.808 for the test set. The PubMac-based model (PubMac-GB) achieved best AUC values of 0.935 for test set. The SHAP analysis results emphasized that MaccsFP62, ECFP_624, ECFP_724, and PubchemFP213 were the crucial molecular features. Applicability domain analysis was also performed to determine reliable prediction boundaries for the model. The PubMac-GB model was applied to virtual screening for potential GABAA agonists and the top 100 compounds were given. CONCLUSION: Overall, our ensemble learning-based model (PubMac-GB) achieved comparable performance and would be helpful in effectively identifying agonists of GABAA receptors.

Reducing overconfident errors in molecular property classification using Posterior Network.

Deep-learning-based classification models are increasingly used for predicting molecular properties in drug development. However, traditional classification models using the Softmax function often give overconfident mispredictions for out-of-distribution samples, highlighting a critical lack of accurate uncertainty estimation. Such limitations can result in substantial costs and should be avoided during drug development. Inspired by advances in evidential deep learning and Posterior Network, we replaced the Softmax function with a normalizing flow to enhance the uncertainty estimation ability of the model in molecular property classification. The proposed strategy was evaluated across diverse scenarios, including simulated experiments based on a synthetic dataset, ADMET predictions, and ligand-based virtual screening. The results demonstrate that compared with the vanilla model, the proposed strategy effectively alleviates the problem of giving overconfident but incorrect predictions. Our findings support the promising application of evidential deep learning in drug development and offer a valuable framework for further research.

Highly accurate carbohydrate-binding site prediction with DeepGlycanSite.

As the most abundant organic substances in nature, carbohydrates are essential for life. Understanding how carbohydrates regulate proteins in the physiological and pathological processes presents opportunities to address crucial biological problems and develop new therapeutics. However, the diversity and complexity of carbohydrates pose a challenge in experimentally identifying the sites where carbohydrates bind to and act on proteins. Here, we introduce a deep learning model, DeepGlycanSite, capable of accurately predicting carbohydrate-binding sites on a given protein structure. Incorporating geometric and evolutionary features of proteins into a deep equivariant graph neural network with the transformer architecture, DeepGlycanSite remarkably outperforms previous state-of-the-art methods and effectively predicts binding sites for diverse carbohydrates. Integrating with a mutagenesis study, DeepGlycanSite reveals the guanosine-5'-diphosphate-sugar-recognition site of an important G-protein coupled receptor. These findings demonstrate DeepGlycanSite is invaluable for carbohydrate-binding site prediction and could provide insights into molecular mechanisms underlying carbohydrate-regulation of therapeutically important proteins.

Learning protein fitness landscapes with deep mutational scanning data from multiple sources.

One of the key points of machine learning-assisted directed evolution (MLDE) is the accurate learning of the fitness landscape, a conceptual mapping from sequence variants to the desired function. Here, we describe a multi-protein training scheme that leverages the existing deep mutational scanning data from diverse proteins to aid in understanding the fitness landscape of a new protein. Proof-of-concept trials are designed to validate this training scheme in three aspects: random and positional extrapolation for single-variant effects, zero-shot fitness predictions for new proteins, and extrapolation for higher-order variant effects from single-variant effects. Moreover, our study identified previously overlooked strong baselines, and their unexpectedly good performance brings our attention to the pitfalls of MLDE. Overall, these results may improve our understanding of the association between different protein fitness profiles and shed light on developing better machine learning-assisted approaches to the directed evolution of proteins. A record of this paper's transparent peer review process is included in the supplemental information.

LogD7.4 prediction enhanced by transferring knowledge from chromatographic retention time, microscopic pKa and logP.

Lipophilicity is a fundamental physical property that significantly affects various aspects of drug behavior, including solubility, permeability, metabolism, distribution, protein binding, and toxicity. Accurate prediction of lipophilicity, measured by the logD7.4 value (the distribution coefficient between n-octanol and buffer at physiological pH 7.4), is crucial for successful drug discovery and design. However, the limited availability of data for logD modeling poses a significant challenge to achieving satisfactory generalization capability. To address this challenge, we have developed a novel logD7.4 prediction model called RTlogD, which leverages knowledge from multiple sources. RTlogD combines pre-training on a chromatographic retention time (RT) dataset since the RT is influenced by lipophilicity. Additionally, microscopic pKa values are incorporated as atomic features, providing valuable insights into ionizable sites and ionization capacity. Furthermore, logP is integrated as an auxiliary task within a multitask learning framework. We conducted ablation studies and presented a detailed analysis, showcasing the effectiveness and interpretability of RT, pKa, and logP in the RTlogD model. Notably, our RTlogD model demonstrated superior performance compared to commonly used algorithms and prediction tools. These results underscore the potential of the RTlogD model to improve the accuracy and generalization of logD prediction in drug discovery and design. In summary, the RTlogD model addresses the challenge of limited data availability in logD modeling by leveraging knowledge from RT, microscopic pKa, and logP. Incorporating these factors enhances the predictive capabilities of our model, and it holds promise for real-world applications in drug discovery and design scenarios.

Tora3D: an autoregressive torsion angle prediction model for molecular 3D conformation generation.

Three-dimensional (3D) conformations of a small molecule profoundly affect its binding to the target of interest, the resulting biological effects, and its disposition in living organisms, but it is challenging to accurately characterize the conformational ensemble experimentally. Here, we proposed an autoregressive torsion angle prediction model Tora3D for molecular 3D conformer generation. Rather than directly predicting the conformations in an end-to-end way, Tora3D predicts a set of torsion angles of rotatable bonds by an interpretable autoregressive method and reconstructs the 3D conformations from them, which keeps structural validity during reconstruction. Another advancement of our method over other conformational generation methods is the ability to use energy to guide the conformation generation. In addition, we propose a new message-passing mechanism that applies the Transformer to the graph to solve the difficulty of remote message passing. Tora3D shows superior performance to prior computational models in the trade-off between accuracy and efficiency, and ensures conformational validity, accuracy, and diversity in an interpretable way. Overall, Tora3D can be used for the quick generation of diverse molecular conformations and 3D-based molecular representation, contributing to a wide range of downstream drug design tasks.

Computing the relative binding affinity of ligands based on a pairwise binding comparison network.

Structure-based lead optimization is an open challenge in drug discovery, which is still largely driven by hypotheses and depends on the experience of medicinal chemists. Here we propose a pairwise binding comparison network (PBCNet) based on a physics-informed graph attention mechanism, specifically tailored for ranking the relative binding affinity among congeneric ligands. Benchmarking on two held-out sets (provided by Schrödinger and Merck) containing over 460 ligands and 16 targets, PBCNet demonstrated substantial advantages in terms of both prediction accuracy and computational efficiency. Equipped with a fine-tuning operation, the performance of PBCNet reaches that of Schrödinger's FEP+, which is much more computationally intensive and requires substantial expert intervention. A further simulation-based experiment showed that active learning-optimized PBCNet may accelerate lead optimization campaigns by 473%. Finally, for the convenience of users, a web service for PBCNet is established to facilitate complex relative binding affinity prediction through an easy-to-operate graphical interface.

Uncertainty quantification: Can we trust artificial intelligence in drug discovery?

The problem of human trust is one of the most fundamental problems in applied artificial intelligence in drug discovery. In silico models have been widely used to accelerate the process of drug discovery in recent years. However, most of these models can only give reliable predictions within a limited chemical space that the training set covers (applicability domain). Predictions of samples falling outside the applicability domain are unreliable and sometimes dangerous for the drug-design decision-making process. Uncertainty quantification accordingly has drawn great attention to enable autonomous drug designing. By quantifying the confidence level of model predictions, the reliability of the predictions can be quantitatively represented to assist researchers in their molecular reasoning and experimental design. Here we summarize the state-of-the-art approaches to uncertainty quantification and underline how they can be used for drug design and discovery projects. Furthermore, we also outline four representative application scenarios of uncertainty quantification in drug discovery.

Graph neural network approaches for drug-target interactions.

Developing new drugs remains prohibitively expensive, time-consuming, and often involves safety issues. Accurate prediction of drug-target interactions (DTIs) can guide the drug discovery process and thus facilitate drug development. Non-Euclidian data such as drug-like molecule structures, key pocket residue structures, and protein interaction networks can be represented effectively using graphs. Therefore, the emerging graph neural network has been rapidly applied to predict DTIs, and proved effective in finding repositioning drugs and accelerating drug discovery. In this review, we provide a brief overview of deep neural networks used in DTI models. Then, we summarize the database required for DTI prediction, followed by a comprehensive introduction of applications of graph neural networks for DTI prediction. We also highlight current challenges and future directions to guide the further development of this field.

Facing small and biased data dilemma in drug discovery with enhanced federated learning approaches.

Artificial intelligence (AI) models usually require large amounts of high-quality training data, which is in striking contrast to the situation of small and biased data faced by current drug discovery pipelines. The concept of federated learning has been proposed to utilize distributed data from different sources without leaking sensitive information of the data. This emerging decentralized machine learning paradigm is expected to dramatically improve the success rate of AI-powered drug discovery. Here, we simulated the federated learning process with different property and activity datasets from different sources, among which overlapping molecules with high or low biases exist in the recorded values. Beyond the benefit of gaining more data, we also demonstrated that federated training has a regularization effect superior to centralized training on the pooled datasets with high biases. Moreover, different network architectures for clients and aggregation algorithms for coordinators have been compared on the performance of federated learning, where personalized federated learning shows promising results. Our work demonstrates the applicability of federated learning in predicting drug-related properties and highlights its promising role in addressing the small and biased data dilemma in drug discovery.

Cocrystal Prediction of Bexarotene by Graph Convolution Network and Bioavailability Improvement.

Bexarotene (BEX) was approved by the FDA in 1999 for the treatment of cutaneous T-cell lymphoma (CTCL). The poor aqueous solubility causes the low bioavailability of the drug and thereby limits the clinical application. In this study, we developed a GCN-based deep learning model (CocrystalGCN) for in-silico screening of the cocrystals of BEX. The results show that our model obtained high performance relative to baseline models. The top 30 of 109 coformer candidates were scored by CocrystalGCN and then validated experimentally. Finally, cocrystals of BEX-pyrazine, BEX-2,5-dimethylpyrazine, BEX-methyl isonicotinate, and BEX-ethyl isonicotinate were successfully obtained. The crystal structures were determined by single-crystal X-ray diffraction. Powder X-ray diffraction, differential scanning calorimetry, and thermogravimetric analysis were utilized to characterize these multi-component forms. All cocrystals present superior solubility and dissolution over the parent drug. The pharmacokinetic studies show that the plasma exposures (AUC0-8h) of BEX-pyrazine and BEX-2,5-dimethylpyrazine are 1.7 and 1.8 times that of the commercially available BEX powder, respectively. This work sets a good example for integrating virtual prediction and experimental screening to discover the new cocrystals of water-insoluble drugs.

Blood-brain barrier penetration prediction enhanced by uncertainty estimation.

Blood-brain barrier is a pivotal factor to be considered in the process of central nervous system (CNS) drug development, and it is of great significance to rapidly explore the blood-brain barrier permeability (BBBp) of compounds in silico in early drug discovery process. Here, we focus on whether and how uncertainty estimation methods improve in silico BBBp models. We briefly surveyed the current state of in silico BBBp prediction and uncertainty estimation methods of deep learning models, and curated an independent dataset to determine the reliability of the state-of-the-art algorithms. The results exhibit that, despite the comparable performance on BBBp prediction between graph neural networks-based deep learning models and conventional physicochemical-based machine learning models, the GROVER-BBBp model shows greatly improvement when using uncertainty estimations. In particular, the strategy combined Entropy and MC-dropout can increase the accuracy of distinguishing BBB + from BBB - to above 99% by extracting predictions with high confidence level (uncertainty score < 0.1). Case studies on preclinical/clinical drugs for Alzheimer' s disease and marketed antitumor drugs that verified by literature proved the application value of uncertainty estimation enhanced BBBp prediction model, that may facilitate the drug discovery in the field of CNS diseases and metastatic brain tumors.

Drug target inference by mining transcriptional data using a novel graph convolutional network framework.

A fundamental challenge that arises in biomedicine is the need to characterize compounds in a relevant cellular context in order to reveal potential on-target or off-target effects. Recently, the fast accumulation of gene transcriptional profiling data provides us an unprecedented opportunity to explore the protein targets of chemical compounds from the perspective of cell transcriptomics and RNA biology. Here, we propose a novel Siamese spectral-based graph convolutional network (SSGCN) model for inferring the protein targets of chemical compounds from gene transcriptional profiles. Although the gene signature of a compound perturbation only provides indirect clues of the interacting targets, and the biological networks under different experiment conditions further complicate the situation, the SSGCN model was successfully trained to learn from known compound-target pairs by uncovering the hidden correlations between compound perturbation profiles and gene knockdown profiles. On a benchmark set and a large time-split validation dataset, the model achieved higher target inference accuracy as compared to previous methods such as Connectivity Map. Further experimental validations of prediction results highlight the practical usefulness of SSGCN in either inferring the interacting targets of compound, or reversely, in finding novel inhibitors of a given target of interest.