Computational Drug Discovery Methods

This study uses a 3D convolutional neural network to predict GPCR binding sites with high accuracy.

This paper predicts antibody clearance using deep learning and physicochemical features, showing comparable performance to physicochemical models.

The paper identifies miR-135b as a differentially expressed miRNA in colorectal cancer and models small molecule ligands targeting its Dicer cleavage site using computational methods.

The paper improves drug target prediction by combining gene and compound morphological profiles using a transformer model, showing better performance than direct matching methods.

This paper uses conditional neural processes to predict adverse drug reactions from sparse data and finds they are well-calibrated and effective across organ systems.

This paper introduces a graph-based model to identify druggable vulnerabilities in solid cancers, finding 43 novel targets by leveraging cooperativity in molecular networks.

DrugGym simulates drug discovery processes to evaluate how different strategies affect success rates and costs in achieving target candidate profiles.

This paper introduces the AI-Enhanced RAIN protocol to optimize drug combinations for rectal neoplasms using GraphSAGE, systematic reviews, and meta-analysis, finding effective combinations with significant p-values.

This paper develops an AI framework to generate and optimize BACE1 inhibitors for Alzheimer's disease, identifying 14 candidates with improved binding properties.

This paper introduces GExMix, a data augmentation method for gene expression data, which improves drug response prediction by enhancing representation learning and prioritizing gene-centric information.

This study uses network pharmacology to identify the molecular mechanism of Huanglian Jiedu Decoction in treating Alzheimer's disease, revealing its anti-inflammatory effects through multiple targets and pathways.

This paper improves drug sensitivity prediction using multitask learning, showing better accuracy for sensitive cell lines and identifying tissue-specific predictors.

The paper develops ML-QSAR models and knowledge-driven neural networks to predict bioactivity of chalcone derivatives against disease targets and identify key molecular mechanisms.

This paper develops DIF-based MD-ANN models to predict HIV-1 RTI resistance using genotypic data, showing improved performance over IF-based models in single-drug benchmarks.

This study compares machine learning models for predicting drug action on nuclear type I receptors, finding decision tree and naive Bayes models to be most efficient and accurate.

This study identifies fragments binding to EV-D68 3C protease through crystallography and expands them into lead-like compounds, revealing novel binding modes and structural insights for drug design.

This paper develops a framework for predicting single-cell drug responses using pharmacogenomic embeddings and few-shot transfer learning, showing improved accuracy and generalization.

This study identifies a novel allosteric pocket on APOBEC3B using computational methods but finds that metal contamination in screening samples led to false positive inhibitor results.

This study uses a graph convolutional network to predict effective herbal prescriptions for metabolic diseases by integrating protein-protein interaction networks and network pharmacology data.

The paper assesses current chemical coverage of the human Reactome, finding that 3% of proteins are targeted by chemical probes, but they cover 53% of pathways due to pathway overlap.

The paper introduces an algorithm for optimizing protein-ligand complex structures, improving data curation and analysis for drug discovery.

This study introduces ProteinReDiff, a diffusion-based model for redesigning ligand-binding proteins using sequence and ligand SMILES inputs, showing effectiveness in affinity and structural preservation.

The paper proposes a method using graph neural networks to recommend drug combinations for lung adenocarcinoma by targeting genes and proteins, finding Gefitinib, Paclitaxel, and Icotinib as the most effective combination.

The paper introduces a 3D molecular auto-encoder with a new training objective, achieving better performance than existing methods.

PertKGE improves compound-protein interaction prediction using knowledge graph embedding of perturbation transcriptomics, outperforming baselines in cold-start settings and identifying novel therapeutic targets.

This paper introduces a multi-state modeling protocol for kinases using AlphaFold2 to improve structure-based virtual screening by overcoming structural biases and enhancing the identification of diverse inhibitors.

iScore is a machine learning-based scoring function for predicting protein-ligand binding affinity, outperforming existing methods in accuracy and efficiency for drug discovery.

The paper introduces a model called CSNN that uses network homophily to improve in-distribution predictions of drug-target interactions for human GPCRs.

This paper demonstrates that conventional ML can achieve high accuracy in virtual screening with proper data and representation, outperforming deep learning methods.

This paper constructs a comprehensive biomedical knowledge graph, MegaKG, integrating 23 data sources to support drug development by generating explainable hypotheses and facilitating practical applications.