Computational Drug Discovery Methods

The paper identifies FDA-approved drugs with potential binding activity against CDK2 using computational methods and selects Carteolol as a candidate.

GRAFT predicts drug responses across cell types without direct measurements by transferring genetic perturbation data to chemical perturbations.

The paper improves molecular docking by using targeted side-chain Monte Carlo sampling of an apo structure, achieving better ligand pose recovery than apo or restrained MD approaches.

This paper introduces an AI system for automated small molecule structure elucidation from NMR spectra, combining generative models with reasoning agents to outperform existing tools.

PSICHIC+ is a bioactivity foundation model that predicts binding affinities from sequence data, improving hit identification and hit-to-lead optimization using heterogeneous bioactivity data.

This study compares docking and short MD rescoring for ranking p38α inhibitors and finds no significant improvement from MD rescoring over docking.

This study evaluates toxicity prediction tools on approved drugs, finding no method outperforms simple fingerprints, and highlighting the need for inputs beyond molecular structure.

This paper benchmarks active learning strategies for ADMET prediction, finding benefits vary by endpoint and strategy, with Deep Ensembles showing better predictive performance than MC Dropout.

The paper benchmarks GNNs, pretrained models, and fingerprints for ADMET prediction under fair evaluation, finding no single model dominates and highlighting the importance of endpoint-specific selection and leakage-controlled testing.

The paper introduces DIMER, a framework for drug repurposing and virtual screening in Parkinson's disease, identifying 423 candidates and validating four in organoids.

This paper uses a graph neural network to predict antibacterial activity against Acinetobacter baumannii and identifies piperazine as a key contributor to predicted activity, leading to the testing of CE3F4.

The paper shows that transcriptomic similarity can rank candidate sources for response prediction better than random, but does not resolve aggregation or transfer uncertainty.

The paper introduces NK-RCPP, a workflow for prioritizing neurokinin-receptor peptide candidates using receptor-matched scores and structural-neighbour comparisons.

This study identifies potential neurodegeneration targets of sulforaphane using computational methods across multiple datasets.

The paper develops a computational-experimental framework to identify mutant-selective EGFR inhibitors by leveraging paired activity measurements and structural analysis.

This study identifies PabB as a potential drug target for tuberculosis by pathway analysis and screens FDA-approved drugs for repurposing using molecular docking.

This paper describes the development of CAMOLE-1, a machine learning platform for designing potent and selective MMP12 peptide inhibitors, achieving results in about one year with 270 synthesized peptides.

The paper designs two novel PDE4 inhibitors for plaque psoriasis with improved ADMET properties compared to Roflumilast using computational methods.

Agent3Diso identifies disease-relevant molecular forms by integrating molecular data with agentic reasoning to guide therapeutic discovery beyond gene-level targets.

This study benchmarks AlphaFold 3's interaction scores against BANDOCK+ for drug discovery, showing AF3's slightly better performance in disease ranking but BANDOCK+'s broader coverage.

The paper develops a platform integrating multisource target evidence with machine learning for Alzheimer's drug discovery, improving potency prediction and model performance.

This study investigates the adaptive binding of seven molecules to the Synaptotagmin 1 C2AB domain using molecular docking and simulations, revealing that desolvation and structural flexibility, not just binding energy, govern ligand interactions.

The paper uses network analysis of protein binding sites to reveal organizational patterns across scales, showing both diversification within protein categories and local similarities across boundaries.

MIEN improves glycan pose ranking by distinguishing experimentally observed conformations from non-native ones in protein-carbohydrate interactions.

This study uses supervised machine learning to predict lifespan-extending compounds in C. elegans by comparing different feature representations and models.

The paper uses scaffold-seeded enumeration to explore chemical space around a first-in-class 2AAR agonist, identifying active compounds with analgesic effects.

LigRNA predicts RNA-ligand binding sites using a graph neural network that combines language model embeddings and physicochemical features, achieving improved performance over existing methods.

PATHNet improves protein-ligand interaction prediction by combining physics-inspired modeling with neural networks, achieving high accuracy across multiple targets.

This paper benchmarks large language models for drug-target binding affinity prediction, showing that assay descriptions provide predictive value beyond molecular and protein data.

Large-language-model-assisted repurposing web tool recovered clemastine for multiple sclerosis remyelination only through mechanism-worded, model-dependent queries; its rankings missed later evidence of harm in progressive disease.