Computational Drug Discovery Methods

THEOBROMA: an aggregated open database of 1.13 million natural products with per-compound license auditing, three-tier classification, and stereochemistry-aware deduplication

THEOBROMA aggregates 1.13 million natural products with per-compound license auditing, three-tier classification, and stereochemistry-aware deduplication to support license-compatible virtual screening and bioactivity analysis.

Cancer Cell Line Heterogeneity Imposes a Primary Bottleneck for Virtual Perturbation Screening at Scale

The paper extends a perturbation prediction model to 162 cancer cell lines and finds cancer cell line heterogeneity significantly reduces prediction accuracy compared to normal and primary cell lines.

LEN-Seek: Fast and scalable ligand binding-site similarity search in the latent space of an SE(3)-invariant graph VAE

LEN-Seek is a method for ligand binding-site similarity search using a graph VAE that encodes structural information into a latent space, achieving faster and more scalable results than existing methods.

Unlocking Scalable Ligand Residence Time Predictions with Koffee Unbinding Kinetics Simulations

This paper introduces Koffee Unbinding Kinetics, a method for fast, scalable prediction of ligand-protein residence times to improve drug discovery.

PepXPro: a framework for curating, generating, and optimizing structure-affinity protein-peptide datasets

PepXPro curates and generates standardized protein-peptide datasets for benchmarking computational methods, providing a reproducible framework for dataset creation and evaluation.

DQHTFI: Dynamic-Query Hypergraph Transformer for Fine-Grained Drug-Target Interaction and Affinity Prediction

This paper introduces DQHTFI, a framework for drug-target interaction classification and binding affinity regression, achieving competitive results on benchmark datasets.

Human-supervised Agentic AI for Hypothesis Generation and Experimental Assistance in Drug Repurposing

A human-supervised multi-agent LLM system covered drug repurposing from knowledge graph building to assay analysis in three case studies; retrospective COVID-19 screen AUC-ROC 0.986 without predefined thresholds.

Learning from human and chemical languages to predict biological function

This paper introduces a deep learning model that predicts biological function from chemical structure using scientific literature, identifying effective inhibitors of beta-lactamases.

MERIT: Mechanism driven model predicts drug outcomes and nominates indications for failed drugs

MERIT predicts drug trial outcomes using molecular and disease features, outperforming existing methods and identifying potential indications for failed drugs.

In Silico Targeting of Calpain-5 in Autosomal Dominant Neovascular Inflammatory Vitreoretinopathy (ADNIV) with Peptides and Natural Products

This paper identifies druggable pockets of CAPN5 and screens natural products and peptides to find inhibitors for ADNIV.

Structural Context Determines Docking Engine Performance: A Family-Stratified Benchmark of Six Engines

This paper benchmarks six molecular docking and co-folding engines across 14 protein families, finding that engine performance depends on structural and physical characteristics of the target.

Joint Modeling of Transcriptomic and Morphological Phenotypes for Generative Molecular Design

Pert2Mol generates molecular structures from paired transcriptomic and morphological data using a multi-modal framework, achieving better performance than single-modality or diffusion-based methods.

AI-Driven Computational Design of Peptide-Based WWP1 Inhibitors as Promising Therapeutic Agents Against Breast Cancer, Including Triple-Negative Subtype

This study uses AI to design a peptide inhibitor of WWP1, showing potent activity against breast cancer, especially TNBC, with promising cytotoxic effects and synergistic potential with PI3K inhibitors.

PandaMap: A Python Package for Comprehensive Visualization of Protein-Ligand Interaction Networks

PandaMap generates 2D/3D protein-ligand interaction diagrams, reports, and CSV files from PDB files, using validated distance thresholds and open-source tools.

ASOCompass: Context- and Chemistry-Aware Activity Prediction for Transferable Antisense Oligonucleotide Screening

ASOCompass predicts ASO activity by integrating sequence, chemistry, and biological context, achieving higher correlation than baselines across diverse conditions.

Collaborative multi-agent intelligence uncovers subtype-selective allosteric sites at GPCR-lipid interaction interface

The paper identifies subtype-selective allosteric sites at GPCR-lipid interfaces using a multi-agent workflow and screens compounds to find ligand candidates.

ClinOracle: Hierarchical AI Prediction of Target Binding and Patient-Derived Functional Activity Across Diverse Therapeutic Targets

ClinOracle uses a hierarchical graph neural network to predict target binding and patient-derived functional activity, improving compound prioritization through a Priority Score combining translational probability and developability.

A Cluster-Specific First-principles Network Pharmacology Framework for Molecular-Level Mechanism Deduction: Application to the HL-60-Selective Cytotoxicity of 3-Deoxycardiobutanolide

This paper introduces a network pharmacology framework to generate molecular-level hypotheses for drug mechanisms, applying it to explain the selective cytotoxicity of 3-deoxycardiobutanolide in HL-60 cells.

CellAwareGNN: Single-Cell Enhanced Knowledge Graph Foundation Model for Drug Indication Prediction

Adding OneK1K immune cell-type expression edges to an updated PrimeKG modestly improved drug-indication link prediction, AUPRC 0.816 to 0.826 overall and 0.847 to 0.864 for 40 autoimmune diseases.

Computational and Structure-Guided E-Pharmacophore-Based Virtual Screening for the Identification of Novel NEK2 Kinase Inhibitors as Potential Anticancer Agents

This study identifies three novel NEK2 kinase inhibitors through computational pharmacophore modeling and virtual screening, showing promising binding stability and pharmacokinetic properties.

BENDER DB: a database of protein binding sites across neglected disease proteomes

BENDER DB compiles predicted protein binding sites from neglected disease pathogens using AlphaFold structures and multiple predictors, offering a consensus and visualization tools to aid drug discovery.

Boltz-Perturb: Improving Diversity and Accuracy in Protein-Ligand Co-Folding through Training-Free Conditioning Perturbation

Boltz-Perturb improves ligand binding pose prediction by introducing inference-time perturbations, achieving higher success rates with less computation.

MolX: A Geometric Foundation Model for Protein-Ligand Modelling

MolX is a geometric foundation model that jointly learns representations of protein pockets and ligands from 3D structural data, achieving state-of-the-art performance in predicting molecular interactions.

DrugLM: A Unified Framework to Enhance Drug-Target Interaction Predictions by Incorporating Textual Embeddings via Language Models

DrugLM enhances drug-target interaction predictions by integrating textual embeddings from large language models into DTI models, showing strong performance without fine-tuning.

Highly potent antiviral drug candidates targeting SARS-CoV-2 nsp3 and nsp5

The paper identifies two small molecules, Amb929 and Amb701, as potent inhibitors of SARS-CoV-2 nsp3 and nsp5, with Amb929 showing higher antiviral activity and lower cytotoxicity.

Overcoming the accuracy-generalization tradeoff in docking and scoring for prospective virtual screening

This paper introduces DODock and DOScore, hybrid ML/physics frameworks for docking and scoring that improve accuracy while generalizing better to novel molecules, showing significant improvements in hit rates for virtual screening.

Nesso-1: Accelerating Open-Source Binding Affinity Predictions

Nesso-1 is an open-source binding affinity prediction tool that is significantly faster than Boltz-2 while maintaining or improving accuracy on both in-distribution and out-of-distribution data.

Structure-based discovery of inhibitors of Mac1 domain of nonstructural protein-3 of SARS-CoV-2 by machine learning-augmented screening of chemical space

The paper uses machine learning-augmented virtual screening to discover a novel series of Mac1 inhibitors for SARS-CoV-2 NSP3, validated by experimental assays and crystallography.

PG-MLD: Physics-Guided Molecular Representation Learning via Dynamic 3D Trajectory Distillation

PG-MLD transfers dynamic 3D physical knowledge to SMILES-based molecular models, improving property prediction without requiring 3D conformer generation.

MOZAIC: Compound Growth via In Silico Reactions and Global Optimization using Conformational Space Annealing

MOZAIC generates chemically diverse drug-like molecules by combining fragment-based growth with global optimization, improving lead-like properties such as QED and synthetic accessibility.