Computational Drug Discovery Methods

KODA uses LLMs and knowledge graphs to identify antimicrobial drug targets in the gut microbiome by analyzing metabolic pathways and microbial interactions.

This study evaluates AlphaFold3-derived models for drug binding affinity ranking, showing they outperform existing methods and require fewer resources.

This paper evaluates drug response prediction models and finds that deep learning models perform similarly to simple baselines, highlighting issues in current model evaluation practices.

This paper evaluates how design choices in chemical language models affect their performance and interpretability, finding that molecular representation and tokenization strategies significantly influence internal representations.

DELTA provides a balanced ligand-target database for AI drug discovery, showing active ligands are larger than inactive ones and similar targets bind similar ligands.

The paper introduces iCliff, a linear complexity method to quantify activity cliffs, showing strong correlation with SALI-based metrics and identifying key molecules contributing to activity cliffs.

CrypToth identifies cryptic sites on proteins using mixed-solvent molecular dynamics and topological data analysis, achieving high accuracy in hotspot prediction.

The paper introduces a protocol to integrate diverse bioactivity data into the Chemical Checker compound universe, enabling the generation of novel bioactivity signatures for small molecules.

This study designs an efficient siRNA for PKMYT1 knockdown and identifies natural ligands with multi-target binding affinities to overcome drug resistance in breast cancer.

The paper develops ChemBERTaPolyPharm to model polypharmacy side effects using ChemBERTa and PubMed encoders, achieving an F1 score of 0.93.

The paper identifies potential inhibitors of PBP2a in MRSA using pharmacophore modeling, 2D-QSAR, docking, and MD simulations, highlighting two compounds with favorable ADME properties and binding potential.

SyntheMol-RL generates novel and synthesizable antibiotics by optimizing for antibacterial activity and solubility, outperforming prior models and showing in vitro and in vivo efficacy.

Riemann-GNN uses hyperbolic geometry and causal inference to predict drug-disease associations with high accuracy and interpretability.

This paper introduces BAITSAO, a unified model for drug synergy prediction using large language models, showing improved performance over existing methods.

This paper evaluates combining graph neural networks with expert descriptors for ligand-based virtual screening, finding that simpler GNNs with descriptors perform comparably to complex ones and that expert descriptors excel in scaffold-split scenarios.

This paper develops an AI-driven drug discovery pipeline for LDHA inhibitors, identifying 100+ novel compounds with high inhibition potential and drug-like properties.

The paper screens over one million compounds to identify GIRK2-PIP2 modulators using docking, MM-GBSA, ADMET, and molecular dynamics.

This study uses in silico methods to analyze Forskolin's anticancer mechanisms, revealing its targeting of the PI3K/Akt pathway and related oncogenic networks.

MedSAGE improves small molecule drug design by generating synthesizable, drug-like molecules with high predicted affinity and selectivity for therapeutic targets.

This paper introduces ML models that predict relative growth inhibition for drug combinations, enabling personalized treatment recommendations beyond traditional synergy scores.

The paper identifies eight small molecules that bind to NPSL2, a regulatory element in oncomiR-1, using virtual screening and experimental validation.

The paper introduces ConfSeq, a conformation description language for 3D molecular structures, enabling sequence-based modeling and achieving state-of-the-art results in 3D molecular tasks.

This paper introduces HOLIgraph, a graph neural network model that predicts OATP1B1 inhibitors by integrating ligand features and protein-ligand interactions, achieving over 90% balanced accuracy.

This paper uses AI to predict how post-translational modifications affect drug binding by modeling PTM-modified protein structures and ligand interactions.

EVA predicts drug responses in immunology and inflammation diseases using a pre-trained foundation model, improving early trial outcome prediction and guiding drug development strategies.

This paper predicts chemical compound taste profiles using machine learning on a labeled dataset, finding that binary relevance models with Morgan fingerprints outperform other methods.

This paper introduces Rag2Mol methods for structure-based drug design using retrieval-augmented generation, producing drug candidates with better binding affinities and drug-likeness than existing models.

This paper introduces RT-DMF, a deep learning framework to correct noise and impute missing values in pharmacogenomic datasets, improving downstream analysis.

SynVerse evaluates DL models for drug synergy prediction, finding none outperform a naive baseline and highlighting poor generalization to unseen drugs and cell lines.

This paper introduces GenVS-TBDB, an AI-generated and virtual-screened small-molecule library targeting Mycobacterium tuberculosis, identifying bioactive compounds with potential for tuberculosis drug discovery.