Computational Drug Discovery Methods

This paper uses the CANDO platform to identify repurposable drugs for schizophrenia, finding 25 high-corroboration candidates with validated targets and biological relevance.

BitBIRCH-Lean efficiently clusters large molecular libraries on workstations, achieving 2X speedup over the original BitBIRCH with no loss in cluster quality.

The paper identifies pyruvate carboxylase as a potential target of L-AAA, explaining its astrocyte-selective toxicity through metabolic disruption.

The paper evaluates hydropathic scoring functions to identify near-native ligand-protein poses using a benchmark dataset of 1000 complexes, achieving 90% predictive accuracy when accounting for conformational stress.

The paper introduces two conditioning strategies for flow-matching models to generate bioisosteric 3D molecules without retraining, demonstrating effectiveness on drug-relevant tasks.

The paper improves binding free energy calculations by enabling accurate water reorganization through enhanced sampling methods.

This paper designs and tests covalent kinase activity probes using computational methods and shows they target multiple kinases in live cells with distinct kinetics and conformations.

Slogen unifies fragment generation and screening for lead optimization, improving chemical diversity and synthetic tractability while matching state-of-the-art performance.

This paper introduces a multitask graph neural network for chemical reaction modeling that jointly predicts atom mappings, reaction centers, and classifications, achieving state-of-the-art results on a benchmark.

This paper identifies effective repurposed antivirals against MPXV by combining molecular evolutionary theory with virtual screening, showing nilotinib has 69% efficacy compared to tecovirimat.

This study identifies ursolic acid's therapeutic mechanism against MRSA-induced osteomyelitis through network pharmacology, bioinformatics, and experimental validation, showing immunomodulation, anti-inflammatory effects, and direct antibacterial activity.

The paper analyzes the structural and dynamic differences between ChikV Mac1 and SARS-CoV-2 Mac1 to explain differences in ligand binding affinity and designs inhibitors for ChikV Mac1.

This paper presents a scalable end-to-end pipeline for blind virtual screening using DiffDock and UniDock Vina, achieving high screening performance with minimal pose generation.

This paper introduces EMMA, a transformer-based multi-task learning framework for predicting enzyme-small molecule interactions, showing strong performance across different data-splitting regimes.

This paper improves protein-ligand binding affinity prediction by using linear regression and artificial neural networks with optimized distance-based descriptors, achieving higher accuracy than traditional methods.

PickPocket identifies ligand-binding residues across proteomes using evolutionary and geometric data, finding unannotated sites and improving drug discovery.

ComplexMatrixComb predicts drug combination IC50 doses using complex numbers and matrix factorization, outperforming traditional methods and validating with experimental assays.

This paper introduces ElixirSeeker2, a computational framework for de novo design of anti-aging peptides, which identifies novel candidates effective in delaying cellular senescence and improving locomotor activity in aged C. elegans.

The paper introduces FUSED, a method that integrates foundation models across domains to improve cancer drug response prediction by reducing input features and enhancing performance.

The paper introduces ontology pre-training to improve machine learning predictions for metabolites by integrating bio-ontology knowledge, showing better performance and faster training.

This paper combines DNA-encoded libraries with generative AI to expand hit compounds, finding AI-selected hits show better drug-likeness and diversity than initial DEL hits.

GatorAffinity improves protein-ligand binding affinity prediction by combining synthetic structural data with experimental data, achieving better accuracy and generalizability.

The paper establishes a statistical framework to identify synergistic anticancer drug interactions using large-scale cell line data and reference null distributions.

This study improves virtual screening of AChE inhibitors by combining multi-structure covalent docking with a machine learning consensus model, achieving higher predictive accuracy than single-structure approaches.

This paper benchmarks sequence-based models for predicting protein-ligand binding sites across different ligand availability regimes, showing improved performance with ligand conditioning and chemical language model integration.

This study identifies KKL-35 as a novel CNDP2 inhibitor through in silico screening, showing it has favorable drug-like properties despite lower potency than bestatin.

This paper develops a machine learning model to predict TLR4 binding affinity using molecular descriptors, identifying key features that guide the design of TLR4-targeted therapeutics.

This paper improves molecular generation by steering chemical language models to control properties like solubility and lipophilicity using activation steering techniques.

The paper introduces a soft Metropolis-Hastings correction for molecular diffusion models to improve chemical validity and structural stability in molecule generation.

SMARTBind uses deep learning and contrastive learning to identify small molecule binders to RNA without structural input, outperforming existing methods and reducing computational cost.