Hongliang Duan

This paper introduces RA-GHN, a framework for antibody sequence-structure co-design that captures multi-body interactions and structural constraints, outperforming existing methods.

HighFold-Lasso predicts lasso peptide structures using topology-guided methods, achieving perfect topology success and low RMSD on a test set of 52 structures.

MINT predicts enzyme-cofactor specificity using a multimodal framework with transfer learning, achieving high accuracy in cofactor binding prediction.

CAPTOR-1, a small molecule inhibitor of NOS1AP-tau interaction, reduces tau pathology and improves cognition in AD models.

This study identifies cyclic peptide ligands targeting CAPON, showing reduced neuronal toxicity and tau phosphorylation under Alzheimer's disease stress conditions.

This paper designs AI-generated cyclic peptides that modulate the CD28 immune checkpoint with controllable pharmacology and therapeutic efficacy in preclinical models.

High-PepBinder generates target-specific peptides using a pLM-guided diffusion model and a comprehensive dataset, producing peptides with favorable binding properties and sequence diversity.

The paper designs cyclic peptides targeting TREM2 using AI tools and validates their binding and stability.

This study evaluates AlphaFold3's accuracy in predicting non-canonical cyclic peptides and provides guidelines for its application.

HighFold3 improves cyclic peptide structure prediction, especially for those with unnatural amino acids, outperforming existing models.

This study introduces cycledesigner, a method combining RFdiffusion, ProteinMPNN, and HighFold to design cyclic peptide binders for specific targets, followed by scoring functions for screening.

The paper introduces PPIKB, a database and platform for protein-peptide interactions, compiled from literature and patents with manual validation, supporting drug discovery and systems biology research.

HighPlay uses reinforcement learning and protein structure prediction to design cyclic peptides with good binding affinity, validated through molecular dynamics simulation and experimental testing.

NCPepFold predicts structures of non-canonical cyclic peptides with improved accuracy using multigranular representation and cyclization optimization.

HighFold2 predicts 3D structures of cyclic peptides with unnatural amino acids, achieving a median RMSD of 1.891 Å for monomers and complexes.

MeDCycFold accelerates cyclic peptide structure prediction using a deep learning-based Rosetta distillation model, achieving 49x speedup while maintaining Rosetta-level accuracy.

The paper develops a deep learning model called PepMSND to predict peptide blood stability using multi-level feature engineering and a comprehensive database of 635 samples, achieving high accuracy and AUC.

GAPS uses geometric attention mechanisms and transfer learning to identify protein-peptide binding sites, achieving state-of-the-art performance.

Highfold predicts cyclic peptides and their complexes using AlphaFold, improving accuracy by incorporating structural information.

This paper introduces Multi_CycGT, a deep learning model for predicting cyclic peptide membrane permeability, achieving high accuracy and AUC on a benchmark dataset.

The paper introduces a deep learning model called CcBHLA for predicting HLA class I peptide binding, showing improved performance over existing methods.