Liangzhen Zheng

This paper introduces Pinal, a foundation model that designs proteins from natural language descriptions, achieving functional alignment and experimental success in four design tasks.

This paper introduces zPolaron, a graph neural network for functional protein design that integrates structural, interaction, and evolutionary information, improving fitness prediction beyond structural compatibility.

OpenKinetics Predictor unifies thirteen enzyme kinetic predictors behind one interface and API; on E. coli iML1515, eight kcat predictors correlate pairwise at r 0.08 to 0.67.

The paper identifies a novel biosynthetic pathway for baccatin III production using engineered microbes, achieving its complete de novo synthesis through enzymatic validation and pathway engineering.

This study introduces AF-pred, an Alphafold-based tool for cross-species prediction of MHC-II restricted epitopes, showing better performance for non-human MHC-II than sequence-based methods.

TransCodon is a Transformer-based model that predicts optimal synonymous codons by integrating species-specific information, 5'UTRs, and RNA structure, improving translational efficiency in heterologous hosts.

This paper identifies the enzymatic mechanism for forming the gamma-thiolactone ring in thiolactomycin biosynthesis, revealing the roles of TlnCCy and TlnA.

This study benchmarks AlphaFold3-like methods for predicting protein-peptide complex structures, showing improved accuracy but challenges in binding site prediction.

This study benchmarks reverse docking pipelines using AlphaFold2 human proteome to predict drug-target interactions and off-target effects, finding that Glide_sfct (PS) performs best.

The paper curates a database of disulfide bonds and their effects on protein thermostability, develops machine learning models to predict beneficial disulfide bonds, and demonstrates the utility of AlphaFold2 for this task.

This paper introduces IGModel, a deep learning framework that predicts protein-ligand docking RMSD and binding strength, achieving state-of-the-art performance on multiple benchmarks.

The paper introduces Z-align, a molecular alignment method that improves binding pose prediction by using topological structural information and flexible optimization of bond lengths and angles.

The paper introduces a generalist model for drug design that predicts multiple properties simultaneously, achieving faster virtual screening and drug development with explainability.

fastMSA accelerates multiple sequence alignment by using a dual encoder based on a protein language model, achieving a 34-fold speed-up while maintaining homolog recall and compatibility with downstream tasks like AlphaFold.