Jianlin Jack Cheng

The paper introduces TransFew, a transformer model that integrates protein sequence and function label representations to improve protein function prediction, especially for rare function terms.

This study identifies molecular subgroups and epigenetic alterations in Beckwith-Wiedemann Syndrome and large offspring syndrome across methylome, transcriptome, and chromatin architecture.

The paper introduces MULTICOM, a system that enhances AlphaFold-Multimer for protein complex structure prediction, achieving higher TM-scores in CASP15.

This work introduces an equivariant graph neural network to predict tertiary structural changes from single-site mutations, outperforming AlphaFold in mutant structure prediction.

The paper improves AlphaFold2-based protein structure prediction using MULTICOM_refine, achieving higher accuracy in CASP15 compared to standard AlphaFold2.

The paper introduces an E(n) Equivariant Graph Neural Network to predict ligand-protein binding structures, showing promising results for drug design.

This paper introduces DProQA, a gated graph transformer for assessing the quality of predicted protein complex structures, achieving third place in CASP15 quality assessment.

This paper introduces TransFun, a method that combines protein sequences and structures using transformers and equivariant graph neural networks to improve protein function prediction.

The paper introduces a hybrid method combining pairwise similarity and deep learning for estimating protein complex model accuracy, achieving high correlation and low ranking loss in CASP15.

This paper introduces DeepProLigand, a deep learning pipeline that improves protein-ligand interaction modeling using cryo-EM data and templates, achieving top performance in a benchmark challenge.

The paper introduces EnQA, a 3D-equivariant graph neural network for assessing protein model quality, achieving state-of-the-art performance and outperforming AlphaFold2's own scores.

CDPred predicts inter-chain residue-residue distances in protein complexes using 2D attention-based deep learning, outperforming existing methods on homodimer and heterodimer datasets.

This paper introduces DistDom, a deep learning method that predicts protein domain boundaries using 1D sequence features and 2D distance maps, achieving higher accuracy than existing methods on CASP14 data.

ATOMRefine improves protein structure prediction by refining all-atom coordinates using a SE(3)-equivariant graph neural network, enhancing both backbone and side-chain accuracy.

This paper introduces DRLComplex, a deep reinforcement learning method for reconstructing protein dimer quaternary structures from interchain contacts, achieving high accuracy on standard datasets.

DISTEMA uses a 2D CNN with attention to predict protein model quality from raw distance maps, achieving lower ranking loss than existing methods.

This paper introduces DRCon, a deep learning model for predicting interchain contacts in homodimers, achieving higher accuracy than existing methods on benchmark datasets.

This paper introduces a gradient descent algorithm to reconstruct protein quaternary structures from inter-chain contacts, showing it outperforms other methods and can produce high-quality models even with predicted contacts.

This paper introduces 4DMax, a method that predicts dynamic chromosome structures from time-series Hi-C data, effectively interpolating chromatin positions and recovering structural features at unobserved time points.

The paper improves protein tertiary structure prediction by enhancing deep learning and distance prediction methods in the MULTICOM system, achieving better performance in CASP14.

The paper uses deep learning on monomeric multiple sequence alignments to predict interchain contacts in homomultimeric proteins, achieving high precision in some cases.

The paper develops deep learning-based EMA predictors for protein model accuracy estimation in CASP14, achieving top rankings in GDT-TS score loss.

TAPIOCA predicts chromatin topology using epigenetic features without Hi-C data, outperforming existing methods across cell lines.

The paper develops deep learning-based protein distance predictors for CASP14, achieving top rankings in contact prediction and showing the importance of MSAs and features in accuracy.

VEHiCLE enhances Hi-C contact map resolution using a variational autoencoder and adversarial training, improving downstream analysis capabilities.

The paper evaluates 43 methods for predicting intrinsically disordered regions and binding regions in proteins, finding that deep learning methods outperform older physicochemical approaches but binding prediction remains challenging.

This study introduces an attention-based convolutional neural network for protein contact prediction, showing improved interpretability and predictive performance on CASP13 targets.

This paper introduces DeepGRN, an attention-based model for transcription factor binding site prediction, showing improved performance on ENCODE-DREAM datasets.

DeepDist predicts real-value inter-residue distances using a deep residual network, improving contact prediction and tertiary structure reconstruction compared to multi-class methods.