Jiangning Song

CellReg3D integrates DNA language models and 3D genome structure to model cell-type-specific gene regulation, achieving state-of-the-art performance in predicting epigenetic marks and gene expression.

MolX is a geometric foundation model that jointly learns representations of protein pockets and ligands from 3D structural data, achieving state-of-the-art performance in predicting molecular interactions.

NicheTrans translates single-omics data into spatial multi-omics insights by integrating spatial and multi-modal information, outperforming existing methods in disease and tissue analysis.

This paper introduces TCRDiff, a conditional diffusion model for generating antigen-specific TCR sequences, validated through in vitro assays.

ExplainBind predicts protein-ligand binding likelihood and identifies key residues and non-covalent interactions without structural inputs, outperforming existing methods and enabling drug discovery insights.

EvoRMD predicts RNA modification types using a model that integrates sequence data and biological context, achieving strong performance and interpretability.

This study introduces Termini, an AI framework for designing antimicrobial peptides with terminal modifications, achieving high hit rates and therapeutic potential in vitro and in vivo.

OmniCleave models protease-substrate specificity using a geometric graph framework, outperforming existing methods and identifying novel substrates.

ACENet predicts enzyme pHmin using a graph neural network integrating structural and evolutionary features, achieving high accuracy and stability across conditions.

This paper introduces GDOT-Net, a graph diffusion optimal transport network that aligns structural and functional brain connectivity for improved diagnosis of neuropsychiatric disorders.

AMR-GNN predicts antimicrobial resistance using a graph neural network framework that integrates multiple genomic representations from whole-genome sequencing data.

The paper develops a disease-specific histone modification prediction model using a large language model and MoE architecture, achieving high AUROC scores and identifying AD-associated variants.

PRISM improves spatial multi-omics analysis by imputing missing data and identifying spatial domains, outperforming existing methods on simulated and real datasets.

SpatialZ generates dense 3D cell atlases from planar spatial transcriptomics data by reconstructing virtual slices, validated with real 3D data and enabling new spatial analyses.

HyperSTAR identifies spatial domains across resolutions using hypergraph learning, outperforming existing methods in spatial omics analysis and revealing breast cancer spatial heterogeneity.

This paper introduces SpatialEx and SpatialEx+ frameworks to integrate spatial multi-omics data using histology as an anchor, achieving high-parameter co-profiling across tissue sections.

The paper develops ACCESS, a multimodal graph neural network to annotate functional dark matter proteins from extreme environments beyond homology-based methods.

This paper introduces AMP-D3, a framework for designing diverse and druggable antimicrobial peptides using diffusion and multimodal deep learning, achieving high accuracy in prediction and validation.

ImmuScope predicts CD4+ T cell immunogenic epitopes using self-iterative multiple-instance learning, improving accuracy with multi-allelic and single-allelic data.

KcatNet predicts enzyme catalytic efficiency using geometric deep learning, outperforming existing methods and providing insights into enzyme-substrate interactions.

The paper introduces tFold-TCR, a model for predicting TCR-pMHC structures, achieving higher accuracy and speed than existing methods and creating a large database of these structures.

This paper introduces ConBoTNet, a deep learning model that improves MHC-II peptide binding affinity prediction using supervised contrastive learning and bottleneck transformers, achieving better performance than existing methods.

This paper introduces EPACT, a deep-learning framework for predicting TCR-antigen binding pairs, showing high accuracy in epitope recognition and TCR repertoire analysis.

The paper introduces Mimosa, a Transformer-based model for miRNA target prediction that incorporates non-canonical base-pairing patterns, achieving better performance than traditional methods.

HPClas is a machine learning classifier using catBoost to identify halophilic proteins with high accuracy, achieving AUROC scores of 0.877 and 0.845 on training and test sets respectively.

SCREEN is a graph neural network that predicts catalytic residues and assesses mutation tolerance in enzymes by integrating structural and functional information.

scRCA improves cell type annotation using noisy reference data by combining a Siamese network with noise-robust loss functions and outperforms existing methods.

MERITS provides a comprehensive database of PE/PPE proteins from Mycobacterium, offering structural and functional information to aid in vaccine and diagnostic development.

This paper introduces scpDeconv, a deep learning method for deconvolving cell type mixtures in proteomic data, showing improved performance across species and technologies.