Zhentao He

Email / CV / Scholar / LinkedIn / GitHub

I am Zhentao He (James He), a Ph.D. student at Fudan University in the ISTBI (Institute of Science and Technology for Brain-inspired Intelligence), working under the supervision of Prof. Shanfeng Zhu. I am also a Zhuobo Fellow at Fudan University (university doctoral excellence program). My research focuses on machine learning and data mining for biological data. I am interested in developing robust, interpretable, and transferable representations, with applications to multiomics analysis, perturbation modeling, and protein function.

I received my B.E. in Computer Science & Technology from Wuhan University (2021–2025), where I was a Hongyi Honor College student and graduated as an Outstanding Graduate of Wuhan University (Class of 2025). My undergraduate research at the BIOD Lab focused on single-cell multiomics data integration methods under the guidance of Prof. Lihua Zhang.

Research Interests:

Representation Learning
Generalization & Adaptation
Multimodal Learning for Multiomics
Perturbation Prediction & Protein Function Modeling

I am currently seeking opportunities for research visits and joint doctoral training, and I am always happy to discuss potential collaborations. Feel free to reach out via email or check out my GitHub!

News

Feb 04, 2026	🎉 Our paper “Interpretable Data Integration for Single Cell and Spatial Multi-Omics” has been published in Cell Systems! Read the paper here
Sep 01, 2025	Embarking on a PhD Journey at Fudan University.

Selected publications

Cell Syst.

Interpretable Data Integration for Single Cell and Spatial Multi-Omics

Chenghui Yang, Zhentao He, Qing Nie, and Lihua Zhang

Cell Systems, Feb 2026

Published online February 4, 2026

Abs DOI Code

FGOT is a feature-guided optimal transport method for paired and unpaired single-cell multi-omics data as well as paired spatial multi-omics data. It jointly reveals cellular heterogeneity and transcriptional regulatory links, while providing post hoc interpretability for existing integration methods.
PR

MoE-TCR: Mixture-of-experts framework for pan-specific TCR-epitope binding prediction

Wei Qu, Jinxing Li, Zhentao He, Jiayi Wang, and Shanfeng Zhu

Pattern Recognition, Nov 2026

Abs DOI Code

MoE-TCR is a sparsely gated mixture-of-experts framework for pan-specific TCR-epitope binding prediction. It combines diverse expert architectures, including the criss-cross attention-based TECC model, to improve predictive accuracy and generalizability across benchmark datasets.
Commun Biol

SpaDC enables sequence-based integrative analysis and regulatory inference of spatial chromatin accessibility data

Chuanlong Ma, Chenghui Yang, Caiwei Zhen, Zhentao He, Yong Luo, and Lihua Zhang

Communications Biology, Jun 2026

Published June 6, 2026

Abs DOI Code

SpaDC is a graph-regularized convolutional neural network that integrates DNA sequence, chromatin accessibility, and spatial location for spatial epigenomics data denoising, integration, spatial domain identification, and regulatory-network inference.