Zhangsheng Yu

Probing Physical Intuition in Large Language Models through Partner-Dependent Mutation Prediction

This study evaluates LLMs' ability to predict mutation effects based on structural data without using dedicated predictors, finding comparable accuracy to existing methods.

Ultra-efficient High Resolution 3D Reconstruction of Spatial Omics Data with Neural Transcriptomic Field

This paper introduces NTF, a deep learning framework for high-resolution 3D reconstruction of spatial omics data, achieving fast and scalable results across multiple species and datasets.

Features fusion or not: harnessing multiple pathological foundation models using Meta-Encoder for downstream tasks fine-tuning

The paper introduces Meta-Encoder, a framework that combines multiple pathological foundation models to improve downstream task performance in computational pathology.

ProDualNet: Dual-Target Protein Sequence Design Method Based on Protein Language Model and Structure Model

ProDualNet designs dual-target protein sequences by integrating sequence-structure information from two receptors, showing better recovery and success rates than existing methods.

Unveiling Fine-scale Spatial Structures and Amplifying Gene Expression Signals in Ultra-Large ST slices with HERGAST

HERGAST identifies spatial structures and amplifies gene expression signals in ultra-large spatial transcriptomics data, outperforming existing methods and revealing novel biological insights in cancer.

AI-Driven Efficient De Novo design of GLP-1RAs with Extended Half-Life and Enhanced Efficacy

This paper designs 10,000 de novo GLP-1RAs using AI, with 60 passing stability, efficacy, and diversity criteria, and two showing extended half-lives and superior efficacy in vivo.

A foundation model for generalizable cancer diagnosis and survival prediction from histopathological images

This paper introduces BEPH, a lightweight foundation model for cancer diagnosis and survival prediction from histopathological images, showing improved performance over existing models with limited data.

RGAST: A Relational Graph Attention Network for Multi-Scale Cell-Cell Communication Inference from Spatial Transcriptomics

RGAST improves cell-cell communication inference from spatial transcriptomics by integrating spatial and transcriptional data, revealing directional communication and enhancing spatial domain identification.

HEARTSVG: a fast and accurate method for spatially variable gene identification in large-scale spatial transcriptomic data

HEARTSVG identifies spatially variable genes in spatial transcriptomic data with high accuracy and efficiency, uncovering biologically significant patterns in colorectal cancer.

Predicting Gene Spatial Expression and Cancer Prognosis: An Integrated Graph and Image Deep Learning Approach Based on HE Slides

The paper introduces an IGI-DL model that predicts gene spatial expression from HE images and improves cancer prognosis using a risk score derived from predicted gene expression.