Zhangyang Gao

A set-aware latent transport model predicts treated cell populations from unpaired controls, keeps perturbation-specific signal in CRISPR, drug and cytokine data, and guided a cytokine choice tested in three donors.

This paper introduces modality-chain reasoning for multimodal protein modeling and design, showing improved performance in structure prediction, inverse folding, and fitness prediction compared to existing models.

This paper introduces FoldToken4, a model that generates consistent and hierarchical multiscale fold languages for protein structures, reducing redundant training across scales.

FoldToken3 compresses protein structures into 256 tokens or less while maintaining reconstruction quality comparable to FoldToken2.

FoldToken2 improves protein structure representation by creating a compact, invariant language that outperforms previous methods in reconstruction accuracy.