Westlake University

Distributional flow matching model for single-cell perturbation prediction that models population-level expression shifts using a graph-aware differential-attention transformer.

A language model that generates small-molecule structures directly from transcriptomic phenotypes — gene up/down-regulation signatures — for phenotype-driven drug discovery.

A 16B-parameter framework for de novo protein design from natural language, converting text descriptions into functional protein sequences via two-stage structure-conditioned generation.

Multimodal, multidomain, multilingual medical foundation model that performs zero-shot clinical diagnosis and report generation from chest X-ray and CT images across English and Chinese.

An 80B-parameter multimodal protein-language model that decodes protein function through natural language dialogue, integrating sequence, structure, and evolutionary context.

Foundation model pre-trained on 65M cryo-EM particle images via contrastive learning, enabling zero-shot classification, pose clustering, and quality assessment.

Tri-modal protein language model unifying sequence, structure, and function via contrastive learning, enabling natural-language protein search across billions of entries.

Protein language model combining amino acid and Foldseek 3Di structural tokens, outperforming ESM-2 across 10 downstream tasks including mutation effect prediction.

A multimodal generative AI assistant for pathology, pairing the PathCLIP vision encoder with a Vicuna-13B LLM and a toolkit of eight pathology-specific models.