Autoregressive model for structure-based drug design. Generates drug-like molecules conditioned on protein pocket structures using GPT-style generation over molecular tokens.

Generative pre-trained transformer for single-cell multi-omics. Trained on 33 million cells across tissues and species. Excels at cell type annotation, gene perturbation prediction...

Context-aware gene network transformer from Harvard. Pre-trained on 30 million single-cell transcriptomes. Enables transfer learning for disease modeling, gene dosage sensitivity, ...

Large-scale cell language model that learns universal cell representations from tens of millions of cells. Supports zero-shot cell type annotation and cross-tissue transfer.

Open-source multimodal biomedical foundation model from Tsinghua AIR (China). BioMedGPT-R1 has 10B+ parameters, achieving 67.1% on USMLE. Supports molecular, protein, and text moda...

Generative pre-trained transformer from Microsoft for biomedical text mining and generation. Pre-trained on 15 million PubMed abstracts. Strong on relation extraction, QA, and docu...

Teaching LLMs to use NCBI Web APIs for genomics tasks. Uses GPT-4 as backbone with tool-augmented prompting to query Gene, OMIM, and other NCBI databases. Pioneering example of bio...

Google's medical LLM reaching expert-level performance on medical licensing exams. Achieves 85%+ on MedQA. Demonstrates strong clinical reasoning and medical knowledge.

Automated machine learning system for biological sequence analysis from MIT. Automatically selects models, features, and hyperparameters for DNA, RNA, protein, and glycan sequence ...

LLM benchmark and model for bioinformatics code generation. Tests AI ability to write correct bioinformatics pipelines across 253 real-world tasks spanning genomics, proteomics, an...