Unaffiliated researchers working outside academia and industry, releasing open models across protein design, genomics, and machine learning.
ECG foundation model that reads any subset of the 12 standard leads natively, encoding recordings as variable-size spatiotemporal graphs.
Genomic foundation model with 120M parameters that learns adaptive DNA token boundaries by dynamic chunking, not fixed k-mer or byte-pair tokens.
Protein backbone generator running a diffusion transformer over SaProt structural tokens, with an IPA token cache to speed up de novo design.
De novo drug design model generating target-conditioned ligands by latent diffusion over 1D SELFIES strings, conditioned on protein sequence alone.
Full-precision LoRA fine-tuning of ESM-2 for per-residue binding site prediction, where low-rank constraints curb overfitting on small datasets.
4-bit QLoRA fine-tunes of ESM-2 for per-residue protein binding site prediction, released as a checkpoint family spanning 8M to 650M parameters.