UC Berkeley

Molecular linker design model fine-tuned from Llama 3 that emits PROTAC and fragment linkers as SMILES from natural-language geometry prompts.

Protein evolution model that learns indel dynamics and epistasis from unaligned sequences, simulating trajectories that yield functional proteins.

Phylogeny-aware genomic language model scoring variant effects from whole-genome alignments and species trees across three evolutionary timescales.

Activation domain predictor scoring transcriptional activator strength from protein sequence, with a 20-model ensemble that reports uncertainty.

Plant immune receptor-ligand classifier that scores MAMP epitope immunogenicity from sequence, reaching 73% accuracy on a held-out test set.

End-to-end protein structure alignment from 3D coordinates, matching TM-align quality at LDDT 0.56 and beating Foldseek on SCOPe40 search.

Encoder-decoder codon language model that reverse-translates a protein into species-specific coding sequences for synthetic mRNA design.

Text-to-protein generative model designing de novo sequences from plain-language function descriptions, with designs confirmed by crystallography.

Synthesis planning model that generates full synthetic routes from Enamine building blocks and reaction templates for any target small molecule.

Genomic language model for variant effect prediction that scores deleterious mutations from a single DNA sequence, with no alignment at inference.

Functional genomics models predicting RNA-seq, CAGE, and DNase coverage tracks from DNA sequence using striped bidirectional Mamba and attention.

Autoregressive transformer pretrained on 2,000 hours of intracortical spiking activity, decoding motor intent across subjects, species, and tasks.

Text-guided protein design framework aligning language with sequences for text-conditioned generation, zero-shot editing, and property prediction.

Latent diffusion model for controllable all-atom protein generation that co-designs sequence and structure while training on sequences alone.

Codon-resolution language model suite pairing a bidirectional encoder with an autoregressive decoder over protein-coding sequences.

Blind protein-ligand docking that transfers to binding domains absent from training, scoring 22.6% top-1 on DockGen and 50% on PoseBusters.

DNA language model for genome-wide variant effect prediction, trained by masked language modeling on multispecies genomes with no labeled data.

DNA language model for variant effect prediction across coding and non-coding regions, using whole-genome alignments of 100 vertebrate species.

Protein-nucleic acid complex structure prediction from sequence, folding protein, DNA and RNA chains in one network with confidence estimates.

Small-molecule generation over SMILES through a fixed-size latent space dense enough that gradient-free search over it optimizes chemical properties.

Inverse folding model predicting protein sequence from backbone coordinates, trained on 12 million AlphaFold2-predicted structures.

Protein structure and complex prediction from sequence, in a three-track network that reasons over alignments, distances, and 3D coordinates at once.

Chemical language model pretrained on up to 77 million PubChem SMILES strings for molecular property prediction on the MoleculeNet benchmarks.