University of California, San Diego

Variant effect and disease phosphosite prediction that fuses frozen ESM-2 embeddings with normal-mode protein dynamics over AlphaFold residue graphs.

Brain MRI morphometry model that estimates cortical thickness, surface area, and volume in milliseconds instead of hours.

Meta-learning framework that transfers perturbation responses across cell lines, donors, and drugs from a few measured seed perturbations.

Vision-language foundation model linking human brain activation maps and neuroscience text for text-to-brain and brain-to-text generation.

Single-cell foundation model running self-attention across all 27,874 human genes, with Gene Ontology priors injected through a graph network.

Cancer genomics foundation model embedding clinical gene-panel mutations into tumor subtype vectors. Pretrained on 30,328 tumors and 8 networks.

Binding free-energy surrogate trained on 1.4M molecular dynamics frames, ranking docking poses ~28,000x faster than physics-based MMGBSA.

Structure-based drug design pipeline generating 3D molecules in binding pockets, raising zero-shot CrossDocked2020 docking success from 53% to 64%.

Structure-based 3D molecule generation with one diffusion backbone for fragment growing, linker design, scaffold hopping, and side-chain decoration.

MSA design model generating alignments from protein language model embeddings to improve folding accuracy on orphan and low-homology proteins.

RNA inverse folding with an attention-free RWKV language model that designs sequences for a target secondary structure with tunable G-C content.

Multimodal large language model that writes free-form natural-language gene function descriptions directly from a nucleotide sequence and a prompt.

Enhancer prediction from DNA sequence alone, reaching 88.05% accuracy and 76.22% MCC on held-out ENCODE cCRE regions and annotating the whole genome.

Protein structure embedding model that compresses each 3D fold into a single fixed-length vector for proteome-wide similarity search and clustering.

Prostate cancer detection model for MRI and transrectal ultrasound, trained with patch-level contrastive learning across 4,401 patients.

Multimodal foundation model for wearable physiological sensing across PPG, ECG, EEG, GSR, and IMU signals, using channel-aware attention.

Histopathology model predicting extrachromosomal DNA status from routine H&E slides by first inferring the tumor transcriptome from tile features.

Generative masked protein language model with an interpretable concept layer, letting designers set 718 biophysical and annotation concepts directly.

Multi-modal protein foundation model aligning 3D structure and literature text to a sequence anchor through contrastive pretraining.

Diffusion model for de novo protein backbone design that learns a 32-dimensional latent code for global fold geometry and generates conditioned on it.

Template-guided protein design model that miniaturizes, diversifies, or expands a natural protein by decoding a fixed-size probabilistic encoding.

Self-supervised vision foundation model pretrained on 1.3M medical images via second-order graph matching, for segmentation and classification.

Multi-modal LLM answering free-form questions about a compound's indications, pharmacodynamics and mechanism of action from its SMILES string.