University College London

Conditional flow matching model that reconstructs 4D left-ventricular geometry and motion from a 12-lead ECG and basic demographic variables.

Protein domain embeddings that assign CATH homologous superfamilies by nearest-centroid lookup, staying accurate below 20% sequence identity.

Pediatric brain MRI segmentation model trained on synthetic scans, so a single 3D U-Net labels clinical acquisitions from neonates to young adults.

Protein language model pretrained on structural domain segments, encoding fold and contact signals for remote-homology detection from sequence alone.

Promptable DNA language model that generates multi-kilobase plasmid sequences from plain-language component specs, refined with verifiable rewards.

Protein function prediction model that fuses sequence, structure, text, and interaction embeddings with learned gating to assign Gene Ontology terms.

Protein-family language model trained on unaligned homolog sets for zero-shot variant fitness prediction and design. ProFam-1 holds 251M parameters.

Histopathology model predicting homologous recombination deficiency from H&E slides in ovarian cancer, reaching 0.846 AUC and 0.938 specificity.

Modality-agnostic foundation model for human brain imaging that runs five core neuroimaging tasks across uncalibrated CT and MRI without retraining.

Structure-based drug design model that inpaints a 3D ligand density into an empty protein pocket, then decodes those voxels into valid SMILES.

Brain foundation model unifying EEG and MEG in a single encoder via a shared discrete tokenizer that transfers across sensor layouts and montages.

Cryptic protein binding site prediction from sequence, backed by a database of 5,151 cryptic sites mined from 6 million apo-holo PDB alignments.

Whole-heart segmentation foundation model for CT and MRI, pretrained self-supervised on unlabeled cardiac scans with an xLSTM-UNet backbone.

Zero-shot pathogenicity scoring for in-frame insertions and deletions from protein language model likelihoods over residues both alleles share.

Antibody chain pairing model that scores whether a heavy and a light chain are cognate partners, fine-tuned from an antibody-specific language model.

Crossmodal diffusion model synthesizing bulk tumor gene expression from H&E whole-slide images, so grading and survival prediction need no RNA assay.

Equivariant graph network that embeds a protein domain structure as a 128-dimensional vector, turning fold recognition into nearest-neighbour search.

Neuroscience literature language model that predicts experimental outcomes from study methods, LoRA-tuned on 1.3 billion tokens of papers.

Latent diffusion model for de novo DNA sequence generation, sampling species-conditioned promoter and gene regions of 256 or 2,048 base pairs.

Self-supervised foundation model for retinal imaging, pretrained on 1.6 million unlabelled fundus and OCT scans to detect ocular and systemic disease.

Invariant point attention network that segments a protein structure into domains, clustering residues bottom-up on experimental and AlphaFold models.