University of Toronto

Single-cell foundation model pretrained on human brain organoids that predicts transcriptome-wide responses to knockdown of any protein-coding gene.

Binary protein classification over frozen ESM3 embeddings, using transformer contextualization and learned attention pooling instead of fixed pooling.

Whole-transcriptome inference from label-free live-cell phase-contrast microscopy, predicting 18,085 genes without staining or lysing the cells.

Antisense oligonucleotide activity prediction from sequence, position-specific chemistry, dose, and cell context. Spearman 0.5970 on ASO Atlas.

Protein language model for microbial smORF-encoded small proteins, pairing multi-scale convolutions with transformer layers in a compact encoder.

Contrastive promoter-protein pretraining that aligns bacterial promoters with their encoded proteins to learn regulatory genomics representations.

Multimodal reasoning LLM for protein function prediction, fusing protein language model embeddings to emit interpretable GO-term reasoning traces.

Protein function prediction model that autoregressively generates Gene Ontology terms from amino acid sequence instead of classifying fixed labels.

Tokenizer-free genomic foundation model that adaptively chunks raw nucleotides, enabling zero-shot variant fitness and gene essentiality prediction.

Genome language model that adds evolutionary-rate prediction to pretraining, improving representations for variant effect and regulatory genomics.

Joint-embedding predictive foundation model for echocardiography, pretrained on 18M cardiac ultrasound videos for artifact-robust representations.

Micronuclei instance segmentation for fluorescence microscopy, using an anchor-tuned Mask R-CNN to detect micronuclei and link them to parent nuclei.

Image-to-image translation from label-free phase-contrast microscopy to H&E-like images, so pretrained histopathology models run on live cells.

Flow matching model that builds any non-canonical amino acid into a protein pocket from its SMILES string, at 1.43 Å mean RMSD on held-out ncAAs.

Synthesizable 3D molecule generation that jointly samples building blocks, reactions, and atomic coordinates, returning a synthesis route per design.

Single-cell perturbation foundation model predicting transcriptomic responses to CRISPR and small-molecule interventions in cancer cells.

DNA-LLM reasoning model fusing genome foundation model embeddings with an LLM to produce step-by-step pathway and variant effect explanations.

Bacterial genome language model tokenizing whole genomes as ordered conserved elements; frozen embeddings beat Pfam baselines on 23 of 25 phenotypes.

Promptable 3D medical image and video segmentation foundation model fine-tuned from SAM 2.1, cutting lesion annotation time by up to 92%.

Protein segmentation that locates folded domain, sub-domain, and disordered region boundaries from frozen ProtT5 embeddings without any training step.

Chest X-ray encoder that detects CT-level abnormalities by aligning radiographs with 3D CT volumes and radiology reports in a shared embedding space.

Multimodal small-molecule foundation model contrastively pretrained over SMILES, molecular graphs, and fingerprints for antibiotic screening.

Spatial transcriptomics foundation model continually pretrained on 30 million profiles, with a protocol-aware mixture-of-experts decoder.

3D medical imaging foundation model self-supervised on roughly 100,000 MRI, CT, and PET volumes spanning more than ten organs and three modalities.

Multi-omics instruction-tuned LLM that reads DNA, RNA, protein, and multi-molecule sequences and answers natural-language questions about them.

Latent diffusion model that paints high-resolution Cell Painting images of cells responding to a chemical compound or an over-expressed gene.

Chemical language model of the human metabolome that generates and ranks candidate structures for unidentified mass spectrometry peaks.

DNA methylation foundation model over 49,156 array CpG sites. Imputes missing values, embeds samples, and predicts epigenetic age and disease risk.

Cellular senescence prediction from protein sequence, pairing ESM-2 embeddings with a hybrid BiLSTM-CNN classifier at 86.43% test accuracy.

Mamba-based mature RNA foundation model, contrastively trained on splice isoforms and 400+ mammalian species orthologs for mRNA property prediction.

Open transformer foundation model for 12-lead electrocardiograms, pretrained on 1.5 million unlabeled ECGs with a wav2vec 2.0 self-supervised recipe.

Generative pretrained transformer trained on 33 million human cells for single-cell annotation, batch correction, and perturbation prediction.

Promptable foundation model for universal medical image segmentation, fine-tuned from SAM on 1.57M image-mask pairs across 10 imaging modalities.

Discrete diffusion model for protein sequence and MSA generation, enabling controllable de novo design directly in sequence space without structure.

All-atom peptide conformational sampling from sequence, using hypernetwork-conditioned diffusion trained by energy against a molecular force field.

EEG foundation model pretrained on clinical recordings with a wav2vec 2.0-style contrastive task, transferring to BCI decoding and sleep staging.

Histopathology patch encoder giving 1024-dimensional embeddings for slide search and cancer subtyping, trained on TCGA with weak slide labels.

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