University of Toronto

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

A multimodal reasoning LLM that fuses protein-language-model embeddings with biological context to generate interpretable reasoning traces for protein function and GO-term annotation.

A tokenizer-free, hierarchical autoregressive genomic foundation model that adaptively chunks raw nucleotides, enabling efficient long-context learning and zero-shot variant and gene predictions.

A genome language model that adds evolutionary-rate prediction as a pretraining task, improving representations and variant effect prediction over sequence-only training.

A Joint-Embedding Predictive foundation model for echocardiography, pretrained on 18M cardiac ultrasound videos to learn artifact-robust anatomical representations.

Mamba-based mature RNA foundation model using contrastive learning on splice isoforms and 400+ mammalian species orthologs for mRNA property prediction.

An open transformer foundation model for 12-lead electrocardiograms, pretrained on 1.5M ECGs with hybrid contrastive and generative self-supervision.

A generative pre-trained transformer for single-cell multi-omics, pretrained on 33 million human cells for cell annotation, batch correction, and perturbation prediction.

A promptable foundation model for universal medical image segmentation, fine-tuned from SAM on 1.57M image-mask pairs spanning 10 imaging modalities and 30+ cancer types.