DEVOURED
Genome language models can design cancer vaccines
Radical Numerics is using its Omnii genome language model to predict which tumor mutations are most immunogenic for personalized mRNA cancer vaccines.
What: Radical Numerics demonstrated that their multimodal model, Omnii, outperforms specialized tools like BigMHC-IM and NetMHCpan-4.1-BA in predicting how likely a patient's immune system is to recognize and attack specific tumor mutations. The system takes patient tumor DNA and MHC sequences to output optimized mRNA sequences.
Why it matters: This signals that general-purpose biological models are reaching parity with, or exceeding, specialized bioinformatics tools, suggesting that foundation models will play a central role in personalized medicine design.
Takeaway: Researchers can request early access to Omnii for cancer vaccine design by contacting health@radicalnumerics.ai.
Deep dive
- Pipeline: Reads tumor DNA to identify unique mutations.
- Presentation: Predicts if mutated peptides bind to patient MHC molecules.
- Immunogenicity: Predicts if a T-cell will actually mount an immune response against the presented peptide.
- Design: Encodes top-ranked candidates into an mRNA cassette.
- Performance: Omnii achieved AUROC scores of 0.750 for Class I immunogenicity and 0.778 for Class II, significantly higher than task-specific tools.
- Evaluation: Precision was measured at k=5, 10, and 20 candidates, with Omnii showing the greatest advantage in low-budget settings.
Decoder
- MHC (Major Histocompatibility Complex): Proteins on the cell surface that display peptide fragments to T-cells.
- Neoantigen: A new protein created by a cancer mutation that the immune system may recognize as foreign.
- mRNA cassette: The specific sequence of mRNA used to instruct cells to produce desired antigens for a vaccine.
- AUROC (Area Under the Receiver Operating Characteristic curve): A performance metric for classification models; 0.5 is random guessing, 1.0 is perfect prediction.
Original article
Full article content is not available for inline reading.
Explore all 54 articles in today's digest
Fresh dev news, AI-digested daily. Never miss a trend at devoured.fyi