Radical Numerics introduces Omnii to select targets for personalized mRNA cancer vaccines

Radical Numerics introduces Omnii to select targets for personalized mRNA cancer vaccines

On September 8, Radical Numerics described its post-trained Omnii model: it takes a patient's tumor mutations and MHC variants as inputs, selects vaccine targets, and designs an mRNA cassette sequence. In computational comparisons published by the company, the model ranked immunogenic targets more effectively than several specialized tools.

In June, Radical Numerics introduced Omnii as a genomic model for DNA-based predictions. The company has now trained it for a narrower task: selecting targets from an individual tumor's mutations for a personalized vaccine.

A single tumor can have hundreds of mutations, while an mRNA cassette has room for only a few targets. Some mutations alter a protein, creating peptides, short protein fragments that are absent from healthy cells. The vaccine is intended to direct immune cells called T-cells to recognize these fragments.

Each candidate undergoes two checks. First, the model predicts presentation: whether the peptide will bind to the patient's MHC, a molecule that displays protein fragments on the cell surface, and whether that fragment will appear on the tumor cell. It then assesses immunogenicity: whether a T-cell will recognize and respond to the displayed fragment. Only a target that passes both checks can be included in the cassette.

In a comparison of MHC class I presentation predictions, Omnii achieved an AUROC of 0,939, compared with 0,933 for BigMHC-EL. AUROC measures how often a model ranks immunogenic candidates above nonimmunogenic candidates. For immunogenicity predictions within the same class, the company reports 0,750 for Omnii and 0,558 for BigMHC-IM.

The company also simulated selecting targets from a hundred candidates, with five slots in the cassette and 6% of targets being immunogenic. In this simulation, Omnii selected an average of 1,3 immunogenic targets out of five, compared with 0,7 for BigMHC-IM. After selection, the system designs an mRNA sequence that encodes the selected neoantigens.

Omnii is available as a research preview; Radical Numerics is seeking partners with expertise in cancer immunology and vaccine development.