Automating neoantigen selection for personalized cancer vaccine design

Area of Science:

• Immunogenomics
• Immuno-oncology
• Computational Biology

Background:

• Personalized cancer vaccines (PCVs) leverage cancer cell-specific somatic variants to elicit immune responses.
• Neoantigens, derived from these variants and presented by MHC molecules, are key targets for PCVs.
• Over 100 clinical trials are investigating PCVs, highlighting their therapeutic potential.

Purpose of the Study:

• To address the bottleneck in PCV design caused by manual neoantigen selection.
• To introduce NEAT (Neoantigen Evaluation & Automated Triage), a machine learning tool for automated neoantigen prioritization.
• To enhance the scalability and reproducibility of PCV development workflows.

Main Methods:

• Developed a machine learning model trained on data from 33 patients and 1,943 peptides across 3 clinical trials.
• Utilized features including variant allele frequency, RNA expression, driver gene status, and binding/presentation scores.
• Integrated the NEAT model into pVACtools v7.0.0 for automated prediction of peptide suitability for PCVs.

Main Results:

• The NEAT model achieved high performance with a sensitivity of 0.847, specificity of 0.924, and an AUC of 0.955.
• The model automatically categorizes peptides as accepted, rejected, or requiring further review.
• Successful incorporation of the model into pVACtools signifies a step towards automated PCV design.

Conclusions:

• NEAT significantly reduces the manual effort and time required for neoantigen selection in PCV development.
• The automated triage system enhances the efficiency and scalability of transitioning from patient samples to vaccine manufacturing.
• This advancement supports the broader implementation and accessibility of personalized cancer vaccines.