Immunogenic Variant Candidates
Prioritizes variants for immunotherapy based on tumor microenvironment
All unit tests passing including TME edge cases
7 immune-promoting + 4 immune-suppressive hallmarks scored
Frameshift (1.5x), missense, nonsense, splice variants scored
Pathway norms calibrated on 9,415 TCGA samples
Overview
Ranks somatic variants by their potential as immunotherapy targets, integrating tumor microenvironment (TME) permissiveness with clonality, expression, and variant characteristics. Uses the VAE's pathway latent to assess immune context — classifying tumors as HOT (immune-infiltrated), WARM, or COLD — which determines whether immunotherapy approaches are likely to succeed. Combines CNV-aware clonal cell fraction, RNA expression confirmation, variant-type bonuses (frameshift, missense), and driver gene status into a composite prioritization score.
Inputs
4 inputsSomatic variantsVariableVariant list with gene, VAF, variant type, and driver status
z_pathway200 (50×4)Pathway activities for 7 immune + 4 suppressive hallmarks
RNA expressionPer variant geneGene-level expression for transcription confirmation
CNV dataPer variant geneCopy number for CNV-aware clonality estimation
Outputs
3 outputsRanked candidatesRanked listVariants ranked by composite immunogenicity score
TME assessmentHOT/WARM/COLDTumor microenvironment permissiveness classification
Clonality estimatesPer variantCNV-aware cancer cell fraction (CCF) estimates
Mathematical Formulation
Weighted balance of immune-promoting vs suppressive hallmarks
CNV-aware cancer cell fraction from variant allele frequency
Multi-factor prioritization with variant-type and driver bonuses
Key Features
- TME permissiveness scoring from 11 immune/suppressive hallmarks
- CNV-aware clonality estimation (cancer cell fraction)
- Variant-type bonuses: frameshift 1.5x, driver +30% additive
- RNA expression confirmation for transcription validation
- HOT/WARM/COLD TME classification for immunotherapy triage
- Structured abstention for COLD tumors
Key Innovations
- 1VAE-derived immune context for neoantigen prioritization
- 2Integration of clonality + expression + TME in single scoring framework
- 3TME-gated recommendations prevent futile immunotherapy suggestions
- 4Designed as upstream filter for HLA binding / pMHC prediction tools
Hyperparameters
7 (IL2, IFNγ, IFNα, TNFα, complement, allograft, inflammatory)4 (TGF-β, angiogenesis, hypoxia, hedgehog)1.5×+30% additive> 0.6 TME score< 0.3 TME scoreTraining Details
Not a trained model — uses calibrated VAE pathway norms, curated hallmark weights, and evidence-based scoring rules. TME classification thresholds set from TCGA distribution analysis. Designed as upstream prioritization for downstream HLA/pMHC binding prediction.