← Về thư mục
Neoantigen Pipeline - Usage Guide
Overview
This workflow identifies tumor-specific neoantigens from somatic mutations for personalized cancer vaccine design. It integrates HLA typing, MHC binding prediction, and multi-factor immunogenicity scoring to rank vaccine candidates.
Prerequisites
pip install pvactools mhcflurry vatools pandas numpy matplotlib seaborn
mhcflurry-downloads fetch
conda install -c bioconda ensembl-vep arcas-hla optitype samtools
Required databases:
- VEP cache for annotation
- IEDB tools (optional, for additional algorithms)
Quick Start
Tell your AI agent what you want to do:
- "Find neoantigens from my somatic VCF for vaccine design"
- "Predict MHC binding for tumor mutations"
- "Rank neoantigen candidates by immunogenicity"
- "Run the neoantigen pipeline with my HLA types"
Example Prompts
Complete pipeline
"Run neoantigen discovery on my tumor VCF with HLA-A02:01,HLA-B07:02"
"Find vaccine candidates from my annotated somatic mutations"
HLA typing
"Determine HLA types from my tumor RNA-seq BAM"
"Extract HLA alleles for neoantigen prediction"
Binding prediction
"Predict MHC Class I binding for my mutant peptides"
"Find strong binders (<500nM) in my neoantigen candidates"
Ranking
"Score neoantigens by immunogenicity and expression"
"Rank my neoantigen candidates for vaccine prioritization"
| Input |
Format |
Description |
| Somatic VCF |
VCF (VEP-annotated) |
Tumor somatic mutations |
| HLA types |
String |
Comma-separated 4-digit HLA alleles |
| Expression (optional) |
TSV |
Gene-level TPM from tumor RNA-seq |
| Tumor BAM (optional) |
BAM |
For HLA typing if types unknown |
What the Agent Will Do
- HLA Typing - Determines patient HLA alleles from RNA-seq (if not provided)
- VCF Annotation - Adds protein consequences with VEP
- Binding Prediction - Predicts peptide-MHC binding with multiple algorithms
- Neoantigen Calling - Identifies tumor-specific peptides with pVACseq
- Immunogenicity Scoring - Ranks candidates by binding, expression, VAF, and specificity
- Visualization - Generates summary plots of candidate distribution
Key Parameters
| Parameter |
Default |
Description |
| IC50 threshold |
500 nM |
Strong binder cutoff |
| Epitope lengths |
8,9,10,11 |
MHC-I peptide lengths |
| VAF minimum |
0.1 |
Variant allele frequency filter |
| Expression minimum |
1 TPM |
Gene expression filter |
| DAI threshold (WT/MT IC50 ratio) |
>2 moderate, >10 strong |
Agretopicity; >1 = MT binds better than WT |
Tips
- Expression data improves ranking: Include tumor RNA-seq TPM when available
- Use multiple algorithms: MHCflurry + NetMHCpan gives more robust predictions
- Consider Class II: CD4+ T cell help improves vaccine efficacy
- Clonal mutations first: Prioritize high-VAF variants for broader tumor coverage; clonal neoantigen burden (McGranahan 2016) predicts ICI response better than total
- Validate HLA typing: Clinical-grade HLA typing is more reliable than computational; T1K (Song 2023 Genome Res) is the 2024-2026 all-rounder for class I + II + KIR from short read
- Check HLA-LOH: LOHHLA (McGranahan 2017) or DASH (Pyke 2022); somatic loss of HLA abolishes neoantigen presentation in ~17% of pan-cancer (Montesion 2021; >30% HNSCC / NSCLC / cervical); flag and exclude lost-allele predictions
- Pair with TMB / MSI: TMB-H pan-tumor (Vega 2021 panel-calibrated) and MSI-H / dMMR (KEYNOTE-016/164/158) are co-biomarkers; MSI-H supersedes TMB-H per Sha 2020
- immunoinformatics/mhc-binding-prediction - MHCflurry parameters
- immunoinformatics/mhc-class-ii-prediction - Class II (CD4) neoantigens for vaccine help
- immunoinformatics/neoantigen-prediction - pVACtools details
- immunoinformatics/immunogenicity-scoring - Ranking algorithms
- clinical-databases/hla-typing - HLA typing + HLA-LOH detection
- clinical-databases/tumor-mutational-burden - TMB-H ICI biomarker
- clinical-databases/msi-detection - MSI-H / dMMR ICI biomarker
- clinical-databases/somatic-signatures - Mutational mechanism context