Computational oncology tools for tumor mutational burden, microsatellite instability, and immunotherapy biomarker prediction.

Retired tools

hrd_score, liquid_biopsy_ctdna and tumor_fraction_estimate are retired and can no longer be run; their log2-ratio and VAF-median estimators were removed on 2026-10-04. For HRD scoring use an allele-specific copy-number HRD caller such as scarHRD, for tumor fraction use ichorCNA on low-pass WGS, and for ctDNA or MRD use a dedicated ctDNA/MRD pipeline. tmb_msi_calculate still computes TMB from an annotated VCF.

Characterize tumors and predict immunotherapy biomarkers computationally — compute tumor mutational burden and microsatellite instability, predict MHC binding, and identify neoantigens for personalized immunotherapy. Reach for these tools when you have somatic variant or peptide data and need clinically anchored readouts such as checkpoint inhibitor eligibility. They sit in the oncology-specific analytics arc, feeding downstream into therapy selection and clinical-trial matching. See the Targets & Modalities overview for where this fits.

MHC binding prediction is a simple anchor-motif heuristic, not NetMHCpan

mhc_binding_predict's backend (oncology_analytics_app/main.py) does not run NetMHCpan-4.1/NetMHCIIpan-4.0 (the cited method) — it scores 9-mer peptides with a BIMAS/SYFPEITHI-style position-weight matrix over the P2/P9 anchor residues, then converts that score into a pseudo-ic50_nm value through an arbitrary exponential formula (method: "anchor_pwm" in the output). Only exact 9-mers are scored — 8, 10, and 11-mer class I peptides and all class II (13-25mer) peptides are silently skipped, not processed. Treat its IC50/percentile-rank output as a rough anchor-motif heuristic, not a calibrated binding-affinity prediction, and don't rely on it for lengths other than 9.

Oncology Analytics overview

The Oncology Analytics toolkit provides three tools for tumor characterization and immunotherapy biomarker prediction: TMB/MSI calculation, MHC binding prediction, and neoantigen identification. All tools run on Modal GPU compute; depending on the tool and input size, expect anywhere from a few minutes (MHC binding prediction) to tens of minutes (neoantigen prediction).

Architecture

User Query (VCF, mutation list, peptide sequences, HLA types)
  --> Intent Detection
       |- Genomic instability --> TMB/MSI Calculate
       |- Immunotherapy --> MHC Binding Predict or Neoantigen Predict
  --> Modal GPU Compute (minutes to tens of minutes)
  --> Structured Results with Clinical Thresholds

Tool inventory

Genomic instability tools

Tool	Name	Description	Key Output
TMB/MSI Calculate	`tmb_msi_calculate`	Computes tumor mutational burden and microsatellite instability status	TMB (mut/Mb), MSI status (MSS/MSI-L/MSI-H)

Immunotherapy biomarker tools

Tool	Name	Description	Key Output
MHC Binding Predict	`mhc_binding_predict`	Predicts peptide binding affinity to MHC class I/II molecules	Binding affinity (nM), percentile rank, binding classification
Neoantigen Predict	`neoantigen_predict`	Identifies neoantigens from somatic mutations for immunotherapy	Ranked neoantigen candidates, binding scores, expression levels

Capabilities

Tumor mutational burden and microsatellite instability

TMB measures the total number of somatic mutations per megabase of coding genome. The FDA-recognized threshold of 10 mut/Mb identifies patients likely to respond to immune checkpoint inhibitors. MSI status is classified as microsatellite stable (MSS), low instability (MSI-L), or high instability (MSI-H). MSI-H tumors are eligible for pembrolizumab regardless of tumor type.

MHC binding prediction

MHC binding prediction scores 9-mer peptides against common HLA-A/B alleles with an anchor-residue heuristic (see the note above — it isn't the cited NetMHCpan model, and it only handles exact 9-mers, not the fuller 8-11mer class I / 13-25mer class II range). Strong binding (IC50 < 50 nM) or weak binding (IC50 < 500 nM) by that heuristic's own pseudo-IC50 are candidates for further, more rigorous evaluation.

Neoantigen prediction

Neoantigen prediction identifies tumor-specific peptides generated by somatic mutations that may be recognized by the immune system. The pipeline processes somatic mutations through peptide generation, MHC binding prediction, expression filtering, and clonality assessment to produce ranked neoantigen candidates for personalized immunotherapy.

When to use each tool

I want to...	Use this tool
Determine checkpoint inhibitor eligibility	`tmb_msi_calculate`
Evaluate peptide immunogenicity	`mhc_binding_predict`
Design a personalized cancer vaccine	`neoantigen_predict`
Run a full immunotherapy biomarker panel	`tmb_msi_calculate` + `neoantigen_predict`

When to use other tools instead

I want to...	Use this instead
Search for cancer-associated genes	An NCBI Entrez search or OpenTargets associations (Genomics)
Look up known cancer variants	A ClinVar search (Genomics)
Predict protein structure of a mutant	ESMFold
Dock a drug to a cancer target	DiffDock
Assess drug safety and toxicity	ADMET-AI or a Safety Passport
Search clinical trials for a cancer type	ClinicalTrials.gov

Clinical thresholds reference

Biomarker	Threshold	Interpretation	Clinical Use
TMB	≥ 10 mut/Mb	TMB-high	Pembrolizumab eligibility
MSI Status	MSI-H	Microsatellite instable	Checkpoint inhibitor eligibility
MHC Binding	IC50 < 50 nM	Strong binder	Neoantigen candidate
MHC Binding	IC50 < 500 nM	Weak binder	Possible neoantigen

Citations

Marquard AM, et al. Pan-cancer analysis of genomic scar signatures associated with homologous recombination deficiency suggests novel indications for existing cancer drugs. Biomark Res. 2015;3:9. doi: 10.1186/s40364-015-0033-4

Marabelle A, et al. Association of tumour mutational burden with outcomes in patients with advanced solid tumours treated with pembrolizumab: prospective biomarker analysis of the multicohort, open-label, phase 2 KEYNOTE-158 study. Lancet Oncol. 2020;21(10):1353-1365. doi: 10.1016/S1470-2045(20)30445-9

Reynisson B, et al. NetMHCpan-4.1 and NetMHCIIpan-4.0: improved predictions of MHC antigen presentation by concurrent motif deconvolution and integration of MS MHC eluted ligand data. Nucleic Acids Res. 2020;48(W1):W449-W454. doi: 10.1093/nar/gkaa379

Chat usage

Example prompts

"Calculate TMB and MSI status from these somatic variants"
"Predict MHC class I binding for peptides with HLA-A02:01 and HLA-B07:02"
"Identify top neoantigen candidates from these mutations for a cancer vaccine"

Workflow chaining

Feeds into: Compound Passport (therapy selection), Target Passport (druggability), Clinical Trial matching, Neoantigen vaccine design, Safety Passport (toxicity assessment)
Accepts from: Genomics tools (gene/variant data), VCF uploads, BLAST (mutation identification), Literature search (biomarker evidence)

Oncology Analytics on the platform