← Về thư mục
📄 / / profiles / r-and-d / raw_evidence / OpenAI4S / skills / bioskills / bio-clinical-databases-msi-detection / SKILL.md

name: bio-clinical-databases-msi-detection description: Calls microsatellite instability from WES/WGS/targeted-panel with MSIsensor, MSIsensor-pro, MSIsensor-ct (panel-aware), mSINGS, and MANTIS for FDA pembrolizumab MSI-H pan-tumor / Lynch syndrome / dMMR ICI biomarker. Use when stratifying ICI eligibility (Le 2015), pairing MSI with TMB-H (Sha 2020 / Salem 2018), screening Lynch syndrome (universal IHC + MSI), or distinguishing MSI-H tumors from POLE-exo hypermutator with overlapping signatures. origin: openai4s category: bioskills/clinical-databases metadata: tool_type: cli primary_tool: MSIsensor-pro third_party: name: GPTomics/bioSkills repository: https://github.com/GPTomics/bioSkills commit: d91ed3d563019e649dc854c56ccd62551359488a license: MIT

Version Compatibility

Reference examples tested with: MSIsensor-pro 1.2+, MSIsensor 0.6+, MANTIS 1.0.5+, samtools 1.19+, mSINGS 5.6+, pandas 2.2+, cyvcf2 0.30+. FDA pembrolizumab MSI-H / dMMR pan-tumor approval is from 2017 (Le 2015 NEJM; KEYNOTE-016/164/158); approval extended to colorectal first-line in 2020.

Before using code patterns, verify installed versions match. If versions differ: - Python: pip show <package> then help(module.function) to check signatures - CLI: <tool> --version

If code throws ImportError, AttributeError, or TypeError, introspect the installed package and adapt the example to match the actual API rather than retrying. MSIsensor-pro replaces MSIsensor for tumor-only assays; MSIsensor-ct is the bTMB-equivalent for ctDNA panels.

MSI Detection; The Companion ICI Biomarker to TMB

'Detect MSI status from this somatic sequencing data' -> Profile microsatellite instability across canonical loci (Bethesda 5 panel + extended NGS-derived sites); classify MSI-H / MSS / MSI-L per Bethesda / FDA / KEYNOTE convention.

The Regulatory and Trial Landscape

Event Year Threshold Notes
Le 2015 NEJM 2015 MSI-H + ICI in CRC The seminal paper: pembrolizumab in MSI-H CRC ORR 40% vs 0% MSS
FDA pembrolizumab MSI-H / dMMR pan-tumor 2017 MSI-H First tissue-agnostic FDA approval (KEYNOTE-016/164/158)
FDA pembrolizumab first-line MSI-H CRC 2020 MSI-H + first-line CRC KEYNOTE-177
CheckMate 142 2017-2018 MSI-H + nivolumab/ipilimumab Pan-tumor MSI-H second-line
ESMO 2024 2024 MSI-H Maintained pan-tumor MSI-H biomarker
Universal Lynch screening -- IHC + MSI on all CRC <= 70 yr NCCN / ACG / EGAPP guidelines

MSI vs dMMR vs TMB-H: The Conceptual Hierarchy

Term Definition Method Relationship
dMMR (deficient MMR) Loss of MMR protein function IHC (MLH1, MSH2, MSH6, PMS2) Causes MSI
MSI-H Microsatellite instability high PCR-based Bethesda or NGS Consequence of dMMR
Lynch syndrome Germline MMR mutation Germline sequencing Causes ~50% of MSI-H CRC; rest are sporadic (MLH1 hyper-methylation)
TMB-H >= 10 mut/Mb NGS panel / WES Statistical correlate of MSI-H
POLE-exo hypermutator POLE proofreading defect Sequencing / signatures Hypermutator WITHOUT MMR-D; MSI-stable typically

MSI-H + TMB-H overlap (Chalmers 2017 Genome Med 9:34): - ~83% of MSI-H tumors are TMB-H. - ~16% of TMB-H solid tumors are MSI-H. - Sha 2020 Cancer Discov: MSI-H is the more established dMMR biomarker for ICI decisions; TMB-H not additive.

POLE-exo vs MMR-D: - POLE-exo (SBS10a/10b): hypermutator (100-300 mut/Mb pure); typically MSI-stable. - MMR-D (SBS6/15/26/44 + ID1/2): 30-50 mut/Mb typical; MSI-H. - POLE-exo + MMR-D (SBS14 + SBS20): ultra-hypermutator >=500 mut/Mb; MSI-H.

Tool Taxonomy

Tool Paired Tumor-only ctDNA Algorithm Fails when
MSIsensor (Niu 2014 Bioinformatics) Yes No No Bayesian + read-length distribution Tumor-only data (no baseline); cohort baseline missing
MSIsensor-pro (Jia 2020 Genom Proteom Bioinform) Optional Yes No Distribution comparison to baseline Baseline cohort not provided; panel < 50 loci
MSIsensor-ct (Han 2021 Brief Bioinform) -- -- Yes cfDNA-aware Tumor fraction < 3%; low ctDNA shed
MANTIS (Kautto 2017 Oncotarget) Yes No No Step-wise difference Tumor-only; low coverage at microsatellites
mSINGS (Salipante 2014 Clin Chem) -- Yes No Background panel (unstable-loci fraction) Background panel poorly characterized for cohort

Operational consensus 2024-2026: - Tumor + paired normal WES: MSIsensor or MANTIS. - Tumor-only assay (commercial panels, often unpaired): MSIsensor-pro with reference baseline. - ctDNA / liquid biopsy: MSIsensor-ct. - Lynch screening: IHC FIRST (rules out 90%+); MSI-PCR / NGS confirmatory.

Decision Tree by Scenario

Scenario Recommended path Why
Tumor + paired normal WES MSIsensor (standard) Reference paired-normal comparison
Tumor-only WES/panel MSIsensor-pro with panel baseline No matched normal needed
ctDNA / liquid biopsy MSIsensor-ct cfDNA-aware
Lynch syndrome screening Universal IHC + MSI (NCCN) IHC catches 90%+; MSI for IHC-equivocal
FDA pembrolizumab eligibility Validate per FoCR PCR + IHC + NGS concordance Cross-platform required
MSI-H + TMB-H concurrence MSI-H is primary biomarker Sha 2020; TMB-H not additive
POLE+MMR ultra-hypermutator Sigprofiler signatures (SBS14, SBS20) Mechanism beyond MSI alone
Sporadic MSI-H Confirm MLH1 hypermethylation; rule out Lynch Distinguishes sporadic vs germline
MSI-stable + TMB-H Investigate POLE-exo signature (SBS10a/10b) POLE-exo causes hypermutator without MSI
Pan-tumor screening MSI + IHC + TMB combined Multiple modalities for ICI eligibility

Bethesda Panel and Modern NGS-Derived Loci

The original NCI/Bethesda reference panel (Boland 1998) used BAT-25 and BAT-26 plus three dinucleotide markers (D2S123, D5S346, D17S250); >= 2 of 5 loci unstable -> MSI-H. Modern PCR assays use the mononucleotide pentaplex (the current clinical standard), which replaced the dinucleotide markers for improved cross-population specificity: - BAT-25 (chr4) - BAT-26 (chr2) - NR-21 (chr14) - NR-24 (chr2) - MONO-27 (chr2)

NGS-based MSI panels use 50-1000+ microsatellite loci. MSI-H requires unstable status at >=40% of tested loci typically (varies by panel calibration).

Standard Workflow: MSIsensor-pro Tumor-Only

Goal: Compute MSI status from tumor-only WES/panel.

Approach: Generate baseline from population reference; compare patient tumor.

# Generate microsatellite list from reference genome (one-time)
msisensor-pro scan -d /reference/GRCh38.fa -o microsatellites.list -p 1 -m 5

# Generate baseline from N normal control samples (one-time per panel)
msisensor-pro baseline -d microsatellites.list -i normal_samples.list -o baseline.list -b 16

# Score tumor sample. The `-i sample_id` flag is uncommon: in typical msisensor-pro
# usage the sample identifier is derived from the BAM file -- verify the flag set
# against `msisensor-pro pro --help` for the installed release.
msisensor-pro pro \
    -d microsatellites.list \
    -t tumor.bam \
    -o msi_output \
    -b 16 \
    --baseline baseline.list

# Output: msi_output_all (raw); msi_output_unstable (unstable loci); msi_output.txt (summary)
# Critical column: %_unstable. Threshold MSI-H typically >= 20-30% depending on panel.

Paired Tumor-Normal MSIsensor

msisensor msi \
    -d microsatellites.list \
    -n normal.bam \
    -t tumor.bam \
    -o msi_paired_out \
    -b 16

# Output: %_unstable in paired comparison
# MSI-H threshold: >= 20% by FoCR guidance; varies 10-30% across studies

MANTIS Step-wise Difference

mantis.py \
    -t tumor.bam \
    -n normal.bam \
    -b microsatellite_targets.bed \
    --threads 8 \
    -o mantis_output

# Output: mantis_output.kmer_counts (raw), mantis_output (status)
# Threshold MSI-H: stepwise difference > 0.4 (default)

MSI-H Classification Logic

import pandas as pd


def classify_msi(unstable_percentage, panel_calibrated_cutoff=20.0):
    '''Classify MSI status from percentage of unstable loci.

    Bethesda PCR: >=2 of 5 unstable -> MSI-H (40% loci)
    NGS: panel-specific cutoffs typically 10-30%
    Concordance: MSI-PCR + IHC + NGS should agree (FoCR)
    '''
    if unstable_percentage >= panel_calibrated_cutoff:
        return 'MSI-H'
    elif unstable_percentage >= panel_calibrated_cutoff / 2:
        return 'MSI-L (intermediate; treat as MSS clinically per FDA)'
    else:
        return 'MSS'


def msi_lynch_workflow(msi_status, ihc_results, mlh1_methylation_status, germline_test):
    '''Standard Lynch syndrome workflow.

    Args:
        msi_status: 'MSI-H' / 'MSS' / 'MSI-L'
        ihc_results: dict {MLH1: 'retained' or 'loss', MSH2, MSH6, PMS2}
        mlh1_methylation_status: 'methylated' (sporadic) / 'unmethylated' (Lynch suspect)
        germline_test: 'positive' / 'negative' / 'not_performed'
    '''
    if msi_status != 'MSI-H':
        return 'No further Lynch screening indicated'

    ihc_loss = [gene for gene, status in ihc_results.items() if status == 'loss']
    if not ihc_loss:
        return 'MSI-H with retained IHC; consider Lynch with germline testing'

    if 'MLH1' in ihc_loss:
        if mlh1_methylation_status == 'methylated':
            return 'Sporadic MSI-H (MLH1 hypermethylation); not Lynch'
        elif mlh1_methylation_status == 'unmethylated':
            return 'Lynch suspect (MLH1 loss without methylation); proceed with germline testing'
        else:
            return 'MLH1 loss; perform methylation test'

    return f'MSH2/6/PMS2 loss ({", ".join(ihc_loss)}); strong Lynch suspect; germline testing'


def msi_tmb_ici_decision(msi_status, tmb_value, tumor_type=None, dmmr_ihc=None):
    '''Integrated ICI eligibility from MSI + TMB.

    Sha 2020: MSI-H is primary biomarker; TMB-H not additive.
    McGrail 2021: TMB-H NOT endorsed for breast/prostate/glioma alone.
    '''
    msi_high = msi_status == 'MSI-H'
    dmmr_positive = dmmr_ihc == 'positive'
    tmb_h = tmb_value >= 10

    if msi_high or dmmr_positive:
        return ('ICI eligible: MSI-H or dMMR (FDA pembrolizumab 2017 pan-tumor; KEYNOTE-016/164/158); '
                'TMB-H is not additive (Sha 2020).')
    if tmb_h and tumor_type and tumor_type.lower() in ('breast', 'prostate', 'glioma'):
        return ('TMB-H but tumor type excluded by ESMO 2024 / McGrail 2021. '
                'Consider tumor-type-specific cutoff.')
    if tmb_h:
        return 'TMB-H pan-tumor (FDA pembrolizumab 2020); ICI eligible.'
    return 'MSS + TMB-low. Standard chemo per tumor type.'

Per-Operation Failure Modes

1. Tumor-only with paired-normal tool - Trigger: Run MSIsensor on tumor-only BAM. - Mechanism: MSIsensor requires paired normal for baseline comparison. - Symptom: Tool errors or produces unstable noisy result. - Fix: Use MSIsensor-pro for tumor-only; or use mSINGS background-panel approach.

2. Panel size too small - Trigger: 30-locus panel called MSI-H based on 20% threshold (= 6 unstable loci). - Mechanism: Small panel + stochastic unstable rates produce high false-positive rates. - Symptom: False-positive MSI-H in WES-comparable panels with < 50 microsatellite loci. - Fix: Validate panel calibration with reference cohort; use panel-specific cutoff; minimum 50 informative loci.

3. IHC vs MSI discordance not investigated - Trigger: IHC retains all four MMR proteins; MSI-H by sequencing. - Mechanism: IHC may miss subtle loss; MSI may include MSH6-only subtype (more variable); rare germline POLE+MMR ultra-hypermutators show MSI. - Symptom: Apparent discordance; classification ambiguous. - Fix: Cross-check with germline MMR sequencing; check for POLE-exo on Sigprofiler.

4. MSI-H + Lynch syndrome confusion - Trigger: Report MSI-H tumor as "Lynch syndrome". - Mechanism: ~50% of MSI-H CRC is sporadic (MLH1 hypermethylation, not germline Lynch). - Symptom: Incorrect family counseling; wrong screening. - Fix: Apply IHC + MLH1 methylation + germline testing workflow.

5. POLE-exo hypermutator labeled MSI - Trigger: Tumor with 200 mut/Mb POLE-exo signature labeled MSI-H. - Mechanism: Pure POLE-exo causes hypermutator WITHOUT MSI (different repair mechanism); apparent MSI-H call may be a false positive in high-mutation context. - Symptom: Misclassification; ICI eligibility still positive but for different mechanism. - Fix: Run Sigprofiler signatures (SBS10a/10b vs SBS6/15/26/44); confirm POLE-exo via SBS10 contribution.

6. ctDNA MSI without sufficient tumor fraction - Trigger: Run MSIsensor-ct on cfDNA with <1% tumor fraction. - Mechanism: Low ctDNA fraction produces noise-dominated unstable locus counts. - Symptom: False-negative or unstable MSI call. - Fix: Estimate tumor fraction first (ichorCNA); require >= 3% for reliable cfDNA MSI.

7. Universal screening missed - Trigger: CRC patient < 70 yr without IHC / MSI. - Mechanism: NCCN / ACG universal Lynch screening required; without it, Lynch syndrome undiagnosed. - Symptom: Family loses screening benefit. - Fix: Universal IHC + MSI on all CRC < 70; institute reflex testing.

8. MSI-L treated as actionable - Trigger: Report MSI-L (intermediate) as ICI-eligible. - Mechanism: FDA approval specifies MSI-H; MSI-L = MSS clinically. - Symptom: ICI given on insufficient indication; reimbursement issues. - Fix: Apply MSI-H threshold strictly per FDA; MSI-L = MSS.

Reconciliation: When Sources Disagree

Pattern Likely cause Action
PCR Bethesda MSI-H vs NGS MSS Bethesda panel uses 5 loci only; less sensitive Trust NGS with >=50 informative loci
NGS MSI-H vs IHC retained Subtle MMR loss; MSH6-only subtype; or POLE-exo Confirm with germline + POLE-exo signature analysis
Paired-normal MSI-H + tumor-only MSS Sample swap or low tumor purity in tumor-only Re-validate; check purity (>=20% required)
MSIsensor-pro vs MSIsensor (paired) Different baseline thresholds Apply panel-specific calibration
MSI-H suspected but tools differ Borderline mutational burden Use signature analysis (SBS6/15/26/44) as orthogonal evidence
ctDNA MSI vs tissue MSI Tumor fraction low Trust tissue; estimate ctDNA fraction

Quantitative Thresholds and Conventions

Threshold Convention Source
Bethesda MSI-H >= 2/5 unstable Boland 1998
NGS MSI-H cutoff 10-30% unstable loci (panel-specific) Various
MANTIS MSI-H threshold Step-wise difference > 0.4 Kautto 2017
MSIsensor MSI-H threshold >= 20% by FoCR Friends of Cancer Research
Minimum informative loci >= 50 NGS loci Panel-design convention
ctDNA tumor fraction minimum >= 3% for reliable cfDNA MSI (depth-dependent operational floor; MSIsensor-ct reports 0.05% LOD only at >= 3000x) Operational convention
Tumor purity minimum >= 20% Standard
FDA approval MSI-H or dMMR pan-tumor (2017) KEYNOTE-016/164/158
First-line MSI-H CRC KEYNOTE-177 (2020) --
MSI-H -> TMB-H rate ~83% Chalmers 2017
TMB-H -> MSI-H rate ~16% Chalmers 2017
Sporadic MSI-H mechanism ~50% MLH1 hypermethylation Various
Universal screening cutoff CRC <= 70 yr NCCN / ACG

Common Errors

Symptom Cause Solution
MSI-H + IHC retained discordance Subtle loss; MSH6-only; or rare hypermutator Cross-check germline + signatures
Borderline MSI call Panel too small Use >= 50 informative loci
Tumor-only MSI low confidence Background subtraction needed Use MSIsensor-pro with cohort baseline
MSI-H + TMB-H reported additive Tautology per Sha 2020 MSI-H is primary; TMB-H not additive
POLE-exo labeled MMR-D Different mechanism; mutation count differs Run Sigprofiler; SBS10a/10b is POLE-exo
Sporadic MSI-H mis-labeled Lynch Need MLH1 methylation test Confirm MLH1 methylation + germline

Anticipated Reviewer Pushback

Pushback Standard response
"MSI-H + TMB-H both reported additive" Sha 2020 Cancer Discov: MSI-H is the primary biomarker; TMB-H is statistical correlate. We report MSI-H first; TMB-H reported but noted not additive.
"Why MSIsensor-pro instead of MSIsensor?" MSIsensor requires paired normal; MSIsensor-pro handles tumor-only via cohort baseline. Most commercial panels are tumor-only.
"MSI-PCR vs NGS discordant" Bethesda 5-locus panel is less sensitive; we use NGS >=50 informative loci for confirmation.
"Universal Lynch screening?" NCCN / ACG recommend reflex IHC + MSI on all CRC <= 70 yr; we implemented universal screening protocol.
"POLE-exo hypermutator with MSI-H?" Sigprofiler signature analysis distinguishes: SBS10a/10b = POLE-exo (typically MSI-stable); SBS6/15/26/44 = MMR-D. POLE+MMR concurrent produces ultra-hypermutator.
"MSI-L?" FDA approval specifies MSI-H; MSI-L = clinically MSS; we apply MSI-H threshold strictly.
"ctDNA MSI viability?" MSIsensor-ct works if tumor fraction >= 3%; we estimate via ichorCNA; below threshold falls back to tissue.

References