Pan-Cancer MSK MiMSI MSI Cohort (2024)

Overview

Prospective secondary cohort of 5,037 MSK-IMPACT–sequenced tumors with paired mismatch repair (MMR) immunohistochemistry (IHC) used to validate the MiMSI deep multiple-instance learning MSI classifier. Spans 42 cancer types. The training/test microsatellite vectors and somatic alteration data are released as cBioPortal study pancan_mimsi_msk_2024 PMID:39746944.

Composition

  • 5,037 samples from 42 cancer types with paired MMR IHC; 4,195 MMR-proficient and 842 MMR-deficient (580 MLH1 loss, 166 MSH2 loss, 60 MSH6 loss, 36 PMS2 loss) PMID:39746944.
  • Cancer types most represented: COAD (n=2,448), UCEC (n=1,212), esophagogastric carcinoma (n=475), cancer of unknown primary (n=114), BLCA (n=73), small bowel cancer (n=60), PRAD (n=55) PMID:39746944.
  • Orthogonal MMR ground-truth: MMR IHC confirmed by board-certified pathologist; MSI-PCR for training/test cohort PMID:39746944.

Assays / panels (linked)

  • MSK-IMPACT (IMPACT468) — targeted hybridization-capture sequencing; typical mean coverage ~600×; microsatellite locus list: 1,755 loci covered by MSK-IMPACT (generated using msisensor scan v0.2) PMID:39746944.

Papers using this cohort

  • PMID:39746944 — Ziegler et al. 2025: MiMSI, a deep multiple-instance learning MSI classifier trained on 741 MSK-IMPACT cases and validated prospectively. MiMSI achieved sensitivity 0.895 and auROC 0.971 on held-out test samples, outperforming MSISensor (sensitivity 0.67, auROC 0.907), with the largest advantage in tumors with <30% purity (MiMSI 85.1% vs MSISensor 72.8%).

Notable findings derived from this cohort

  • MMR-deficient (MMR-D) tumors had significantly higher TMB than MMR-proficient (median 39 vs 5.27 mut/Mb, Mann-Whitney P<2.2×10⁻¹⁶); MSH2 loss had higher TMB than MLH1 loss (median 46.5 vs 37.7, P=0.0013) PMID:39746944.
  • Median indel/SNV ratio 0.18 in MMR-P vs 0.5 in MMR-D; MSH6 loss had the lowest (0.09) and MLH1 loss the highest (0.57) among MMR-D subgroups PMID:39746944.
  • MiMSI sensitivity in this prospective cohort: 91.6% (95% CI 89.5–93.4%) vs MSISensor 86.1% (83.3–88.6%); MiMSI correctly classified 226/247 (91%) MSISensor-indeterminate cases PMID:39746944.
  • MiMSI advantage over MSISensor was most marked in tumors with <30% purity (85.1% vs 72.8%, McNemar’s chi-squared P=8.244×10⁻⁷) PMID:39746944.
  • Per-cancer-type sensitivity advantage for MiMSI in small bowel cancer (100% vs 94.1%), PRAD (93.8% vs 85.7%), and UCEC (89.7% vs 79%) PMID:39746944.

Sources

  • cBioPortal study pancan_mimsi_msk_2024. Note: raw sequencing reads are not available (patient-consent restrictions); only precomputed microsatellite vectors and somatic alteration data are public. Ziegler et al. MiMSI: A deep learning classifier to detect microsatellite instability in next generation sequencing data. 2025. PMID:39746944.

This page was processed by crosslinker on 2026-04-30.