MSIFinder

MSIFinder classifies microsatellite instability (MSI) status from sequencing data using a random forest classifier to distinguish MSI-H and MSS tumors as a predictive biomarker for cancer immunotherapy.

Key Features:

• Algorithm: Uses a random forest classifier (RFC) for MSI classification.
• Implementation: Implemented as a Python package.
• Feature markers: Employs a set of 54 selected microsatellite feature markers.
• Training data: RFC model was built using 19 MSI-H and 25 MSS samples for feature selection and model training.
• Validation (retrospective): Validated on 21 MSI-H and 379 MSS samples with sensitivity 1.0, specificity 0.997, accuracy 0.998, positive predictive value 0.954, F1 score 0.977, and AUC 0.999.
• Validation (prospective): Prospective validation on 18 MSI-H and 122 MSS samples achieved sensitivity 1.0 and specificity 1.0.
• Robustness to sequencing depth: Maintains concordance of 0.993 at 100× sequencing depth.
• Robustness to panel size: Maintains concordance of 0.99 when the panel size is reduced to 0.5 million bases.
• Sequencing data compatibility: Applicable to genome sequencing and target panel sequencing data.

Scientific Applications:

• MSI detection in solid tumors: Classification of MSI status (MSI-H vs MSS) in cancers such as colorectal cancer and endometrial carcinoma.
• Predictive biomarker for immunotherapy: Determination of MSI status to inform immunotherapy efficacy in advanced or metastatic solid tumors.
• Research and clinical evaluation: Use in research and clinical contexts where MSI classification is required under variable sequencing depths and panel sizes.

Methodology:

Implemented in Python and based on a random forest classifier trained and feature-selected using 19 MSI-H and 25 MSS samples (54 markers), with validation on retrospective (21 MSI-H, 379 MSS) and prospective (18 MSI-H, 122 MSS) cohorts.