MSIFinder: a python package for detecting MSI status using random forest classifier.

Area of Science:

• Genomics and Bioinformatics
• Cancer Research
• Computational Biology

Background:

• Microsatellite instability (MSI) is a key genomic alteration in various solid tumors, including colorectal and endometrial cancers.
• MSI status, classified as microsatellite instability-high (MSI-H) or microsatellite stable (MSS), is crucial for predicting immunotherapy response.
• Existing computational methods for MSI detection are sensitive to sequencing depth and panel size.

Purpose of the Study:

• To develop a robust and accurate computational tool for classifying MSI status.
• To create a machine learning-based approach that is less affected by variations in sequencing data.
• To provide a reliable tool for both scientific research and clinical applications.

Main Methods:

• Development of MSIFinder, a Python package utilizing a random forest classifier (RFC) for MSI classification.
• Training the RFC model with 19 MSI-H and 25 MSS samples, selecting 54 feature markers.
• Validation using a test set of 21 MSI-H and 379 MSS samples, and a prospective cohort of 18 MSI-H and 122 MSS samples.

Main Results:

• MSIFinder demonstrated high performance with an accuracy of 0.998 and an AUC of 0.999 on the test set.
• The tool achieved perfect sensitivity (1.0) and specificity (1.0) on the prospective cohort.
• MSIFinder showed minimal impact from low sequencing depth (0.993 concordance at 100×) and small panel size (0.99 concordance at 0.5 Mbp).

Conclusions:

• MSIFinder is a robust and effective tool for MSI classification.
• The tool provides reliable MSI detection suitable for scientific and clinical use.
• MSIFinder's performance is stable across varying sequencing depths and panel sizes.