ML Model for MSI Detection from Tumor NGS Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for determining microsatellite instability (MSI) status, such as MSI-PCR and IHC, are not always feasible for limited tissue samples and lack quantitative analysis, while NGS-based methods require matched normal samples or additional microsatellite loci for improved accuracy.
Innovation Solution
A trained machine learning model is used to predict MSI status from large-panel clinical targeted NGS data, accounting for at least six microsatellite loci, without the need for a matched normal sample, by assigning weights to features like peak width, peak height, and SSR type, enhancing sensitivity and specificity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Extent of automation
If NGS-based MSI detection methods such as MANTIS and MSIsensor are used, then automated analysis and quantitative statistics are provided, but matched-normal samples are required for evaluation
Solution Approach 1:
The patent extracts and removes the requirement for matched-normal samples from the MSI detection process. By training the machine learning model exclusively on tumor samples with known MSI status, the system eliminates the need for normal tissue pairing while maintaining automated analysis capabilities through quantitative statistics on microsatellite locus features.
Solution Approach 2:
The machine learning model performs self-training by learning to distinguish MSI-H and MSS statuses directly from tumor sample features without external reference to normal samples. The model uses internal feature relationships (peak patterns, read depths, locus characteristics) to autonomously determine MSI status, making the system self-sufficient.
2Measurement precision
If MSI-PCR assay is used for MSI status detection, then fragment analysis can be performed, but manual interpretation and qualitative analysis are required
Solution Approach 1:
The patent replaces the manual mechanical interpretation process with an automated machine learning system. Instead of human experts qualitatively analyzing fragment patterns, the ML model quantitatively processes sequencing data features (peak heights, widths, positions) to automatically classify MSI status, eliminating subjective variability and enabling high-throughput automated analysis.
3Reliability
If MMR-IHC is used for MSI status detection, then MMR protein expression can be tested, but loss of mutated proteins from missense mutations may not be detected
Solution Approach 1:
The patent segments the MSI detection approach from protein expression testing to direct DNA sequence analysis. By examining microsatellite locus features (peak patterns, read depths, allele frequencies) at the DNA level rather than protein level, the system can detect mutations including missense mutations that do not affect protein expression, thereby improving detection sensitivity while maintaining reliability.
4Measurement precision
If more microsatellite loci are added to improve NGS-based MSI testing accuracy, then detection robustness may improve, but assay complexity and cost increase
Solution Approach 1:
The patent applies partial action by selecting a focused subset of 11 specific microsatellite loci that provide sufficient discriminatory power for accurate MSI classification. Rather than analyzing all possible microsatellite regions, the ML model is trained on this optimized panel, achieving high accuracy while limiting assay complexity and cost to a manageable level.
Data Source
AI summary
A method and a system used to determine microsatellite instability (MSI) status utilizing Next-Generation Sequencing (NGS) and a machine learning model are disclosed. The present disclosure further provides a method and a system for identifying a treatment based on the computed MSI status data for the human subject.