Speaker
Description
Microsatellite instability (MSI) describes the accumulation of length alterations in microsatellite loci caused by deficiencies in the DNA mismatch repair system. In clinical diagnostics, MSI status is commonly assessed using polymerase chain reaction (PCR) assays targeting a panel of microsatellite markers. This testing plays an important role in the molecular characterization of colorectal cancer and in identifying patients who may benefit from specific therapeutic strategies. The electropherogram signals represent fragment length distributions whose interpretation is typically based on predefined rules comparing tumor and reference samples. However, signal variability and borderline cases can make this evaluation challenging. We investigate the use of machine learning models for MSI classification based on PCR fragment analysis data, exploring embedding and feature engineering strategies to distinguish MSI-negative and MSI-positive cases by capturing characteristic patterns in the fragment profiles. In addition, the model is used to assess the number and contribution of biomarkers required for reliable prediction, indicating that accurate classification can be achieved with a reduced biomarker set. We further analyze the adaptability of the approach to other tumor types, such as endometrial carcinoma. Our results indicate that machine learning methods can reliably infer MSI status and may support automated interpretation and decision assistance in diagnostic workflows.
