Cancer mutational signatures identification in clinical assays using neural embedding-based representations.

Area of Science:

• Genomics
• Cancer Research
• Bioinformatics

Background:

• Mutational signatures offer prognostic and therapeutic insights but are underutilized in targeted gene panels.
• Predicting dominant signatures from limited mutation data in clinical settings is challenging.

Purpose of the Study:

• To develop a model for predicting dominant mutational signatures using limited mutations in targeted gene panels.
• To delineate the landscape of dominant signatures across diverse cancer types and patient cohorts.

Main Methods:

• A mutational representation model was developed to learn and embed specific mutation signature connections.
• The model was applied to predict dominant signatures across over 60,000 tumors using gene panel data.

Main Results:

• Dominant signature predictions in gene panels hold clinical importance, with UV, tobacco, and APOBEC signatures correlating with better survival.
• Associations between specific genes/mutations and signatures were identified (e.g., SBS5 with TP53, APOBEC with FGFR3^S249C).
• The APOBEC signature emerged as a predictor of resistance to epidermal growth factor receptor-tyrosine kinase inhibitors (EGFR-TKIs).

Conclusions:

• The developed model facilitates the detection of mutational signatures in clinical assays with limited data.
• This approach has significant clinical implications for a large number of cancer patients, potentially guiding treatment decisions.
• Mutational signature analysis in targeted gene panels can provide independent prognostic information and predict therapeutic responses.