Mutational signature refitting on sparse pan-cancer data

Area of Science:

• Genomics
• Computational Biology
• Cancer Research

Background:

• Cancer genomes contain characteristic marks called signatures, resulting from mutational processes.
• Signature exposures provide crucial information for patient stratification and predicting drug response.
• Accurate deciphering of signature exposures is of growing interest, with previous methods being unsupervised.

Purpose of the Study:

• To develop a superior supervised approach for refitting cancer mutation signatures.
• To improve the accuracy of determining signature exposures from genomic data.
• To enhance patient stratification and drug response prediction in cancer.

Main Methods:

• Introduced SuRe, a supervised learning method utilizing a neural network.
• Leveraged neural networks to capture correlations between signature exposures in real data.
• Applied SuRe to sparse mutation data from tumor-specific and pan-cancer datasets.

Main Results:

• SuRe demonstrates superior performance compared to existing unsupervised methods for signature refitting.
• Performance advantage increases as mutation data becomes sparser.
• Successfully predicted homologous recombination deficiency in breast cancer using sparse data.

Conclusions:

• SuRe offers a powerful tool for analyzing sparse mutation data in cancer genomics.
• The method shows significant potential for clinical applications, including disease prediction.
• SuRe outperforms standard methods in unsupervised patient stratification, applicable to large-scale sequencing data.