Artificial Intelligence (AI)

Easy and accurate quantification of mutational signatures in cancer samples

Group Zimmer, Medo   Mutations in the genome are often not random but follow distinct patterns. These patterns, termed mutational signatures, are footprints of biological processes or past exposures to specific carcinogens. Mutational signatures can help to elucidate tumor evolution and have prognostic and therapeutic implications.

Mathematical estimation of signature activities is a high-dimensional statistical problem. Although many tools for signature analysis have been developed, a recent analysis showed that no tool performs best under all conditions.

We propose to identify the best-performing tool for a given task, cancer type, and number of mutations. This will be achieved by extensive prior testing of the existing tools on synthetic data under different conditions. We further propose to design a new tool that will explicitly include mutations that are not well explained by any known signature.

With the two proposed tools, we aim to optimally use costly sequencing data and produce informed decisions for patient treatments.