Abstract Genomic copy number evolution in cancer refers to the acquisition or loss of genome segments over time owing to mutational processes in cancer that result in the loss of biological mechanisms for the maintenance of genome integrity in cells. While genome sequencing can allow us to detect these copy number alterations in cancer cells, it does not directly inform us what the sequence of evolutionary events was that led to the present state of the cancer. The goal of our work is to infer this sequence of events. Specifically, we have developed a novel, first-of-its-kind reinforcement learning based approach for inferring evolutionary trajectories from genomic copy number profiles which we call RLevolution. We show that RLevolution is able to deconvolve the sequence of complex events that may occur during cancer evolution, demonstrated on a combination of simulated and real-world cancer datasets.
Feng et al. (Wed,) studied this question.