In the last few years, the use of Deep Learning has proved to be very effective in many fields including Bioinformatics and, especially, the study of tumor DNA sequencing data, which allows to measure somatic mutations that accumulate the lifetime of an individual

and are fundamental drivers of cancer. With the final aim of understanding and modeling tumor evolution, in general it is possible to exploit Deep Learning techniques

to extract rich and useful information from datasets of somatic mutations from several cancer patients, which can then be exploited for developing targeted drugs and

therapies for patients. Recent works exploit Deep Reinforcement Learning (DRL) to generate phylogenies belonging to a given cancer cohort. Other works attempt to use DRL to study cancer clones and mutations, opening new scenarios for future improvements.

Graph Neural Networks (GNNs) represent another powerful Deep Learning technique that can be applied to tumor evolution. Indeed, the input data usually consists of trees or, more in general graphs. Therefore, the adoption of GNNs can be of crucial interests, for example to extract rich representations for clones or mutations to be then employed for several prediction tasks. Moreover, this kind of networks can also be exploited to access pathways by providing features that also consider the topological relations in the input graphs.

In this project we will develop advance computational methods based on Deep and Reinforcement Learning for modeling and predicting tumor evolution. We will also use

GNNs to extract relevant features from the input data, so to improve the state of the art for complex prediction tasks.