Using machine learning to decipher the adaptive immune system
Machine learning has entered center stage in the biological sciences thanks to its ability to detect, recover and re-create complex signals in large-scale biological data in which noise abounds. One particularly interesting and challenging machine learning problem is to decipher the workings of the adaptive immune system - the part of the immune system that allows our body to detect and fight viruses, bacteria or cancer cells it has never before encountered (e.g. Covid-19). This recognition is driven by a small DNA sequence that is constructed randomly for each individual immune cell and determines what that particular immune cell will recognize. From a machine learning standpoint, the DNA of different immune cells can be considered as small text sequences, where the challenge is to discover the complex patterns that determines what any given cell will recognize. Learning such patterns is important for diagnosing disease, improving vaccines and for developing new treatments for auto-immune diseases or cancer.
The aim of this master project is to develop deep learning methodology that is able to discover complex patterns of immune recognition based on large datasets of DNA sequence of immune cells. A combination of convolutional layers and attention mechanisms seem promising for the problem. You could also consider to make use of techniques connected to model sparsity, causality and the incorporation of prior information.
No prior knowledge of biology or immunity is needed. A good grasp of machine learning is required, and so is a desire to work hard and learn much from the master thesis project.
More details about adaptive immunity, relevant machine learning approaches and the work of our research group is available here.