Using neural networks to learn the folding landscape of DNA

The human genome is composed of billions of base pairs coding for genetic information. DNA is scattered through our chromosomes that all get packed in the nucleus of a cell. Different conformations of DNA folding can be informative of cell type and lead to regulation of specific expression programs by modifying physical accessibility to transcription. The tight structure and its conformations are controlled by multiple DNA binding proteins. Advanced molecular techniques now give information on the spatial contact and interactions of sequences, and their binding proteins and factors distributions along these sequences. Recent work showed that neural networks can predict the structure given binding proteins presence/absence on sequence but the introduction of mutations in the sequence doesn’t give steady predictions yet. The goal of this project is to build another type of neural network, variational autoencoders to the DNA folding problem.

Faculty Supervisor:

Eldon Emberly

Student:

Partner:

Université Paul Sabatier

Discipline:

Life Sciences

Sector:

Education

University:

Simon Fraser University

Program: