Stage M2S3 (2020-2021)

Project summary
Biology is going through an incredible revolution: current experimental techniques allows us to generate GBs of data in each experiment. In addition, international consortiums gather to address comprehensively specific problems in biology. For instance, the Human Cell Atlas Project aims to map and characterize all cells types in the human body using single-cell genomic techniques as a basis for both understanding human health and diagnosing, monitoring, and treating disease. This goal represents new exciting challenges in Mathematics and Computer Science as novel approaches are required to identify patterns, extract valuable information and produce reliable predictions. Machine Learning techniques have been proven to be a very powerful tool for these tasks. In particular, variational autoencoders have been used to learn reduced dimensional latent spaces from genomic data (see Fig. 1). However, the black-box nature of the methods hinders the interpretability of the latent variables. In this project we aim to develop an interpretable variational autoencoder to model gene regulation from single-cell transcriptomic data (see Fig. 1). Briefly, imposing prior knowledge on gene interactions through the network structure of the decoder will allows us to interpret latent variables as activities of regulatory proteins. Thus, we will be able to infer the key regulators that are responsible for the specific transcriptome in each single cell. Furthermore, using optimal transport theory on the latent regulatory space we will be able to predict the minimum number of regulators that need to be modified to promote a transition from one particular cell type to another. Finally, these results may have important implications to help the development of cell therapies.

Required skills
Very good programming skills will be expected. Good background in mathematics will be a plus. Prior knowledge in biology is not required but certain degree of curiosity to learn new fields will be desirable.

Acquired expertise during the internship
Mastering machine learning and deep neuronal networks. Programming with TensorFlow and Pytorch. Basic learning on genomics and gene regulation. Critical thinking and training for oral and written research communication.

Contact
Nacho Molina (This email address is being protected from spambots. You need JavaScript enabled to view it.)