Build and improve image embedding models of cellular phenotypes

The over-arching goal of the project is to explore the use of several recently developed self-supervised image representation learning methods in an attempt to improve performance across several biologically relevant benchmarking tasks at Recursion. At recursion, deep-learning based models are used to generate feature embeddings for our imaging data and these embeddings to generate downstream biological insights for drug programs. A large proportion of our in-house data is unlabeled, and we’re exploring the use of novel self-supervised representation learning models to improve our existing image embedding models. The end objective is to improve the ability to extract high-value insights from the data generated, typically called ‘maps of biology’, and lead to scientific breakthroughs by our disease biology scientists.

Faculty Supervisor:

Rahul Krishnan

Student:

Partner:

Recursion Canada Inc.

Discipline:

Computer science

Sector:

Professional, scientific and technical services

University:

University of Toronto

Program: