Reverse engineer spatial patterns in biology

Precise and robust patterns emerge in biology, especially during development. Dissecting the genetic interactions that lead to these patterns has been the corner stone in developmental biology. This is usually done by observing changes in the pattern when perturbing the system, e.g. by deletions and/or mutations of the genes. Mathematical modeling usually starts with the knowledge of the genetic network inferred from those kinds of experiments. Here we set out to try a different and hopefully complementary approach. We ask the question that given a pattern what are the possible interactions or regulation logic that can achieve the pattern. In this talk, I will present examples of this approach using deep learning neural networks. In particular, I will discuss our results and lessons we learnt in the embryogenesis of fruit flies.