The overarching aim of the MechML project is to bridge the gap between machine learning methodology and challenging life sciences applications by exploiting biological knowledge to allow effective, interpretable and scalable learning.

In particular we will adapt and extend existing approaches from machine learning to computational biology, thereby delivering new methods that scale to pertinent needs, including i) enhancing the interpretability of highly parameterized multi-layer models (“deep learning”), ii) combining conventional machine learning methods with mechanistic models and prior knowledge, iii) developing the models to be robust to non-iid sample structure and different forms of measurement noise, iv) developing portable models that can be reused to facilitate the integration of evidence across experiments, labs and different technologies, v) developing software solutions for scalable and parallelizable inferences on large datasets.