Machine learning methods for complex molecular structural data: From self-assembly and folding to macromolecular assemblies.

Structural data underpins the understanding of fundamental biological functions, but is often highly complex, based on the size of bioactive molecules. Simulations can provide some insight, but additionally generate a substantial quantity of temporal data. To tackle these issues, transformative approaches are required. Machine-learning allows us to reduce data sets to key features to provide new insights on structural relationships. We will develop new methodologies to tackle areas such as the folding of mycolic acids, found in tuberculosis and related pathogens, and the interactions of proteins for biological regulation, that provide enhanced ways of rapidly understanding and ascribing relevant functional information to these important systems.