Towards computational screening of (bio)materials

Professor Agur Sevink

Department of Chemistry, University of Leiden

CECAM-IRL Talk April 2, 2019, at 3 PM UCD Physics Room SCN 128, University College Dublin

Computational investigation of soft (bio)materials is becoming an integral part of modern chemical & biological research. In-silico characterization is sometimes even employed prior to direct experimental characterization, to obtain the insight that is needed to define a target subspace in the massive design space of all chemicals.

Naturally, most of these in-silico investigations focus on obtaining detailed information at the molecular level via well-established methods that operate with electronic and/or atomistic resolution. Moreover, non-expert access to these techniques is ensured by efficient open-source implementations. Alternative approaches that are simply based on interpolation, such asArtificial Intelligence and Machine Learning, are also readily available and can be useful in providing results that go beyond this molecular scale,provided that the training data sufficiently sample the target chemical subspace.An important drawback of AI/ML, however, is that the physics – providing a deeper insight – is often lost.

In this talk, I will discuss a class of alternative computational developments that have the potential of offering relevant physical insight in processes that substantially exceed the time and length scales associated with molecules.

After motivating their development via a number of challenges associated with modeling relevant phenomena in biology/soft materials, I will discuss their foundation and implementation as well as their limitations and opportunities, focusing on what makes them computationally efficient and difficult to use for the non-expert.In the final part, I will speculate how these new developments may beemployed for efficient computational screening.