SKM 2023 – scientific programme

Parts | Days | Selection | Search | Updates | Downloads | Help

CPP: Fachverband Chemische Physik und Polymerphysik

CPP 35: Poster Session II

CPP 35.38: Poster

Wednesday, March 29, 2023, 11:00–13:00, P1

Rational Design of Novel Photoswitches with Generative Models — •Robert Strothmann, Christian Kunkel, Johannes Margraf, and Karsten Reuter — Fritz-Haber-Institut der MPG, Berlin, Germany

The sheer vastness of chemical spaces poses a daunting challenge to molecular discovery through high-throughput screening based on exhaustive sampling. Generative models (GMs) are an emerging machine learning (ML) approach that enables a more guided discovery. Implicitly learning chemical design rules from large reference data sets and suitable descriptors of a targeted functionality, GMs directly propose promising, yet diverse candidates.

Here we explore the use of GMs for the design of novel molecular photoswitches. In a first step, large general molecular databases are used to train a GM to generate chemically valid photoswitches. In a second step, the creation process needs to be conditioned towards performant switching capabilities. In the absence of sufficient corresponding experimental reference data, this conditioning is based on synthetic first-principles data. For that purpose computationally efficient descriptors are used in a multi-objective fashion to account for the desired key aspects of the switching process.