SPP 2363: Hervorhebung der molekularen Ähnlichkeit durch erklärbare künstliche Intelligenz

Grunddaten zu diesem Projekt

Beschreibung

The project is built on our neural fingerprints that can be used as structure- and activity-sensitive molecular representations for e.g. virtual screening. The final aim of this 2nd funding period is to extend our work regarding neural fingerprints, the analysis of molecular representations and explainable artificial intelligence (XAI) techniques and integrate these approaches into a software package that is usable in scientists’ day-to-day work. The final software tool will allow a medicinal chemist or drug hunter to train their own neural fingerprint based on an available dataset, analyse the performance of the trained neural fingerprint using our uniform benchmark framework, and apply it in cheminformatics applications like a fingerprint-based similarity search or clustering of an existing dataset. In addition, we will provide pretrained neural fingerprints for various domains including a generic bioactive neural fingerprint. As a further special feature, it will ultimately be possible to use XAI methods to highlight the similarity between two molecules indicated by the neural fingerprint also in the underlying molecules. This would give the medicinal chemist the opportunity to trace and further analyse the similarities.