Modern artificial intelligence is playing an increasingly central role in the drug discovery pipeline. A lot of research is being carried out to make portions of the pipeline more automatable, and far more intelligent. We have developed a robotic platform to highlight the utility of modern AI. Our robot is being equipped with state-of-the art methods; so far we have developed algorithms for: 1. Hypothesis generation: We have shown that we can train a recurrent neural networks (RNNs) on a corpus of ~22,000,000 molecules to generate new ‘unseen’ molecules.i Most interestingly, then we can use a technique known as transfer learning to generate libraries that are focused against a specific biological target. 2. Advanced route planning: However, one of the challenges of generating molecules using the above RNNs is determining their synthesizability. We have developed deep learning models and trained them on essentially all organic reactions. These trained models can be used as a rapid filter to remove overly challenging synthetic targets. 3. High throughput analytics and characterization: We have used quantum chemistry to generate training data for deep neural networks in order to try and identify molecular structures from spectra. This can be used to rapidly (and cheaply) assess the outcome of a reaction.

Robogenic chemistry

256th National Meeting (2018)

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.