Key points are not available for this paper at this time.
2023 was the year of ChatGPT and artificial intelligence is becoming increasingly ubiquitous in all walks of life.In scientific research AI is now being applied to almost every aspect of chemical discovery and optimisation from the generation of new molecules, to predicting their synthetic routes and modelling complex pharmacology, not to mention the more mundane tasks such as writing up results (disclaimer, this report was written entirely by human intelligence!).AI in drug discovery was therefore a timely topic for this year's DMCCB Basel symposium, held at the University of Basel's BioZentrum on February 12 th .The meeting was attended by approximately 120 participants from across the academic and industrial sectors with specialisms spanning drug discovery, chemical biology, computational chemistry, and machine learning.At the early stage of the discovery process generative AI can be used as a tool to discover new chemical matter for screening against a target.A generative AI workflow requires a generative model that 'invents' molecules with a particular set of properties, coupled with a scoring algorithm that ranks the generated molecules based on relevant parameters.The results for the scoring algorithm then feed back into the generative model to inform the design of further compounds.Developing models that can generate structures that are novel but chemically and synthetically reasonable is an ongoing challenge.To this end Prof.Hongming Chen (Guangzhou Laboratory) presented his groups research on the 'Development of Novel Generative Models for Molecule Design'.Prof. Chen's group has developed 'Tree-Invent', 1 a molecular generative model capable of producing structures that adhere to specific topological constraints.The model's adaptability extends to various molecule design applications, including scaffold decoration, scaffold hopping, and linker generation, showcasing its utility in diverse design scenarios.Prof Chen also presented EC-Conf, a novel diffusion method for rapid generation of low energy conformers. 2Conformer generation is necessary for many computational methods such as pharmacophore modelling, virtual screening and QSAR modelling and often represents a bottleneck in high throughput computational workflows.EC-Conf demonstrated up to two orders of magnitude improvement in efficiency over traditional diffusion models.
Simon R. Williams (Wed,) studied this question.