The development of novel compounds containing multiple desired pharmacophores has provided a direct means to target several discrete drug targets simultaneously, thus mitigating the requirement for complex treatments administering several drugs in tandem in order to be effective. However, typically the resulting compounds are achieved by linearly fusing individual preselected pharmacophores together, rather than incorporating each into a desirable drug-like scaffold. This leads to compounds that can, by-design, artificially target multiple pathways simultaneously, yet may lack the drug-like qualities that enable such compounds to be readily absorbed and distributed to present a viable treatment strategy.<br/>With the recent surge in popularity of generative deep learning methods, a diverse range of different neural network architectures have now been presented in the literature that exhibit great potential towards achieving a goal of high-throughput, automated, <i>de novo</i> drug design. In particular, generative AI models are now capable of directly targeting desired subregions of chemical space and/or distinct protein targets, as well as biasing towards the inclusion of particular molecular fragments or learned scaffold designs. Ultimately, this provides an effective means for generating diverse sets of novel compounds, whilst preserving the drug-like nature of the output molecules.<br/>Here it will be demonstrated how such neural networks can be utilised as a computational approach to aid the design of novel compounds containing multiple desired pharmacophores, whilst also crucially maintaining predicted activity towards each of several independent desired targets. The results of these generative approaches will be compared against the most recent experimental benchmarks concerning the design of dual- and multi- pharmacophore inhibitors. It is suggested that such generative AI models may provide an efficient utility to advise multi-inhibitor design by optimising the selection of compatible pharmacophore pairings that will result in the most effective treatment response.

AI-driven design of dual-pharmacophore libraries

2019 Spring National Meeting

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