Leiden Early Drug Discovery & Development
Using AI to improve the Design-Make-Test cycle with Galapagos
Researchers at LED3 are working together with biopharmaceutical company Galapagos to develop software for use in early drug discovery (funded by NWO). This software is able to design molecules with several simultaneously optimized characteristics and will also take prediction reliability into consideration in order to better manage the 'Design-Make-Test' cycle.
Enhancing the 'Design-Make-Test' cycle with AI
In the drug discovery process, the 'Design-Make-Test' cycle is repeated until a compound emerges that meets all the requirements of a pre-clinical candidate. Several properties need to be optimized simultaneously, including activity, selectivity, ADME (absorption, distribution, metabolism, and excretion), toxicity, and more. This is an extensive process, and an AI approach that uses all the available data and is able to model all these different endpoints will be very beneficial in making this a much more rational and efficient optimization process.
Both the process of using all relevant data and the generation of new molecules to consider will benefit from the application of computational technologies. Automatic model building for all relevant endpoints, and use of these models to guide the generation of new molecules have been shown to improve both the speed and the quality of the process, leading to the identification of better molecules, faster. Better because the relevant chemical space is better explored; faster because fewer compounds need to be synthesized during optimization.
From exploration to exploitation of molecules
Approaches to generate new molecules will also greatly benefit from better assessment of model prediction uncertainty. Initially, one aims to expand the chemical space covered by the models and to optimize these models, in which case compounds with high prediction uncertainty will be pursued (“exploration” mode; Hit Finding and Hit to Lead phases).
Subsequently, to hone in on the most promising areas of chemical space, compounds with low prediction uncertainty will be pursued (“exploitation” mode; Lead Optimisation phase). Accurate estimation of prediction uncertainty is especially important in exploration mode of a multi-objective optimization process, where molecule generation is mainly driven by endpoints with larger prediction uncertainties.
DrugEx for better drug discovery
LED3 developed software DrugEx has the integrated ability to balance exploration and exploitation in the molecular design process and forms the ideal starting point for this extended development. The practical application of this software, both in academic and in industrial settings, will contribute towards its success and applicability. To test the software it will be applied both at Leiden University and Galapagos. The software will be made available in the public domain after completion.