In Silico Drug Development
Abstract
This Large Storage application is co-submitted with the Large Compute project application for the period Jan 1 - Dec 31 2025, and is a continuation of Large Storage 2023/4-6.
Computational chemistry has an important role to fill in the development of new pharmaceuticals. With HPC clusters coupled to the latest developments in algorithms and software, we are able to screen vast libraries of compounds searching for new drug candidates, create in silico models of target proteins, and explore protein-protein interactions crucial for e.g. signaling pathways in cancer cells.
We focus our studies on several multi-protein complexes as targets for cancer therapy, in order to identify small molecule inhibitors able to block modes of action or signaling pathways. These involve the multimeric UPRosome with focus on the activated IRE1 receptor; a range of protein complexes of the Integrated Stress Response (ISR); and studies of several death inducing or death effector complexes such as the apoptosome, DR4/5, and various protein complexes of the death fold superfamily, as well as interactions of pro-survival complexes wuch as TRADD-TRAF2 and TRAF2-cIAP1/2 .
We follow well-established protocols, involving homology modelling (if needed), protein-protein docking calculations according to a recent ‘consensus protocol' developed in our group, followed by replica MD simulations to determine stabilities and key interactions and mechanistic studies using QM/MM-MD. Normally, the MD simulations carried out are 500-1000 ns each, and performed in triplicate, placing high demands for HPC resources. In the drug development projects, we perform systematic docking of large databases (up to 1bn compounds), and detailed BPMD and MD simulations of resulting complexes, followed by additional hit-to-lead optimizations. The size of the compound libraries we use in our research requires the use of massively parallel execution. We have recently extended our work into the area of Machine Learning for de novo drug discovery and protein interaction network determination, with very promising results. These are significantly expanded upon in the coming year.
The size and extent of the simulations, the amount of data processed in the screening campaigns, and the sizes of the training and benchmarking sets for the AI-based tools, justifies the resources applied for.
