The exa-scale ready tool: LiGen™
Dompé SpA, through its dedicated Drug Discovery Platform, invested since 2010 in a proprietary software for computer aided drug design (CADD). The most relevant tool is the de novo structure based virtual screening software LiGen™ (Ligand Generator), co-designed in collaboration with the Italian super computer center CINECA. LiGen™, as a distinguish features, has been designed and developed to run on High Performance Computing (HPC) architectures. To maintain the performance primate beyond 2020, Dompé decided to take the opportunity of embed in LiGen™ the emerging and most innovative technologies and new programming paradigms, that are leading the transition to exa-scale computing era. In the ANTAREX framework, a LiGen™ implementation, ready for exascale HPC application, was obtained, reaching a 100X of performance increase due to the redesign of part of the code co-developed with the Politecnico di Milano, and a 100X in the scale of the simulation thanks to the CINECA new optimized implementation.
Tangible Chemical Space
Dompé has generated a huge virtual chemical space of hundreds of billions of compounds. This library was built starting from a database of millions of commercial available reagents combined using a set of robust synthetic reactions, in order to obtain a tangible chemical space, meaning that is really achievable in one reaction step. The reactions are encoded by means of smarts language that accounts of detailed information about the reagent substructures directly involved in the reaction and their chemical environment, allowing to have an accurate annotation of each reaction in terms of synthetic feasibility. Dompé has already generated 500 Billions of compounds and more important has developed a technology able to obtain trillions of compounds before the end of 2020. These numbers represent a huge improvement, of about 1,000-fold, respect to the ultra-large virtual docking library launched by NIH (National Institute of Mental Health) researchers. (link: https://www.sciencedaily.com/releases/2019/02/190206131924.htm) In the context of HPC accelerated drug design, having access to a huge virtual chemical space represents a great advantage in virtual screening applications to increase the finding of potentially therapeutic molecules, especially if they are easily synthesizable. A further advantage of starting from billions of virtual compounds is the possibility to obtain still large but very focused chemical spaces applying criteria defined by the specific project needs, like precise physico-chemical and ADMET properties, speeding up the next steps of drug development.
Dompé Ultra High Performance Virtual Screening Platform
The huge tangible virtual chemical space developed by Dompé can be exploited in exascale virtual screening applications, perfectly integrating with the Antarex enhanced LiGen tool. Thus, Dompé developed a Ultra High Performance Virtual Screening Platform (EXSCALATE) based on both LiGen, an exascale software able to screen billion of compounds in a very short time, and a library of trillion of compounds. This HPC architecture exploits the CINECA (and PRACE) super computing infrastructure. Dompé will makes available for the EU (and National) Healthcare System the exascale virtual screening platform providing the fastest response for virus infections and multidrug-resistant bacteria. Exascale computing is also needed to rethink the actual drug discovery and development process for address diseases for which the model one target-one drug is insufficient to account for both the disease complexity and the patient’s peculiarity. Key for the success of the structure-based drug design, performed with Exscalate, is the evaluation of a chemical space big enough to allow the identification of chemical structures having the best complementary pattern of interactions with the biological target under investigation along with other phys-chem characteristics as well as novelty and synthesis feasibility. Thanks to this HPC architecture, in December 2018, Dompé was able to run the first reported exascale-ready structure based virtual screening experiment: 1.2 billion molecules against 1 biological target using on more than 900K hardware threads (> 220K physical cores) on the MARCONI Tier-0 system at CINECA. The simulation requires only 3 hours. Soon, EXSCALATE will be able to count on the new Italian supercomputer, called LEONARDO, that will be hosted at CINECA. This impressive machine will be able to reach a power peak of 270 petaflops, which will make the screening and identification of new drugs process, even faster.