XtalPi and Pfizer collaborate on AI

Aim is to develop artificial intelligence-powered molecular modeling tech for drug discovery

Mel J. Yeates
Register for free to listen to this article
Listen with Speechify
CAMBRIDGE, Mass.—XtalPi Inc., a computation-driven pharmaceutical technology company, announced in early May a strategic research collaboration with Pfizer Inc. to develop a hybrid physics- and artificial intelligence (AI)-powered software platform for accurate molecular modeling of drug-like small molecules. This state-of-the-art platform will combine quantum mechanics and machine-learning algorithms with cloud computing architecture to improve the accuracy and chemical-space coverage of molecular mechanics modeling, and enable the prediction of pharmaceutical properties relevant for drug discovery and development.
As Dr. Shuhao Wen, XtalPi’s co-founder and chairman of the board, tells DDNews, “XtalPi and Pfizer have existing collaborations on crystal structure prediction—a core offering of XtalPi’s ID4 platform. The collaboration began in 2017. Our strong suit lies in combining physics insight with artificial intelligence and machine-learning algorithms to develop new methods and technologies that can surpass the existing efficiency and accuracy bottleneck. The quantum mechanics calculations will generate the target data for the AI and machine-learning models. Our use of AI is rooted in our deep understanding of the physics in molecular interactions, which will allow us to best harness the power of AI to improve on existing methods.”
“We are very excited for the collaboration … When we first developed our CSP platform, we weren’t sure how well it would perform in a [more] rigorous industry setting until we had the chance to run some blind tests with our first pharmaceutical clients,” continues Wen. “This collaboration will be a new experience to us, as we work directly and closely with Pfizer scientists to create a new molecular modeling software platform that is rooted in our technology expertise and Pfizer’s industry insights and knowledge, and will potentially have an impact beyond just the two companies. As a commercial platform that is partially open-source to the academic community, we also hope it will contribute to research advancements and innovations in related fields.”
This research collaboration aims to help XtalPi and Pfizer further advance their capabilities in computation-based rational drug design and solid-form selection. As part of the collaboration, a portion of the molecular mechanics parameters generated with public-domain compounds will be made available to the academic community in hopes of fostering continuous improvement and scientific innovations in related fields.
“Comparing to existing tools, our AI-powered, cloud-based mechanic modeling platform will cover a more comprehensive chemical space while improving the accuracy of its models, all with a special focus on drug-like molecules. Our goal is to develop the platform into one of the most capable and reliable modeling platforms which can support more advanced R&D applications the current options may not be suited for. Apart from modeling molecular interactions, the platform will also be able to predict several of its physical, chemical and pharmaceutical characteristics that are meaningful in drug R&D,” Wen notes.
“Crystal structure affects the safety and stability of a drug. Different crystal structures may have varying physical and chemical properties, such as stability and solubility, which translate to difference in shelf life, performance and bioavailability and are crucial to clinical trial success. A drug company must identify the crystal structure used in the drug product. And the crystal form chosen may affect the formulation and other later steps of drug development.
“With traditional research methods, the different crystal structures identified in the lab may or may not be the most stable forms, and there is no way to know for sure, which poses a considerable risk to drug development. Abbott Laboratories’ Ritonavir was a classic case in point: the crystal structure chosen for the drug product was a metastable form, and after half a year in the market, started to transform into a more stable form. As a result, the biological profile of the drug changed significantly, forcing Abbott to recall and withdraw the drug from the market temporarily, [and] reformulate before it was able to re-release it,” Wen points out.
According to Wen, “XtalPi’s quantum physics-based crystal structure prediction can accurately predict all the possible stable crystal forms and their drug-related characteristics, and provide crucial insights that help experiment scientists to conduct targeted lab work to physically identify the most meaningful crystal structures. It de-risks the molecule in pipeline, supports critical R&D decision-making and cuts down the crystal structure research cycle from a measure of several months, if not years, to a mere few months, even weeks. The platform is optimized to get prediction results within days.”
“In addition, the platform will have increased efficiency. We will also use AI algorithms to automate the workflow, choosing different fitting parameters for different types of tasks and molecules in order to get the best result. Its high-performance cloud computing architecture will further expedite our research efficiency by dozens of times, if not more,” Wen says. “We also intend to make it partially open source to the academia to foster further improvement and innovation in this field.”
When asked what kinds of data the researchers hope will come from this collaboration, he expects “Experimental data, modeling data, molecular structure data and chemical compound data. We will generate a lot of data on chemical structures, molecular energy and the physical, chemical and pharmaceutical properties of molecules. This, combined with the experimental data we will get from this collaboration, will help improve our algorithms and AI models.”
Wen mentions that the near-term plan is to create a platform which can help Pfizer increase efficiency and accuracy in small-molecule drug research. “In the long run, we hope that through similar collaborations, we can continue to develop and grow this platform to serve a wider range of research needs in the industry as well as the academia, and explore new possibilities that come with the enhanced capabilities. We hope it will become a powerful, reliable efficient tool for the research of new therapeutics.”

Mel J. Yeates

Published In:

Subscribe to Newsletter
Subscribe to our eNewsletters

Stay connected with all of the latest from Drug Discovery News.

March 2024 Issue Front Cover

Latest Issue  

• Volume 20 • Issue 2 • March 2024

March 2024

March 2024 Issue