AMPL opportunity: Consortium releases first product for faster drug discovery

The Accelerating Therapeutics for Opportunities in Medicine (ATOM) Consortium, an initiative that aims to expedite research and development for new medicines, is making headway. 

The group recently released its first product: the ATOM Modeling PipeLine (AMPL, pronounced “ample”), an open-source software package that uses data and machine learning to predict chemical compounds’ experimental properties and activities against disease targets. 

The ATOM preclinical drug discovery workflow.
The ATOM preclinical drug discovery workflow. The process starts with existing data that’s used to train machine learning models, while later steps integrate further computing, biomedical data, and laboratory science. (Originally published in Frontiers in Pharmacology; reproduced here under the CC BY 4.0 license

AMPL allows users to design, fine-tune, and deploy computational models fit to biological and chemical data for predictions on new compounds. It was extensively validated by the ATOM team and is now freely available for download on GitHub

“We envision that the ATOM Modeling PipeLine can be applied to any group that’s interested in pharmaceutical drug discovery projects and predicting chemical properties,” said Ben Madej, Ph.D., an ATOM team member and a data scientist at the Frederick National Laboratory (FNL). 

AMPL’s release is a milestone for ATOM, a precompetitive public–private partnership that brings together national laboratories, academic scientists, and the pharmaceutical industry. The consortium members are combining computing, biomedical data, and laboratory science to dramatically shorten the time it takes to develop new medicines. 

“AMPL is the first step toward the eventual ATOM public platform, and it is a key component in many of our ATOM projects,” Madej said. “AMPL model predictions enable new ATOM platform models.” 

The software offers a uniform baseline for devising and creating computational models that help scientists design medicines faster. AMPL is also free and readily available, which is important given the current lack of standardization in machine learning models, a class of computational models in which algorithms are trained to make predictions. 

“Everyone has their own tools and software that they’re using. This creates problems as it becomes difficult to build reproducible models with widely varied workflows,” Madej said. “One of ATOM’s goals is to make a reproducible, open pipeline that anyone can use for modeling.” 

Improving the process with computing

High-performance computing and machine learning using AMPL represent potentially robust ways to speed up the long, expensive, and iterative drug discovery process. 

More than 90% of new compounds developed through traditional methods fail during preclinical testing, and, on average, slightly more than half of the ones that make it to clinical trials also fail. These are costly and time-consuming setbacks. But the predictions gained from computing approaches like AMPL aim to eliminate suboptimal candidate compounds and identify the more promising ones—and to do so faster and cheaper than traditional scientific approaches. 

Even so, innovative approaches don’t happen in a vacuum. Building a package like AMPL and giving the public a framework for drug discovery require a combination of disciplines and data. 

This is where consortia like ATOM can make a difference, as posited in a recent Frontiers in Pharmacology publication by Izumi Hinkson, Ph.D., former ATOM and FNL employee; Madej; and Eric Stahlberg, Ph.D., ATOM co-lead and FNL’s director of Biomedical Informatics and Data Science. 

Within ATOM, the national laboratories combine their expertise in cancer, biomedical data science, high-performance computing, simulation, and machine learning. One national laboratory also houses the powerful computing systems that ATOM uses to develop and train its models and software algorithms. Jumpstarted with experience and initial chemical data provided by the pharmaceutical industry, ATOM now proceeds with a growing emphasis on open and publicly available data. Meanwhile, several medicinal chemistry laboratories in an academic setting perform assays and experiments tightly integrated with the machine learning approaches. 

“There’s quite a few people from different disciplines coming together to try and tackle some of these problems. It’s a really interesting place to work together,” Madej said. 

AMPL also exemplifies ATOM’s unique place among drug discovery consortia: ATOM is actively adding new partners, and certain data, models, and technologies will be made available to benefit the scientific community. 

“I think that’s the key aspect,” Madej said. “All along, there have been plans to get ATOM technologies out beyond ATOM and to the public.”