Three students from Butler University and Hood College participated in a 10-week virtual training at the Frederick National Laboratory for Cancer Research, learning more about tools developed by the Accelerated Therapeutics for Opportunities in Medicine (ATOM).
FNL co-founded the ATOM consortium in 2017 as a public–private collaboration with universities, other national laboratories, and industry to accelerate the drug discovery process. ATOM has developed several open-source tools, including the ATOM Modeling Pipeline (AMPL)—an end-to-end pipeline for building and sharing predictive models.
FNL works with ATOM tools to streamline the lengthy drug discovery process using machine learning—a discipline of artificial intelligence that allows machines to learn from data—to create a workforce knowledgeable in this expertise.
Eric Stahlberg, Cancer Data Science Initiatives director and co-lead of ATOM, is a sponsor of the trainee program. “The ATOM trainee program continues to be a wonderful example to advance science now and into the future. While working on challenging problems of today and having an impact, trainees and students are gaining the critical experience that will enable the discoveries of tomorrow,” he said.
Accessibility, reproducibility and transportability
Hood College student J. Jedediah Smith partnered with FNL mentors Sean Black and Sara Jones, data scientists in the Cancer Research Technology Program (CRTP). With his background in Python (a high-level programming language) from his master’s in bioinformatics, Smith’s mentors tasked him with exploring the use of containers (isolated workspaces on a machine) for AMPL. When employed effectively, containers can improve accessibility, reproducibility, and transportability across different operating systems and computing platforms.
Smith’s aim was to eliminate dependencies with other modules and make AMPL compatible with cloud environments such as Google Cloud Platform and Microsoft Azure. Smith and his mentors built and tested the container together.
“[A container] simplifies and reduces certain errors that you can have based off of dependencies that would be on your machine,” Black said.
“I’d heard good things about FNL, and I’d heard about containers… so it was really great actually to use [the tools] and apply them,” Smith said.
At the end of the program, Smith accepted a job at a government agency in cyber security. He leveraged his newfound knowledge and skillset to apply for the position.
Improving machine learning
Justin Overhulse, data scientist in CRTP, along with Chloe Thangavelu, an intern in CRTP, mentored Butler University doctor of pharmacy students Renate Toldo and Sarah Norris. Together they sought to streamline the process to determine quantum mechanical (QM) properties of the gap between the highest energy occupied molecular orbital (HOMO) and the lowest energy unoccupied molecular orbital (LUMO), related to a compound’s stability and reactivity.
To accomplish this, Toldo and Norris compared three QM datasets with little compound overlap. With each dataset they prepared three models and cross-compared them to evaluate the accuracy of predicting the HOMO-LUMO gap to their reported values within each dataset.
The trainees’ work in cross comparing the datasets and QM model training provided insight into the team’s next steps. The ATOM team will use the QM models to increase the likelihood of synthesizability of the predicted compounds.
In addition, the program demonstrated how the trainees’ education and expertise could apply not only to the traditional pharmacy or in hospital settings but also to a research organization, the government, or data-centric private industries supporting data science efforts.
“This entire opportunity, it was a new side of their field that they could potentially go into,” said Overhulse.
The mentors organized a career development speaker series with professionals who had similar backgrounds that inspired the trainees for future careers.
“From the help on the project to the career advice and life skills, it’s been a very impactful summer for me, and I’ve been very thankful to have this position,” Toldo said.
Learning by doing
The FNL experience allowed students to work with real-world data, which is difficult to replicate in the classroom.
“A project is always unique so that we don’t have a school curriculum that repeats every semester,” said Naomi Ohashi, project manager of the trainee program. She highlighted the importance of the mentors’ work. “Every time we have new challenges to solve,” she said.
“The ATOM trainee program continues to be one of the truly exciting opportunities where FNL bridges the community while advancing science,” Stahlberg said. “The energy the students bring to their projects is amazing and truly leads to impressive impacts quickly. The students are having an impact today, and we will all benefit down the road as their experiences will lead to new discoveries in the future.”