Commodore Alum brings DeepMind to AlphaFold

Posted by daviskd2 on Friday, November 5, 2021 in News.

Vanderbilt alum Dr. John Jumper took a unique path to becoming research lead for the groundbreaking protein structure prediction database, AlphaFold. Jumper is a research scientist at DeepMind and leads a team developing new computational methods to enable accurate predictions of protein behavior, most notably 3-D protein structures. With his scientific accomplishments to date, you might think that this was the plan all along, but his career path in science was not always clear cut.

After a physics and math double major at Vanderbilt, Jumper intended to become a pen-and-paper theoretical physicist. He received the prestigious Marshall Scholarship to study in the UK and was accepted to the University of Cambridge’s PhD program. He soon realized that working on computational methods to study quantum mechanics wasn’t a good fit, so he left the program with a MPhil degree and returned to the US. He then got a job with D.E. Shaw Research working on custom supercomputers for protein simulation. This experience was his gateway into biology.

After three years, he returned to his PhD, this time at the University of Chicago, in an experimental and computational lab. Jumper’s PhD work in statistical and machine learning methods for protein simulation then led to a job at DeepMind working on protein structure prediction.

His role as research lead is multifaceted, facilitating research discussions with his team and other teams at DeepMind in addition to doing research himself. “Lately, a lot of my time has been spent working on the papers, code and database release and talks related to AlphaFold.” Jumper said.

He finds research more fun in a team setting, working towards a common goal. He especially enjoys the fast pace of machine learning, stating, “At various times, new ideas and results were coming in so quickly that I felt we were re-evaluating our approach to the problem on a weekly basis.”

An intuition and understanding about how deep learning systems work is important for developing models, and the importance of using intuition to evaluate results and catch errors, was a key takeaway from Jumper’s education at Vanderbilt. His advice to current trainees is to find “good problems” to work on, noting, “A good problem requires the right mix of importance and tractability.” ~ Katherine Amidon Paulin