Earlier this month, Prize4Life announced the winners of the DREAM-Phil Bowen ALS Prediction Prize4Life Challenge, also known as the ALS Prediction Prize. Lilly Fang (and her partner Lester Mackey) is a member of one of the two first place winning teams. We recently posted a Seeker Spotlight featuring Prize4Life’s Neta Zach which dives into the background of the Challenge and final results. In this post, we’re happy to introduce Lilly Fang and learn about her experience with the Challenge.
I am a recent graduate of Stanford Law School and a new associate at Latham & Watkins in Silicon Valley. Prior to attending law school, I worked at an economic consulting firm in New York and received my bachelor’s degree in mechanical engineering from Princeton.
I first heard of InnoCentive through my boyfriend, Lester Mackey, who is now a postdoc in the statistics department at Stanford. Because I was slated to start work at the end of October, I was looking at a good three months of free time after my bar exam in July. Lester, who had been involved in this kind of competition before, suggested that we work on a machine learning Challenge together during this time.
Among the Challenges we considered, the DREAM-Phil Bowen ALS Prediction Prize4Life Challenge stood out as one that seemed particularly meaningful and well-timed. We were excited by the potential of making a real contribution to the study of ALS (more commonly known as Lou Gehrig’s disease) through this Challenge. ALS causes progressive loss of motor function in patients and typically leads to death in about three years. Currently, there is no cure, and the causes are not well understood.
The task set out by the Challenge was to take patient data from the first three months of a clinical trial and use it to predict the rate of progression of the disease over the following nine months. One of the reasons that this Challenge had so much potential for new findings was that the data provided was a subset of the new Pooled Resource Open-Access ALS Clinical Trials (PRO-ACT) database, which is the largest ALS clinical trials database ever created.
One of our biggest challenges was the lack of information about what aspects of the available data would likely be predictive, as the academic literature had thus far only identified a few relevant variables. This lack of information, combined with our lack of medical expertise, led us to extract a broad spectrum of features from the data and to choose an algorithm that would be robust to irrelevant features. In the end, we hope to have identified some unexpected but useful features for further clinical study.
Winning the prize gave us an opportunity to present our work at the 2012 RECOMB conference on regulatory and systems genomics, where we also had the great pleasure of meeting the other Challenge winners as well as the Prize4Life and DREAM organizers. Undoubtedly, one of the best experiences we had was getting to talk with people so committed to making progress on ALS and excited about the possibilities of further collaboration. Working on an InnoCentive Challenge was a wonderfully rewarding experience. We were able to make a contribution to a problem that we never would have normally encountered, and in the process, raise our awareness of a disease and meet people dedicated to advancing its solution.