Current STRATiGRAD PhD student Sam Cooper was on the winning team at this weekend’s Pistoia Alliance deep learning hackathon, which aimed to put together deep/machine learning researchers with scientists working in the pharmaceutical industry to promote collaborations between the fields.
At the event, a series of challenges and datasets were put forward by companies in the Pistoia Alliance. Teams were then tasked with developing innovative machine learning solutions to these challenges. Sam speaks below about his experience.
What problem were you trying to solve?
We worked on a challenge put forward by Janssen in which we had to predict the activity of a drug across 7 different biological assays given its activity in thousands of other assays. However, drugs were very rarely active and as such the data was very sparse. Such data represents a real challenge to machine learning algorithms.
What was your winning solution?
To solve this challenge we developed a neural network using functions specifically designed for very sparse data, which enabled us to predict the results of the assay more accurately than everyone else. This was helped by the fact we specially configured a cloud computing server provided by Microsoft Azure for the event to train the network using graphics cards, which can train neural networks thousands of times faster rate than the processors used on standard servers and desktop computers.
What do you plan to do with the prize money?
Have a nice relaxing weekend away that doesn’t involve programming!
Find out more about Sam’s work in a recent article giving a new insight on how cancer cells replicate following a publication in Nature Communications this month.
