Collaboration And Machine Learning Key For 23andMe’s Drug Discovery Efforts

January 31, 2019

January 31, 2019 | Companies in the direct-to-consumer (DTC) genetics testing market continue to extend their outreach into the drug discovery and health care space. Having one foot in the consumer world and another in the life sciences, they have a unique perspective on how patients want their data to be used and where developments and innovations in research are going.

23andMe is at the forefront of this collaborative effort, forging partnerships with pharma companies and research institutes to improve drug discovery, as well as using resources to capitalize on the growth of personalized medicine.

On behalf of Bio-IT World, Mana Chandhok spoke with Sarah Laskey, a scientist in the Health R&D department at 23andMe, and Olga Sazonova, a Product Scientist II at 23andMe, about the implementation of tools like AI and machine learning in drug discovery and what recent collaborations mean for personalized medicine as a whole.

Editor’s Note: Mana Chandhok, a Conference Producer at Cambridge Healthtech Institute, is planning a track dedicated to Bioinformatics for Big Data at the inaugural Bio-IT World Conference & Expo West, as part of the Molecular Medicine Tri-Conference, in San Francisco, March 10-15. Both Laskey and Sazonova will be speaking on the program. Their conversation has been edited for length and clarity.

Bio-IT World: In the summer of 2018, a collaboration was announced between 23andMe and GSK. What does this partnership entail? Will 23andMe customers be involved in the collaborative process?

Sarah Laskey and Olga Sazonova: 23andMe and GlaxoSmithKline (GSK) have established a multi-year collaboration expected to identify novel drug targets, tackle new subsets of disease, and enable rapid progression of clinical programs. A joint GSK-23andMe drug discovery team will use their combined resources to identify new therapeutic drug targets and pursue drug discovery and development for new medicines to address serious unmet medical needs.

As in all of our research collaborations, only data from customers who have consented to participate in research will be used. By leveraging this genetic and phenotypic information and combining it with GSK’s incredible expertise and capabilities in drug discovery, we believe we can more quickly make treating and curing diseases a reality.

Has there been any progress from the GSK-collaboration? Do you think this collaboration with GSK signifies anything for 23andMe and its projection moving forward?

We collaborate with some of the best and brightest talent in the world of genetics research, including non-profit foundations, academic institutions, and pharmaceutical companies. Like many of our research collaborations, our collaboration with GSK aims to discover novel drug targets driving disease progression and develop therapies for serious unmet medical needs based on those discoveries.

The GSK collaboration is still in early stages—most of the work so far has been laying a strong foundation for our teams to make progress on a number of exciting projects moving forward. But we are optimistic about this work because we know that drug targets with genetic validation have a significantly higher chance of ultimately demonstrating benefit for patients.

I understand that you are using machine learning and statistical genomics in your efforts at 23andme. Most companies are implementing some sort of machine learning or AI in their drug discovery and development efforts. How are you focusing your use of the technology?

We have used [machine learning] to predict individual disease risk based on genetics and other health data. We are especially excited about the doors this may open for drug discovery in the future. Personalized predictions like the ones we're creating may allow researchers to differentiate responders from non-responders for drugs that work better in specific patient populations. Disease risk stratification based on genetics could also inform clinical trial recruitment, allowing researchers to evaluate drug efficacy in the right patient subgroups. And we believe there is substantial promise in using genetics for better disease subtyping to capitalize on the promise of personalized treatment.

Where should machine learning be going in the next five years?

Research by 23andMe and others has shown the tremendous potential of combining genetic information with different kinds of data, including diet, sleep, exercise, medical records, and more.

But the most important advance we can achieve in the next five years in the fields of genetics and personalized medicine is to collect a broader, more diverse pool of training data. The overwhelming majority of genetic data in public and private databases comes from participants of European descent. We are working toward a future where personalized medicine is a reality accessible to anyone in the world, but for that to be possible, we will need to expand training datasets to fully represent worldwide diversity.

23andme was an early advocate to allow patients and customers to be a part of the solution to help improve health care. Do you think this will continue, and if so, how will this evolve in the future?

Yes! That’s been a key part of our mission from the beginning. Our customers want access to their own genetic health information. And they also want to participate in research—more than 80% of our customers consent to have their data used for research. By using human genetics combined with the participation of a highly engaged community of research participants, we believe we can accelerate the identification of novel therapies and new treatments to fight disease.