Robert Gentleman on His Goals for Drug Discovery at 23andMe

May 19, 2015 | Robert Gentleman’s goal is to “bring bioinformatics and computational drug discovery to complement the really strong genetics” already at 23andMe.

Gentleman recently left Genentech along with his boss Richard Scheller to start 23andMe’s new therapeutics group. Scheller joined 23andMe in mid-March and serves as chief science officer and head of therapeutics at the direct-to-consumer genomics company; Gentleman came on board in early April as vice president of computational biology.

With only weeks behind him thus far, Gentleman expects to spend his first six months building his team and exploring what drug discovery will look like at Silicon Valley’s personal genomics trailblazer. He spoke with Bio-IT World senior writer Aaron Krol about open science at a for-profit venture, the challenges of drug discovery, and what’s left to find via SNP chips. The conversation has been lightly edited for length and clarity.

You won the 2008 Benjamin Franklin Award for Open Science. As someone who has a reputation for supporting open science and open access to scientific data, do you see a harmony between that and the sort of direct-to-consumer approach that 23andMe has taken to genomic health? 

Robert Gentleman: I am an extremely strong proponent of reproducibility in science, and I think it’s essential that things be reproducible, but that doesn’t necessarily mean that you have open access. You can have reproducibility and we can set up systems here that would allow people to verify certain findings in real-time, without them ever having access to the genotype data.

Privacy around genotypes is really paramount, and I think individuals have a right and an expectation that their data will be private and it will be used for the purposes that they’ve agreed to participate. And I think that’s really essential, but I don’t think that’s orthogonal in any way to having reproducibility in science…

I think 23andMe, in terms of not being really just a drug company, has the opportunity to help people across a really broad spectrum of both behavioral changes and therapeutic interventions, and it’s a really interesting approach to the problem.

Do you see 23andMe therapeutics as being another example of the approach to drug discovery that previous companies, like Regeneron or Amgen or even Genentech, have taken? Or do you see something different here with the kind of relationship that 23andMe has with its data donors? 

I think a lot of [23andMe data donors] want to see their genome used for the improvement of health. And most of us, we have relatives that have had all sorts of diseases. And the motivation for doing some of this stuff is really not personal in the sense [of being] about you yourself, but it’s about your family, it’s about your friends, it’s about people around you…

When you start to think about how to run a clinical trial or a study on a subset of those patients, you can reach back out to them, and a lot of them are responsive to that. And I think that that will only get better as we can give them the motivation that, okay, here’s what we’re trying to do with the study that we’re doing. These are the genotypes that we’re interested in having in this study, whether it’s about diet or exercise or other interventions… We’ll find out as we go forward, but I think that will be an extremely motivating piece for the 23andMe community, and so I’m hoping that we’ll be able to get that kind of uptake. And that’s really not possible in the sort of standard drug company model right now.

The other thing I would like to say in that context is, you hit the nail on the head in the sense that most pharma—probably all pharma—realizes the importance of human genetics as a filter for finding targets and ultimately understanding the mechanisms of disease. Where 23andMe is sort of head and shoulders above everybody else is the size of the database, the richness of the phenotype, and the ability to re-contact. 

23andMe [has] over 900,000 [genotyped customers], a fair amount of family structure... Personally, I think this is going to be a very rich resource for understanding a number of human diseases, and the research team here—Joyce Tung and others—has really demonstrated that. They’ve got a lot of nice scientific papers coming out, and more in the pipeline. It has been proven to be a good database.

And then [with] deals with Genentech and Pfizer that have been announced, pharma also clearly sees the benefit here. And so for me, it was a pretty easy decision once I knew Richard was coming and that 23andMe were going to orient themselves in that direction. The opportunity [is] to work on arguably the biggest human database around and to apply all of the things that I learned. I was very lucky to have five years at Genentech. It’s an amazing company with really, really talented scientists, and that five years was pretty instructive coming from academia. And now I have that sort of toolbox and a group here. We’re going to have an awful lot of fun, and I think we’re going to discover some very interesting drugs.

Some really big correlations between disease and the kinds of SNPs that 23andMe measures have been found, things like APOE and BRCA. What kinds of genomic health discoveries do you feel might have evaded detection until now? 

Oh, lots... If you look, for example, at colon cancer, there have been a number of GWAS studies that have been done to identify variants associated with whether or not you get colon cancer, but… colon cancer isn’t one disease. Colon cancer is at least three and maybe ten diseases. If you can identify those subgroups and... do GWAS on those versus a reasonable control group, you’d find a different set of variants. And that’s true for all diseases. Diseases are very complicated.

Another thing that 23andMe has the ability to do that others don’t have so much is… to refine the control group… You can refine that and make sure that your control group really has nobody with the phenotype in the case group. Then you get better odds ratios, you get better data, et cetera.

We have a very challenging time separating out environment from genetics, but [we have] the ability to follow up with people where we think there might be a confounding factor. We hope that we’re going to be able to design trials that will very quickly allow us to figure out whether there is a big environmental [factor] confounding it. If we find a variant that associates with some disease and we think there might be an environmental confounder, we can ask people to behave differently, eat something different, exercise at some point in time, and by being careful, figure out whether it’s really an environment effect that we’re seeing or a genotype effect.

At Genentech your bioinformatics operation was, of course, embedded in a world-class team of biologists who could perform the kinds of in vitro tests and animal tests that were needed to carry your initial discoveries forward. How are you approaching starting your own discovery engine? 

Yeah, that’s a great question. I think part of what I’m hoping is that by not having that reliance, we may be able to better demonstrate how many things we can do computationally. Again, pharma—for good or bad—is sort of a big ship and it’s hard for it to turn. Trying to get people to think hard about how much stuff can you learn simply through computational methods, versus how much do you have to do in animal models, I think, is a good question, because neither of them are perfect models for humans. We have lots of history that demonstrates some of the imperfections of the animal models in pharma… They’re not going to move away from that very quickly. But given that we don’t even have that [at 23andMe], I’m hoping we’ll basically push a lot harder on the really great computational methods.

I also want to emphasize, this is just a fantastic time to be a computational biologist. You’ve got the TCGA, the GTEx, the HapMap. There are so many large projects out there that we won’t be using just internal 23andMe genotype data. We’re going to be leveraging the public and semi-public resources that are available to push things forward. So hopefully we’re going to find that there’s a fair amount that we can do without having to get to animals. And then when we do, I think we’re certainly open to having wet labs here and also to outsourcing a lot of the work.

You made the transition from a more basic research enterprise at Fred Hutchinson to the more demanding world of drug discovery at Genentech. Do you have any warnings to your new colleagues at 23andMe? 

Well, I think the thing is that you just have to be really patient. Drug discovery is extremely hard. It’s a long-term process and it carries with it very high risks. And what we saw over and over again at Genentech was things that looked very promising, there would be some issue that would come along later on. They can look great in mice, they can look great in the first trials, and then they look bad in the second trials. That’s just sort of the reality of drug discovery. Humans are very complicated, and getting things to work well in a lab is not the same as getting them to work well out in the wild where things are different.

I think here we’re hopefully going to be one step above that, so we’ll have the genetics stuff but we’ll actually also be able to work a little bit with people that are interacting with the therapeutics group in some of our trials. And so again, I’m hoping that that de-risks a certain amount of the drug discovery stuff, but hopefully everybody here realizes that [we’re] not going to be making drugs next year.