a16z Podcast - a16z Podcast: The Genetics Of Drug Delivery

Episode Date: March 7, 2017

In this episode of the a16z Podcast introduced by Vijay Pande (based on a presentation at our summit event), Russ Altman, Stanford professor of bioengineering -- and former chairman of their Bioengineering Department -- takes us on a short but deep tour of the possibilities of genomics in drug discovery. Including how building a large bank of human genetic variations will change our understanding and optimization of drug response. Altman (who also hosts his own radio show, "The Future of Everything" on SiriusXM and Stanford radio) describes how in much the same way we inherit our grandmother's eyes, or our great grandfather's ears, we also inherit a response to certain drugs: whether they work or not, what side effects we'll experience, how we react to them. But it's not just genetics information that matters here; it's also molecular, cellular, tissue, and other data about the whole organism. By applying data science and bioinformatics on a more complete data "bank" like this, for the first time, we can see the whole range of actions and side effects -- as well as possible new uses -- that specific drugs will have on specific individuals. The views expressed here are those of the individual AH Capital Management, L.L.C. (“a16z”) personnel quoted and are not the views of a16z or its affiliates. Certain information contained in here has been obtained from third-party sources, including from portfolio companies of funds managed by a16z. While taken from sources believed to be reliable, a16z has not independently verified such information and makes no representations about the enduring accuracy of the information or its appropriateness for a given situation. This content is provided for informational purposes only, and should not be relied upon as legal, business, investment, or tax advice. You should consult your own advisers as to those matters. References to any securities or digital assets are for illustrative purposes only, and do not constitute an investment recommendation or offer to provide investment advisory services. Furthermore, this content is not directed at nor intended for use by any investors or prospective investors, and may not under any circumstances be relied upon when making a decision to invest in any fund managed by a16z. (An offering to invest in an a16z fund will be made only by the private placement memorandum, subscription agreement, and other relevant documentation of any such fund and should be read in their entirety.) Any investments or portfolio companies mentioned, referred to, or described are not representative of all investments in vehicles managed by a16z, and there can be no assurance that the investments will be profitable or that other investments made in the future will have similar characteristics or results. A list of investments made by funds managed by Andreessen Horowitz (excluding investments and certain publicly traded cryptocurrencies/ digital assets for which the issuer has not provided permission for a16z to disclose publicly) is available at https://a16z.com/investments/. Charts and graphs provided within are for informational purposes solely and should not be relied upon when making any investment decision. Past performance is not indicative of future results. The content speaks only as of the date indicated. Any projections, estimates, forecasts, targets, prospects, and/or opinions expressed in these materials are subject to change without notice and may differ or be contrary to opinions expressed by others. Please see https://a16z.com/disclosures for additional important information.

The content here is for informational purposes only, should not be taken as legal business, tax, or investment advice, or be used to evaluate any investment or security and is not directed at any investors or potential investors in any A16Z fund. For more details, please see A16Z.com slash disclosures. Hi, I'm VJ Ponday, general partner A16Z. This episode of the A16 podcast is on the genetics of drug response. This was recorded as part of our A6 and Z inaugural summit. It gives a great deep dive by Stanford Professor of Bioengineering, genetics, and medicine, Russ Altman, and his work, especially around building Farmer GKB. Professor Altman gives a window into how data science and bioinformatics will change the future of drug discovery and drug response. In this case, it turns out that, you know, when we think about inheritance, we might think about inheriting your grandmother's eyes or your grandfather's years, but it turns out

there's a lot more to genetics than just that. How you respond to drugs will be similar to your parents as well. Moreover, we can now see how building a large bank of human genetics variations will transform our understanding on optimizing of drug discovery response, both for understanding side effects and toxicity, as well as making better drugs and going after new indications. Professor Altman gives a really fantastic overview of the space as well as a lot of his own individual contributions. Thanks very much. And I would be Russ Altman from Stanford University. So let me just tell you quickly that I get most of my funding from the National Institutes of Health.

I also have collaborations with Pfizer and Genentech and Karyas and Second Om. And I am a founder of Personnelis, which does immunoancology. So at Stanford, my laboratory focuses on informatics, biomedical informatics and data science for understanding drug response and optimizing it. And so I think the reason I'm talking to you today is maybe, maybe, some of the things that we're doing is form the basis of the next generation of pharmaceutical discovery and development. And I have some confidence in that because we're working with these companies that I mentioned who are thinking about how they might change their way of doing things. So I got into this because we're building a database called FarmGKB, pharmacogenomics knowledge base. We've been doing

this for 16 years. And FarmGKB is a simple idea. It's a database or really a knowledge base of how human genetic variation impacts drug response. So you might not think about this, but your response to drugs was inherited from mom and dad and grandma and grandpa, just like your height and your hair color and your eye color. But there's usually not a family lore about drug response. We all remember grandpa's big ears, but we don't remember that grandpa had terrible side effects when he took coding. So we have to depend on the genome to make measurements, and over the last 15, 16 years, we've actually accumulated quite a large knowledge base of genetic variations in humans. and how they can affect the response to drugs. One quick example, codeine.

Coding is in Tylenol number three. Any of you've had a minor procedure may have gotten Tylenol number three. Codine is actually biologically inactive. It goes through the liver. This is where your liver is. I should say I'm an internist as well, a general practitioner.

Coding goes to your liver, and there's an enzyme in your liver that transforms it into morphine, and morphine is active. Very popular. 7% of people of European descent don't have a version because of genetic differences in that enzyme, they can't turn codeine into morphine.

So codeine is a placebo for them. Any pain relief they experience will be because they felt good about getting a prescription from the doctor and not because it was having any activity. There are other people who turn codeine into morphine super rapidly. So they, for example, get 20 great minutes of morphine, and then they have three hours until their next dose of continued pain. So codeine is a great example, one of hundreds, where knowing a little bit of, bit about your genetics will allow us in the future to implement this vision of an information system

with knowledge of your genome securely, which can then help your prescriber make decisions about the drugs that are most likely to work and the least likely to cause side effects. But that's not what I came to talk about. What I came to talk about is because we're building the farm GKB, which is the genetics of drug response, it's really critical that we understand drug response. And actually, even for drugs that are on the market and have been used for many years, our ability to really describe what they actually do is very limited. And it's not because anybody is doing anything wrong per se, but the companies, when they develop these drugs, have a very focused view of what they're hoping the drug will do, and they design their trials to prove that it does or doesn't do

that. If the drug is on the market, it means the trial was relatively successful. And so they'll say, this drug does X. It treats hypertension. It treats diabetes. And that is true. But because of their focus, they sometimes have blinders to the other things that the drug might be doing, which we might put under the bin of side effects or other idiosyncratic effects that are not understood. But if I'm in charge of understanding the genetics of drug response, I need a full picture of the drug response. So this project over the last 16 years has given me the excuse with my lab to really look at drug response at many levels to try to fully understand what drugs do. And I'll try to argue with you that this is what the pharmaceutical companies of the future

are going to have to do in order to optimize their production and use of drugs. So when I talk about drug responses, they happen. One of the things that makes this, I wouldn't say easy, but one of the fortunate situations is because it's a biological phenomenon, drug response can be characterized at multiple levels. I can talk about the molecular level. How does this small molecule drug interact with its target physically? It forms all kinds of chemical connections. And if there's changes in this protein because of differences in genetics,

It might change how tightly it binds and other molecular properties. So we have a big interest in looking at the low-level molecular interactions to get the full set, for example, of molecules that might interact with a drug, even some of the molecules or targets that were not anticipated by the people who developed the drug. The second level we can think of is the cellular response. Whatever's happening at the molecular level, it will lead to a sequence of signals that has the cell change its physiology. The reason we're giving the drug is we want to shift the cell.

kind of, if you think of it as a network, you want to get it into a new basin of interactions that's more healthy than wherever it was before. And so we're very interested in not fully understanding how a small molecule or large molecule drug changes the cellular milieu, the expression of the genes, which genes are turned on, which genes are turned off, how that cell is working. But we're informatics people and data scientists, so we're not limited to scale. This is the one, I should say, this is the one advantage we have over experimental colleagues. They're awesome, but they tend to be, I'm a cell person, or I'm a molecule person. We can go over all magnitudes of scale and just integrate the data, and this is, I think,

the important theme. So the next level, after cell, is tissues and complete organisms, like humans. And so the electronic medical record and other, and wearables, which you're going to hear a lot about in seven minutes and 50 seconds, these are all unbelievably useful sources that we can use to characterize drug response fully. And then we can get to the population level. We can look at population level databases and say when we give a drug to a million people, yes, it does what we thought it would do based on the approval, but it probably does lots of other things as well, and we can mine public databases to figure out what's going on. And so the themes in our lab, everybody in the lab works at a different scale, but the best

projects are the ones that integrate these scales, because a signal that you get at the molecular level may or may not be true. There's noise in all data sets. However, if you're seeing a signal at the molecular level and at the electronic medical record level, that gives you oodles more confidence that this might be a real signal and not just a weird artifact of the data. And so these levels of abstraction that we have in biology and therefore medicine are incredibly useful for rectifying the signals. And this is what is not done typically. Again, drug companies have been very successful. They've developed a lot of drugs. And usually it's a fragmented look at the data where one unit will look at it from a molecular perspective. And of course they have,

of course they have a mechanism to try to integrate this. But our argument would be you could do this very early. So I just want to end by giving you some examples of some of the things we're doing to kind of make this real. So three things that we'd like to do. We want to fully understand what drugs do. We want to understand drug interactions. And we want to understand new uses for old drugs. And I just want to tell you a couple of stories. So understanding the full effects of drugs. We have published a couple of papers where we looked at FDA databases of adverse events reported by patients and physicians and companies. And we were able to replicate most of the side effects that were listed in the drug label. You know the drug label, that little piece of paper

that seems to be like infinitely expandable and it's actually a puzzle and how many times you can fold a piece of paper. It's the world record holder typically. We were able to find, replicate what was on the drug label, but using the exact same methods, we were able to find tens or hundreds of extra side effects per drug with very high confidence from looking at a combination of FDA records and electronic medical records. That was great for us, because now we have a much expanded view of what a drug actually does, and drugs in that family, we can tell if it's a class effect, all the drugs in this family have the same set of side effects versus a drug-specific effect, which is critical for differentiating in the market and things like that.

So that's a little story about how we look at getting a better sense of all the side effects of drugs. Drug interactions are incredibly important. The average person who's above 70 and who's on any medications is often on seven to ten medications. And whereas all the drugs are approved based on their individual action, there's typically not a careful look at what happens when you have pairs, triplets, quadruplets of drugs, all potentially hitting the drugs. the same pathways at the molecular, cellular, et cetera, level. So we've done some work looking at this, and in one story that I'll just summarize very briefly, we looked for drugs that might cause glucose increases, diabetes, if you will,

in combination where individually they did nothing. So we had a very strong signal that the drugs, when taken alone, had no effect on glucose in the blood. But when people took them together, we saw a huge bump. And in fact, in diabetics, an even huger bump. in the serum glucose. This was not reported. It wasn't on the drug labels at all. And it was because we took data sets from multiple levels. In fact, in this case, we looked at population data, we looked at electronic medical record data, and we looked at organism level data in mice,

combined these all, and we found a very strong signal associated with the use of the two drugs together. This is peroxitine, Paxil, and antidepressant, and pravastatin, a cholesterol medication, not associated typically with glucose changes but with a very clear signal. We actually took that information and went to search logs. In a collaboration with Microsoft, we looked at what people who were on, well, we don't know if they were on these drugs. We just looked at search logs and said, how often do people type in these two drugs and words that might be associated with the symptoms of hyperglycemia or high glucose?

And we compared that to people who just typed in one drug and some words or the other drug and some words. And in a paper that we published, we showed a remarkable increase in the occurrence of words associated with hyperglycemia when they had also typed in the two drugs together. So this opens the obvious, in retrospect, opportunity of doing direct surveillance of people, patients, by looking at social media. And so this was web searches, but I have colleagues who are looking at Twitter feeds. Turns out people tweet their drug response. I don't know why they tweet their drug response, but they do, and you can get tens of thousands of tweets from the Twitter fire hose, and you can start to put together lists of side effects.

It's a huge challenge. The big issue there is taking the words that are used in texting and mapping them to medical concepts, because, as you could imagine, with 140 characters, there's a lot of abbreviations and there's a lot of slang. There are also people, by the way, looking at Facebook, and there are the patient portals where patients get together because they're part of a disease group to share experience. All of these also, it's clear, I think, to most pharmaceuticals have to be sources of data, both for understanding the actions of these drugs, but also understanding the patient preferences

about which symptoms of these diseases they're really most interested in getting treated. And then finally, the third thing is getting new uses for old drugs. This is a very exciting idea. Many of you have heard about it. It's called repurposing. If you're going to repurpose a drug, though, all the information in the drug label that was focused on getting you approved for one indication is not going to be where you're going to get the insight about the new uses. You're going to get the insight about the new uses from the side effects, which I've already

discussed, from the interactions with other drugs. Those are the huge clues that tell you, in addition to this one approved pathway and effect, there are these other parts of the biology that are being tickled, if you will, by the drug, and there's an opportunity to chase that down. And so in the setting of cancer, we published some papers about cancer drugs that are approved for cancer X, but when we look at the genome and when we look at the binding patterns of those molecules, we have very strong predictions that they will also be useful in cancer Y, not an approved use of the drug, but we can come up with pretty compelling evidence that it would at least be enough to start a trial and to evaluate if this might really work, and in some

cases might be enough for a physician to do an off-label use based on their judgment and their assessment that this is going to be safe and worth a try. So in the end, I think I'm very optimistic that the discovery and optimization of drug use in the future is going to benefit from data science because of the integration of these streams. I already had that theme where multiple sources of data when they have a confluence and when they agree are incredibly powerful. And I think this is the future of how we're going to think about drug discovery and how we're going to follow over time the actions of drugs as we expose patients to them and figure out what works and what doesn't. So thanks very much.

Thank you.