00:00hi I'm Vijay Pandey journal' partner a
00:0216z this episode of these 16 podcasts is
00:05on the genetics of drug response this
00:07was recorded as part of our a six and
00:09Zita normal summit it gives a great deep
00:12dive by Stanford professor of
00:13bioengineering genetics and medicine
00:15Russ Altman and his work especially
00:17around building farmer GK be a professor
00:19Altman gives a window into how data
00:21science and buyer from addicts will
00:22change the future of drug discovery and
00:24drug response in this case it turns out
00:26that you know when we think about
00:28inheritance we might think about
00:29inheriting your grandmother's eyes or
00:32your grandfather's ears but it turns out
00:33there's a lot more to Jack's than just
00:35that how you respond to drugs will be
00:37similar to your parents as well moreover
00:39we can now see how building a large Bank
00:41of human genetics variations will
00:43transform our understanding on
00:44optimizing of drug discovery in spots
00:46both for understanding side-effects and
00:48toxicity as well as making better drugs
00:51and going after new indications
00:52professor Altman gives a really
00:54fantastic overview of the space as well
00:56as a lot of his own individual
00:57contributions thanks very much and and I
00:59would be Russ Altman from Stanford
01:01University so let me just tell you
01:02quickly that I get most of my funding
01:03from the National Institutes of Health I
01:05also have collaborations with Pfizer and
01:07Genentech and Carius and second-home and
01:11I am a founder of Personalis which does
01:14immuno Oncology so at Stanford my
01:16laboratory focuses on informatics
01:18biomedical informatics and data science
01:20for understanding drug response and
01:22optimizing it and so I think the reason
01:24I'm talking to you to you today is maybe
01:27that maybe some of the things that we're
01:29doing is form the basis of the next
01:31generation of pharmaceutical discovery
01:34and development and I have some
01:35confidence in that because we're working
01:36with these companies that I mentioned
01:38who are thinking about how they might
01:39change their way of doing things so I
01:42got into this because we're building a
01:44database called pharmgkb
01:46pharmacogenomics knowledge base we've
01:48been doing this for 16 years and
01:50pharmgkb is a simple idea it's a
01:52database or really the knowledge base of
01:54how human genetic variation impacts drug
01:58response so you might not think about
02:01this but your response to drugs was
02:03inherited from mom and dad and Grandma
02:05and Grandpa just like your height and
02:06your hair color and your eye color but
02:09there's usually not a family lore about
02:13we all remember grandpa's big ears but
02:15we don't remember that grandpa had
02:16terrible side effects when he took Cody
02:18so we have to depend on the genome to
02:21make measurements and over the last 15
02:2316 years we've actually accumulated
02:25quite a large knowledge base of genetic
02:28variations in humans and how they can
02:30affect the response to drugs one quick
02:31example codeine codeine is in Thailand
02:34number three any of you have had a minor
02:36procedure may have gotten tylenol number
02:38three codeine is actually biologically
02:41inactive it goes through the liver this
02:43is where your liver is I should say I'm
02:44an internist as well a general
02:45practitioner codeine goes to your liver
02:47and there's an enzyme in your liver that
02:49transforms it into morphine and morphine
02:51is active very popular 7% of people of
02:54European descent don't have a version
02:57because of genetic differences in that
03:00enzyme they can't turn codeine into
03:02morphine so codeine is a placebo for
03:04them any pain relief they experience
03:06will be because they felt good about
03:08getting a prescription from the doctor
03:09and not because it was having any
03:10activity there are other people who turn
03:12coding into morphine super rapidly so
03:15they for example get 20 great minutes of
03:18morphine and then they have three hours
03:21until their next dose of continued pain
03:23so coding is a great example one of
03:26hundreds where knowing a little bit
03:28about your genetics will allow us in the
03:29future to implement this vision of an
03:31information system with knowledge of
03:33your genomes securely which can then
03:36help your your prescriber make decisions
03:38about the drugs that are most likely to
03:40work and a least likely to cause side
03:42effects but that's not what I came to
03:44talk about what I came to talk about is
03:45because we're building the pharmgkb
03:47which is the genetics of drug response
03:49it's really critical that we understand
03:51drug response and actually even for
03:53drugs that are on the market and have
03:55been used for many years our ability to
03:57really describe what they actually do is
04:00very limited and it's not because
04:01anybody is is doing anything wrong per
04:04se but the companies when they develop
04:06these drugs have a very focused view of
04:09what they're hoping the drug will do and
04:10they design their trials to prove that
04:12it does or doesn't do that that if the
04:14tribe if the drug is on the market it
04:16means the trial was relatively
04:17successful and so they'll say this drug
04:20does X it treats hypertension it treats
04:22diabetes and that is true but they
04:24because of their focus they sometimes
04:27winders to the other things that the
04:29drug might be doing which we might put
04:30into the bin of side-effects or other of
04:33idiosyncratic effects that are not
04:35understood but if I'm in charge of
04:37understanding the genetics of drug
04:38response I need a full picture of the
04:40drug despond so this project over the
04:42last 16 years has given me the excuse
04:44with my lab to really look at drug
04:46response at many levels to try to fully
04:48understand what drugs do and I'll try to
04:50argue with you that this is what the
04:52pharmaceutical companies of the future
04:53are going to have to do in order to
04:55optimize their production and use of
04:58drugs so when I talk about drug
04:59responses they happen one of the things
05:01that makes this I wouldn't say easy but
05:03one of the fortunate situations is
05:05because it's a biological phenomenon
05:07drug response can be characterized at
05:09multiple levels I can talk about the
05:11molecular level how does this small
05:13molecule drug interact with its target
05:15physically it forms all kinds of
05:18chemical connections and if there's
05:20changes in this protein because of
05:22differences and genetics it might change
05:24how tightly it binds and other molecular
05:26properties so we have a big interest in
05:28looking at the low level molecular
05:31interactions to get the full set for
05:33example of molecules that might interact
05:35with our drug even some of the molecules
05:38or targets that were not anticipated by
05:41the people who develop the drug the
05:42second level we can think of is the
05:44cellular response whatever's happening
05:46at the molecular level it will lead to a
05:48sequence of signals that has the cell
05:50change its physiology the reason we're
05:51giving the drug is we want to shift the
05:53cells kind of if you think of it as a
05:56network you want to get it into a new
05:57basin of interactions that's more
06:00healthy than wherever it was before and
06:02so we're very interested in a fully
06:04understanding how a small molecule or
06:06large molecule drug changes the cellular
06:09milieu the expression of the genes which
06:11genes are turned on which genes are
06:13turned off how that cell is working but
06:15we're informatics people in data
06:16scientists so we're not limited to scale
06:18this is the one I should say this is the
06:21one advantage we have over experimental
06:22colleagues they're awesome but they tend
06:25to be I'm a cell person or I'm a
06:27molecule person we can go over all
06:29magnitudes of scale and just integrate
06:32the data and this is I think the
06:33important theme so the next level after
06:35cell is tissues and complete organisms
06:37like humans and so the electronic
06:40Eckerd and other and wearables which
06:42you're going to hear a lot about in in
06:44seven minutes and 50 seconds these are
06:47all unbelievably useful sources that we
06:49can use to characterize drug response
06:50fully and then we can get to the
06:52population level we can look at
06:53population level databases and say when
06:55we give a drug to a million people yes
06:58it does what we thought it would do
06:59based on the approval but it probably
07:01does lots of other things as well and we
07:03can mine public databases to figure out
07:06what's going on and so the themes in our
07:09lab everybody in the lab works at a
07:10different scale but the best projects
07:12are the ones that integrate these scales
07:14because a signal that you get at the
07:15molecular level may or may not be true
07:18there's noise in all data sets however
07:20if you're seeing a signal at the
07:21molecular level and at the electronic
07:24medical record level that gives you
07:26doodles more confidence that this might
07:28be a real signal and not just a weird
07:30artifact of the data and so this these
07:33levels of abstraction that we have in
07:36biology and therefore medicine are
07:38incredibly useful for rectifying the
07:41signals and this is what is not done
07:43typically again drug companies have been
07:45very successful they developed a lot of
07:47drugs and usually it's a fragmented look
07:50at the data where one unit will look at
07:51it from a molecular perspective and they
07:53of course they have of course they have
07:54mechanisms to try to integrate this but
07:57our argument would be you could do this
07:58very early so I just want to end by
07:59giving you some examples of some of the
08:01things we're doing to kind of make this
08:03real so three things that we'd like to
08:04do we want to fully understand what
08:06drugs do we want to understand drug
08:08interactions and we want to understand
08:11new uses for old drugs and I just want
08:13to tell you a couple of stories so
08:15understanding the full effects of drugs
08:17we have published a couple of papers
08:19where we looked at FDA databases of
08:21adverse events reported by patients and
08:23physicians and companies and we were
08:26able to replicate most of the side
08:28effects that were listed in the drug
08:29label you know the drug label that
08:30little piece of paper that seems to be
08:33like infinitely expandable and it's
08:35actually a puzzle and how many times you
08:37can fold a piece of paper it's the world
08:39record holder typically we were able to
08:42find replicate what was on the drug
08:44label but using the exact same methods
08:46we were able to find tens or hundreds of
08:49extra side effects per drug with very
08:52high confidence from
08:53looking at a combination of FDA records
08:55and electronic medical records that was
08:58great for us because now we have a much
08:59expanded view of what a drug actually
09:01does and drugs in that family we can
09:05tell if it's a class effect all the
09:06drugs in this family have the same set
09:08of side effects versus a drug specific
09:10effect which is critical for
09:12differentiating in a market and things
09:14like that so that's a little story about
09:16how we look at getting a better sense of
09:18all the side effects of drugs drug
09:20interactions are incredibly important
09:22the average person who's above 70 and
09:24who's on any medications is often on
09:26seven to ten medications and whereas all
09:29the drugs are approved based on their
09:31individual action there's typically not
09:33a careful look at what happens when you
09:36have pairs triplets quadruplets of drugs
09:39all potentially hitting the same
09:41pathways at the molecular cellular etc
09:43level so we've done some work looking at
09:46this and in one story that I'll just
09:48summarize very briefly we looked for
09:50drugs that might cause glucose in combat
09:53glucose increases diabetes if you will
09:55in combination where individually they
09:59did nothing so we had very strong signal
10:01that the drugs when taken alone had no
10:03effect on glucose in the blood but when
10:05people took them together we saw a huge
10:08bump and in fact in diabetics and even
10:10huger bump in the serum glucose this was
10:12not reported it wasn't on the drug
10:14labels at all and it was because we took
10:16data sets from multiple levels in fact
10:19in this case we looked at population
10:20data we looked at electronic medical
10:22record data and we looked at organism
10:24level data in in mice combined these all
10:27and we found a very strong signal with
10:30associated with the use of the two drugs
10:31together this is paroxetine paxil an
10:34antidepressant and pravastatin a
10:36cholesterol medication not associated
10:39typically with glucose changes but with
10:42a very clear signal we actually took
10:44that information and went to search logs
10:46in a collaboration with Microsoft we
10:49looked at what people who were on well
10:51we don't know if they were on these
10:52drugs we just looked at search logs and
10:53said how often do people type in these
10:57two drugs and words that might be
10:59associated with the symptoms of
11:00hyperglycemia or high glucose and we
11:03compared that to people who just typed
11:05in one drug and some words
11:06or the other drug and some words and in
11:09a paper that we published we showed a
11:10remarkable increase in the occurrence of
11:13words associated with hyperglycemia when
11:16they had also typed in the two drugs
11:18together so this opens the the obvious
11:21in retrospect opportunity of doing
11:24direct surveillance of people patients
11:27by looking at social media and so this
11:30was web searches but I have colleagues
11:32who are looking at Twitter feeds turns
11:34out people tweet their drug response I
11:36don't know why they tweet their drug
11:37response but they do and you can get
11:40tens of thousands of tweets from the
11:41from the Twitter firehose and you can
11:44start to put together a list of side
11:45effects it's a huge challenge the big
11:47issue there is taking the words that are
11:50used in texting and mapping them to
11:52medical concepts because as you could
11:54imagine with 140 characters there's a
11:56lot of abbreviations and there's a lot
11:58of slang there were also people by the
11:59way looking at Facebook and there are
12:01the patient portals where patients get
12:03together because they're part of a
12:04disease group to share experiences all
12:07of these also it's clear I think to most
12:09pharmaceuticals have to be sources of
12:11data both for understanding the actions
12:14of these drugs but also understanding
12:15the patient preferences about which
12:17symptoms of these diseases they're
12:18really most interested in getting and
12:20getting treated and then finally the
12:22third thing is getting new uses for old
12:25drugs this is a very exciting idea maybe
12:27many of you have heard about it's called
12:28repurposing if you're gonna repurpose
12:30her drug though all the information in
12:33the drug label that was focused on
12:34getting you approved for one indication
12:36is not going to be where you're going to
12:38get the insight about the new uses
12:41you're gonna get the insight about the
12:42new uses from the side-effects which
12:44I've already discussed from the
12:46interactions with other drugs those are
12:48the huge clues that tell you in addition
12:51to this one approved pathway and and
12:53effect there are these other parts of
12:55the biology that are being tickled if
12:57you will by the drug and there's an
12:59opportunity to chase that down and so in
13:01the setting of cancer we've published
13:02some papers about cancer drugs that are
13:05approved for cancer X but when we look
13:08at the genome and when we look at the
13:09binding patterns of that those molecules
13:11we have very strong predictions that
13:13they will also be useful in cancer why
13:16not not an approved use of the drug but
13:20pretty compelling evidence that it would
13:22at least be enough to start a trial and
13:24to evaluate if this might really work
13:25and in some cases might be enough for a
13:28physician to do an off-label use based
13:30on their judgment and their assessment
13:32that this is going to be safe and worth
13:33a try so so in the end I think I'm very
13:36optimistic that the discovery and
13:38optimization of drug use in the future
13:40is going to benefit from data science
13:43because of the integration of these
13:44streams I already had that theme where
13:46multiple sources of data when they can
13:49cut when they have a confluence and when
13:51they agree are incredibly powerful and I
13:54think this is the future of how we're
13:56going to think about drug discovery and
13:57how we're going to follow over time the
14:00actions of drugs as we exposed patients
14:02to them and figure out what works and
14:03what doesn't so thanks very much