Change your data into knowledge - Kenneth Conway - Millenium Predictive Medicine
Change your data into knowledge Kenneth J Conway President Millennium Predictive Medicine
By on Jan 25, 2001Change your data into knowledge
Kenneth J Conway
President
Millennium Predictive Medicine
I'sd like to thank eyeforpharma and certainly the sponsors of the event for having the opportunity to speak today. And I am glad that with the power of PowerPoint I was able to change quite a few of my slides last night. So, you'sre going to see a very different presentation than I though I was going to give. So, why don'st we get started.
I thought I'sd start off with something a little bit crazy with this slide of one when the velocity of change in the market is greater than the velocity of change in the business. The end is near. This is something that we constantly try to remind all of our employees about and we heard yesterday how knowledge is doubling at a rate of about every 3 years now. So, one of the questions we have from all of our employees, are you twice as smart 3 years from now than you were 3 years ago?'s which most people get very nervous, are they changing that rapidly. And I think the thing we like to think about this is, in this great time of change, and we think about velocity, the thing you need to think about is not only is it speed that it's changing, but the direction that the healthcare industry is going.
So a lot of people talk about, is the world going to change? We think it has changed already. The genome is identified. People now are practicing medicine, really going from clinical symptoms to molecular medicine. We'sve already gone from snail mail to the Internet and we had a lot of discussion yesterday about, are patients really patients anymore or are they consumers? And we can have a lot of debate about which is preferable or where we are. But we'sre certainly moving in that direction. And this is one of the new slides I thought about last night, based on some of the discussions that we had about consumers. And as I thought back to my days in Grammar School and High School, hopefully I got Meslow's Hierarchy correct here. I think I did. But in the lower end we always talked about the things that were most important were food, water, shelter and sometimes some schools may talk about clothing. And I always wondered, well, since we'sre so worried about healthcare, and healthcare isn'st anywhere on Meslow's needs, how come we don'st have our insurance cards to go into the grocery market or a card to actually get whatever type of housing we want?'s So, it's sort of interesting how people have assumed now that healthcare is probably the one underlying need of all of these and probably is driving society in a big way today. But when Meslow thought about this, he hadn'st had it on his list. So, why is this important to me, as we think about this? So, we think medicine is going towards this personalized medicine and eventually I'sll get into how we talk about data and why we organize it the way we do. But we think, in fact we know today, that diseases will be re-classified, based on biological mechanisms and molecular markers. We'sre doing a significant amount of research in the oncology area, in my business, and we already know there are at least 6 different types of breast cancer, based on molecular expression differences in the tumor types. We think that the drugs will be designed, based on the molecular basis for disease. We'sve heard a lot about that and certainly all of you from pharma know that. And I probably should change this. It won'st be individuals that will
receive medicine tailored to their profile, but certainly much smaller groups will receive that medicine tailored.
Now, we think about medicine this way in a very comprehensive effort. And if you think about someone who's born down here and this is their life span and to make this as simplistic as possible, hopefully someone who's born with normal health, they proceed up towards some type of disease, and eventually all of us end up dying. So, we have symptom severity here on the other side. Now, the bad news about medicine today is, it's all practiced above this clinical symptoms line. So, in their lives, some real opportunities. One is, can we actually define some predisposition for patients? We are not focused on that right now. The major reason for that is our whole practice is based on changing healthcare. And in fact, if you can just tell someone that they'sre predisposed but cannot tell them what to do about it, in fact you haven'st really created any value. Screening is the next area that we think is very important. In fact, many people have diseases today that they'sre not aware of because they haven'st shown up as what we today define as clinical symptoms. The question that we really have as we turn some of this data into knowledge is, does this clinical symptoms line come down? Does the line, in fact, if you had, let's say, an ovarian screening test and, in fact, if you had a certain expression profile, is that now a symptom as compared to a symptom is you have a large tumor mass. And then we have prognosis which is really where medicine is starting to practice today. You have a disease. Can we guess whether in fact that disease is an aggressive variety or not? Pharmacogenomics is just starting and a lot of people think about, and then obviously you have the treatment paradigm with the RX there, whether it's a therapy or whether it's some type of surgical intervention. And there's been some discussion about the pharmacogenomics, about how important that may be. We tend to think that at least these 2 items first, if they'sre not done properly, will not give you the right treatment with the right drug, because in fact we'sre treating people, not based on who they are as individuals, but we'sre treating them for the wrong disease. And that, in fact, if we could actually have the right treatment, this may be the whole killer application right here if you could actually prove to the patient that you are doing them some benefit and then drive their actual health status below that symptoms line again.
Probably the best example of this that I can describe today is in the age treatment where, if you could think about this as viral load measurement, where a patient had literally zero viral load, was exposed to the virus, had antibodies so you could actually screen it. If the patient was increasing their viral load dramatically, had some type of treatment, and right now they'sre doing genatyping and phenatyping of the virus, giving different treatments, and in fact if the viral load comes down, you in fact know you are helping the patient. In most diseases what we think we'sre going to see is that if we can in fact have the right data and have the right knowledge, we'sll be able to tell patients whether we'sre actually helping them or not, prior to any change in their clinical symptoms and I'sll give you some examples of that as we go on. So, the bottom line of why the benefits of this is, we'sre going to have early detection and treatment, we should have much more objective therapy and monitoring and eventually this drives you into some type of personalized medicine.
I added this slide last night just to give you our definition of pharmacogenomics because I think there are many definitions and really it's, what are the underlying biological differences that you and I have as human beings that cause us to react differently to the
same drug? And most people talk about DNA or SNPs, but in fact a lot of the differences are symatic changes which is why in fact you need to have a very comprehensive approach doing not only DNA analysis but RNA and Protein. And also, we, 3 years ago, when we started talking about pharmacogenomics, we talked about increasing ethicacy, and having more safe products. Well, for those of you that are here from Novartis, are there any of my colleagues here from Novartis? We did quite a bit of work on Closeapene and Alunsapene, looking, could we identify the differences in patients that would actually, and I guess it's about 1% of them that have the side effect of Neutropenia. And we, in fact, think we could pick those patients out. The problem is, and what we'sve done is we'sve given up on this idea of, can we in fact take compounds that are on the market or maybe compounds that fail from a safety standpoint? Because when you actually do the statistics, you cannot have enough of those patients that have that problem to give you enough statistical power to prove to the FDA or to your medical colleagues that in fact you have made the drug through this testing. Secondly, the reason people are taking the drug in the first place is for ethicacy, not because the safety is somewhat assumed and in fact, as we heard yesterday and will continue to see, the drugs are coming much more quickly. As you bring a new drug out, literally within months later, someone else has another compound that usually has the same ethicacy or maybe slightly better and usually doesn'st have the safety problems that the first drug on the market does. So, this might be interesting to work on but quite frankly there's a lot of difficulties in doing that and so we'sve avoided working on that with most of our partners so far.
So, why is pharmacogenomics so important? Well, we think it's going to increase the success rate in drug development. From having done it we can tell you that for the first programs that you use, they are going to take longer and it is going to be more expensive because you in fact have to do the research to define the markers that you'sll use in the pharmacogenomic studies. And we think that we will be able to improve compliance because, if in fact you can prove to patients that in fact this drug is more efficacious for you, it appears that they will stay on the regiment. And we know we can expand indications by actually taking drugs that might have 10%, 20%, 30% ethicacy rates in small type patient populations. If we can pick those out, hopefully we can get those up into a 70%, 80%, or 90% ethicacy.
Now, this is a classic answer for most people in marketing. They say, well, you'sre going to kill our market. You'sre going to subdivide the market. People are not going to take our drug.'s That may be true, but the reason that you'sre in marketing and you'sre really thinking about what's your sales impact, and the sales impact really is the combination of what your market share is, what your price is, how many indications you actually had to prove for that drug and then what compliance you had. So, it really comes down to, how much of the pharmacogenomics population did you pick? And that may be a smaller population. But in theory at least, your pharmacogenomics share should be much higher and your price should be much higher. And in fact, if you look at some of the areas of oncology where you can improve ethicacy, the different chemo therapeutic regiments sometimes have a 6 times factor in pricing, as they'sre more efficacious.
So, I'sm finally getting to some of the data on how in effect do we do our research and in fact some of the data that we'sre actually looking at. And we have a very comprehensive approach where we'sre looking at DNA of 2 different cohorts, looking for differences, actually testing those patients for different drug effects, looking at changes in RNA
levels, also doing proteomics, looking for the effects on protein and at the same time doing subtractive hybridization and transcriptional profiling, yielding literally billions of data points that now need to be massaged to actually tell you what you have. So, this is an example of one of the studies that we are doing right now on ovarian cancer. Along the one axis here we only have 10 markers. These are the ones, after we took the entire genome through subtractive hybridization and then used 75K rays and then protein profiling, we found 10 markers that looked like they would be very useful in identifying women with early stage ovarian cancer. And what in fact you can find is these are patients right here that do not have ovarian cancer. All these women here do. And so it's our thoughts that through a lot of this work, within the next 2 years, we'sre going to be able to actually identify women in stage one and stage two ovarian cancer. Now, one of the questions we had, I think in the Round Table yesterday was, why doesn'st somebody just buy a super server and do some of this work? And the reason is, it's a lot more difficult. You need to not only know what the markers are, you need to know the histology of the different tumors that you'sre looking at. You need to look, in fact, at whether it's a PS or MS positive, where the tissues came from. And all of this becomes very important when the people that actually understand the biology are given this data in this type of format, then they have to make obviously the human judgment. So it's not all just done on the computer. And then you end up with some fairly simple chart that says, well, these are the markers that look good and these are the people that would be sensitive to specific drugs.'s When you add onto that now doing the proteomics effort, you then take in data and what we have over here on the bottom is, same patient one that didn'st have the drug and one that had the estrogen drug and you can actually see the expression change. Our effort is actually to compare the DNA analysis with the RNA analysis and the protein analysis and try to get a single answer from that where we look for the DNA that expresses a specific RNA which in theory expresses a certain type of protein, and if in fact in 3 different random experiments, on the same patient, taken with samples at the same time, you can get a hit on all 3 of those, then the likelihood that that's going to be valid when you do your clinical trials becomes much higher.
We'sre doing the same type of thing with heart disease. You can see we do high throughput screening, we do lots of database mining, we literally update our database on a nightly basis, both with the research that's done at Millennium and also the research that's done outside. We do multiple types of expressions profiling and from this standpoint, again, the complexity of this has gotten quite high. Not only do you compare human DNA's with rat DNA's, because we know certain things, you'sre looking at multiple types of myocardium and then comparing that, which in fact gave us one gene that we just published on, A Novaless SRBO1 variant, which we then tested with 3 different populations of women and we found, and I think Dr Klaus Lindpaintner talked about this, this morning, that in fact when we'sre talking about genetic testing, it's really no different than some of the biological testing that we'sre doing today. In fact we think SRBO1 actually will identify these women that have high HDL.
The tools that are used at most pharmaceutical companies are similar with probably 2 exceptions. One, we'sve modified the tools to become very high throughput. And secondly, I'sm not aware of, let's say, many companies that have integrated all the data into a single database, where the scientist can literally look at all the experiments that were ever run on those samples, or all the knowledge that's known about either that tissue or those specific genes or that protein all at once. And that, in fact, gives our scientists a huge advantage and it gives our bioinformatics people a headache every
night on how, in fact, as you'sre generating billions of data points, do you continue to bring these together?
So, most of these tools we had to develop ourselves and the ones that I'sll show you here first are tools that you probably see in your own laboratories. All these are all integrated into our database and all can be seen at the desktop of everyone of the scientists. But integrating these now has become even more complex and this is an example, now we'sre doing 3-D analysis where we compare gene by gene and then do a probability density. So, these are all software programs written by our bioinformatics people that have literally taken this data and tried to take it from, certainly, let's say a 2-D dimension into a 3-D, to let the scientists look at this and then make better guesses where in fact we should be doing our research. Which really gets us back to what we started which is a whole personalized medicine mindset of thinking about, can you in fact have some preventive therapies? Do you select people based below the line from inclusion or exclusion? Can you pick people at a better point, select responders and do monitoring? Which gets you back to pre-disposition testing, your screening testing, prognosis, eventually pharmacogenomics, and in fact, as I already stated, most of this testing needs to be done here first, before this is going to be useful. And then hopefully give the patient the right drug and then have surrogate markers that actually say that in fact the drug is working. An example of that is we'sre working on a program in rheumatoid arthritis where you can actually tell, based on the drug, whether in fact the patient is having a response or not. So, instead of the patient waiting 6 months to a year for an x-ray to tell them they in fact are having a positive response, they can get a response literally in a week or two and we believe that's going to change certainly the compliance of whether in fact a lot of the patients stay on that drug or not.
Yesterday you saw a lot of different slides about drug development and it looked like a very linear process and I think this is the basic problem in that we'sre thinking about different stages or different silos of how we actually work a drug through and eventually get it to a launch. The way that we'sre trying to approach this, with our whole bioinformatics programs are thinking about the centre of the whole universe as the molecular markers. And in fact, thinking about drug discovery is a never ending game where, even if you have a drug on the market, what do you know about it that in fact can be useful in phase 4's based on either new validated genes that you identified, animal models that are going on in different experiments, and in fact, can you link, and we don'st have this yet but this is what we'sre trying to do, can you link all this knowledge that actually has literally all of your experiments now being important to every future experiment? And in fact if we think we can do that, and go from this linear process to this more circular process, we'sll solve this phase or the NDA problem of being able to deliver more than one or two NCE's a year.
So, what that drives us to is a new treatment paradigm where we think we'sre going to test patients before we in fact give them a drug. It'sll be done either with a biopsy or some type of body fluid. We'sll then know whether they'sre responders or not responders. And in fact then we'sll have this individualized group therapy which really gives us our view of the medical products of the future and if you think about Hertunu and her septum it's really what in fact we have today which is a therapeutic product, tied to a molecular test which gives us the personalized medicine.
And that's how we approach data management or thinking about it from a holistic stand point that at Millennium it's a very integrated approach where, literally, we have all of our scientists, all of our employees having access to all the data done entirely at the entire company. There are some risks with doing that in that you have lots of people seeing things and you have the concern of, are people going to take a lot of this data with them if they leave the company? That's a possibility. If people want to steal the data, we'sve seen from hackers, they can go in and get anything they want. We just find that giving our employees as much information and integrated approach as possible, is the way to actually change medicine and hopefully bring out drugs and better drugs faster. So, thank you very much.
Q: It sounds great, where's the catch?
A: It's hard to do. It looks a lot easier the way I showed it there than it actually is. We'sve now tried to do this for 7 years, and the reality is we don'st have a drug on the market yet, so that's one of the catches. We have drugs in the clinic and so we'sre now at the beginning stage of actually having genomic derived targets going into the clinic and the proof of the pudding is going to be over the next few years, that those actually go through the clinic better than other compounds, and in fact are they better compounds than in fact are on the market? The real answer is, the catch is, will we end up making more drugs than the standard way of doing it? We don'st know that yet.