A number of people have passed along the recent press stories about Verge Genomics, a new company out of YCombinator that has just raised $32 million for neuroscience drug discovery. Now that, as literally anyone who’s ever done it can tell you, is a hard field of a hard field, and I wish Verge good fortune in getting something to work. There’s a horrifying amount of what we always call “unmet medical need” in that area, and anything that actually goes toward alleviating it is welcome. So again, good luck to them.
But Verge does not, it seems, believe that any luck will be necessary. They are perhaps the purest confluence yet of several trends in biopharma startups. First is the AI/machine learning field, which I’ve written about several times and most certainly will write about again. Second, unfortunately, is “What the stodgy old drug industry needs is some happenin’ young disruptors”. That one comes up as people look at how long it takes to develop a drug, how much money it costs, and how many blind alleys get explored, and decide that the time is ripe to be the Amazon of drug discovery (upending the yawn-sleepy world of retail shopping), the Craigslist that demolished newspaper classified ads, the Apple that triumphed with the iPhone (but definitely not the Apple that almost went under in the 1990s).
Note that all of these examples contain plenty of semiconductors and lines of code. That’s another related trend, which we might call “Silicon Valley To The Rescue”. I’ve written about that a number of times here as well, of course, as regards the Andy Grove Fallacy (biomedical advances should take place at the speed of computing hardware advances), which is a subset of a general confusion about the physical world bred by experience bringing new hardware and software to market. If you can create a new device with new functions and have it adopted by millions of people as part of their everyday lives, you have indeed accomplished something. But that accomplishment might give you an exaggerated idea of your overall ability to affect reality as well.
That’s not because fabricating new chip designs and writing good code are easy: no, those things are hard. But drug discovery and medical advances, sadly, are even harder. The track record so far for the Silicon Valley/drug discovery interface is mixed (followup on that last link). So when I see things like this from a startup like Verge, I wonder:
(Alice) Zhang was three months shy of her MD and PhD graduation from University of California-Los Angeles when she left school to start Verge Genomics in 2015 with Jason Chen, who she met during the program.
“I just became very frustrated with the drug discovery process,” she said. “It’s largely a guessing game where companies are essentially brute force screening millions of drugs just to stumble across a single new drug that works.”
At the time, Zhang also recognized the advancements in neuroscience, machine learning and genomics occurring all around her. Genome sequencing had become more and more affordable, and breakthroughs in understanding how function connects with genes opened a new field of possibilities for exploring disease and health. And there was an opening for an opportunity to (take the) guesswork out of drug discovery. The vision for Verge was to become the first pharmaceutical company that automated its drug discovery engine, helping to rapidly develop multiple lifesaving treatments in diseases like Alzheimer’s disease, ALS, and Parkinson’s disease where no cure exists today.
As you read the rest of the article (and others like it) you come statements about how Verge is using human data “from day one” (unlike those drug companies), and how they’re not just looking at one gene at a time. And then there’s this one “Instead of tediously screening millions of drugs, the algorithm will computationally predict drugs that work“.
Something about that sounds vaguely familiar. Hold on. . .here we are:
. . .(these computer techniques) make drug design more rational. How? By helping scientists learn what is necessary, on the molecular level, to cure the body, then enabling them to tailor-make a drug to do the job… This whole approach is helping us avoid the blind alleys before we even step into the lab… Pharmaceutical firms are familiar with those alleys. Out of every 8,000 compounds the companies screen for medicinal use, only one reaches the market. The computer should help lower those odds … This means that chemists will not be tied up for weeks, sometimes months, painstakingly assembling test drugs that a computer could show to have little chance of working. The potential saving to the pharmaceutical industry: millions of dollars and thousands of man-hours.
That’s the ticket! And that, as some readers will recognize, is from an infamous article from 1981. Soon that piece will have its fortieth anniversary; we should have some sort of celebration. And yes, I understand that one of the whole points of a company like Verge is that it’s not 1981 any more, and that we have both vast amounts of data, and the abilities to deal with vast amounts of data, that no one could have gotten their heads around back then. But what if those still aren’t addressing the rate-limiting steps?
Ash Jogalekar summed it up perfectly on Twitter yesterday, as shown at right. The problems with Alzheimer’s, ALS, and Parkinson’s drug discovery are not data handling problems. The important problems with drug discovery in general are not data handling problems, and unfortunately there are many people who would like to think that they are. Who would perhaps like to think that everything could be solved if we could just obtain and correlate enough data. But what we’re short of is insights, ideas, and understanding, and those come slowly, painfully, and expensively.
Even taking Verge’s approach on its own terms, I very much doubt that we – we humans – have the tools to connect functions with genes in quite the way that the publicity would indicate. Nor do I think that these diseases are necessarily best approached from a pure genomic direction. That’s a part of the puzzle, but it’s a mighty big puzzle. Nor do I think that there are algorithms yet that will take long lists of compounds, correlate them with genomics data, and predict the winners. The Verge folks may have been misquoted, or had their claims exaggerated along the way by the press coverage. And Verge can come and prove me wrong about that, and in a way I hope that they do. But as it stands, unfortunately I think that the only people who can wholeheartedly believe in this approach to drug discovery are people who have never done any drug discovery.
Well, the article says that they have six drugs in development, “closer to the clinical end”, for what that’s worth. Let’s see what happens, and how long it takes, and how much money it costs. That’s the great thing about this field; we occasionally settle questions. Does this compound work? Is this target valid? Do we have a useful drug? My own experience is that we can indeed get those answers, but that it can be a long and viciously expensive process. Over to you, Verge – come on down and take a crack at it, and good luck to you.