BETA
This is a BETA experience. You may opt-out by clicking here

More From Forbes

Edit Story

Andreessen Horowitz Wants To Bring Biology To The Cloud

This article is more than 8 years old.

Can you create biological insight on a laptop? If you could, it might overturn a fundamental paradigm of drug discovery : That it takes a great scientist or team of scientists to find a clear path through the messy complexity of biology. In the conventional model, sometimes the scientist is at a university. Other times she is in a company. But always, always, there is a series of iterative interactions--scientist running experiments in lab, scientist struggling to interpret results, scientist designing new experiments, scientist analyzing new results--until biological insight arises. If it ever does.

Of course, many drug discovery advances over the past thirty years have been driven by technological innovation: Combinatorial chemistry, high-throughput screening, vastly improved imaging and prediction software and rapid and reproducible assays run in some cases by robots on groups of cells or even individual cells leading to large and hopefully meaningful datasets.

But none of these advances has replaced the “Aha” moment of insight that arises from a human being’s engagement with a biological phenomenon that is thorny or one that had not even been perceived to exist. I always expected--and still do expect--to find that kind of insight in labs, not on laptops.

Vijay Pande of Andreessen Horowitz (image courtesy Andreessen Horowitz)

But now a renowned Stanford professor-turned-Silicon Valley venture capitalist, Vijay Pande, has set his sights on this challenge. Pande, the architect of the award-winning Folding@Home project and himself an award-winner in computational biology, recently joined a top Palo-Alto-based venture fund, Andreessen Horowitz, which formed a new $200 million fund to invest in three areas of software:

  • Digital therapeutics, in which companies such as Andreessen Horowitz portfolio company Omada Health will develop behavioral or other interventions to complement or replace pills and injections;
  • Machine learning in medicine, which is already being applied in areas such as radiology and genomics; and
  • Cloud biology, in which software algorithms take in and manipulate lab results to enable drug discovery and other applications. This would fuel innovation via the laptop and the mobile phone.

Pande agreed to describe his interests in a session at the inaugural Digital Health Showcase (DHS), held in San Francisco on January 13. DHS is a daughter conference of Biotech Showcase and both conferences are co-organized by Demy Colton Life Sciences Advisors and EBD Group. In the third area, which Pande emphasized in a follow-up phone conversation is the farthest of the three from realization, Pande would like to “support and fund software companies in the bio space” in order to enhance and augment the work already being done by academic labs, biopharmaceutical companies and other large or small life sciences companies. In moving more of biology to the “cloud”--in both its data storage and computational senses-- Pande wants to enable sideways thinkers, upstarts and even non-biologists to do “more with less” by avoiding getting bogged down in the minutiae of running assays themselves.

In Pande’s model, a life sciences startup of the future might be:

  • Based on software, not hardware (or wetware)
  • Located on the West Coast, not the East Coast
  • Dedicated to unleashing an opportunity to separate even further the insight of biological work in drug discovery and other areas from the laboratory “grunt work”
  • Taking advantage of the “virtualization” of biology in the way that Amazon and other “cloud” providers have virtualized data centers and semiconductor “fabs” like Taiwan Semiconductor Manufacturing Company have virtualized the fabrication of new computer chip designs.

After explaining in more detail what “virtualizing” biology means, I will give some reasons for being cautious about Pande’s aims, then some reasons to believe his approach might eventually work. The “laptop and a credit card” approach is not what drug discovery scientists grew up with, for sure. Then again, those who are disrupted (my word, not Pande’s) are often the last to see it coming.

Video of the entire "fireside chat" is available here.

What does “clouding” biology even mean?

In our discussion, I asked Pande to explain what he meant by “clouding” biology, which he said could allow an investor to avoid the sunk cost of building out a lab. He was not referring to the simple storage of biological data in off-site servers. That has already happened. Instead, he was referring to both maximizing the efficient use of contract research organizations (CROs) as well as something more powerful: Capitalizing on the automated nature of many laboratory assays to hand a remote user full access to a lab. That function, though it has been available to big companies for some time now, has not typically been accessible by early startups. Some companies have sprung up in--where else?--northern California to offer services like this. They include Emerald Cloud Lab, Transcriptic and Mousera. When you run an experiment there, Pande said, “You literally write [software] code. It really feels more like computing.”

To take one example, Emerald Cloud Lab (ECL) is a “web-based wet lab” that conducts experiments at accessible price points in an automated lab exactly as a researcher specifies. The data are organized in the “cloud” so that they can be accessed from anywhere. According to a 2014 post on the site Nanalyze, Emerald’s “lab in the cloud,” funded by Bay Area luminaries (and Pay Pal founders) Peter Thiel and Max Levchin, among others, “is a 15,000 square foot space which contains over $3 million in lab equipment that can now be used by anyone with an Internet connection and a credit card.” The typical experiment costs $25 and the menu is being expanded to include forty experimental types now and an estimated one hundred different offerings by the end of 2016.

In an email exchange, ECL CEO DJ Kleinbaum explained the difference between traditional outsourcing and virtualization this way:

Outsourcing involves contracting with a service provider and giving them high level directions (i.e., "synthesize this molecule...", "determine the toxicity of this compound..."). On the other hand, virtualization is a means of providing remote access to equipment (such as servers, factory equipment, or in our case, scientific instrumentation) to a user and giving them the ability to specify all the exact, low level instructions on how to operate those resources.

How Emerald Cloud Lab “virtualizes” biology. Image courtesy Emerald Cloud Lab

Mousera is stealthier than ECL despite having raised $30 million from Founders Fund, Lux Capital and other investors. Piecing together various reports, it seems like Mousera aims to capture a vast amount of data from mouse experiments by attaching sensors to the mice and analyzing their movements using custom software and algorithms in an approach reminiscent of the much more established, still privately held Psychogenics.

For the record, Andreessen Horowitz is not an investor in any of these companies.

I pushed back a bit on Pande and said, sure, but there are plenty of biotechs that have CROs run their experiments today. Some stay “virtual” for years, in fact. So how is “cloud biology” different from conventional CROs? Pande explained that “cloud biology” is a direct analogy to cloud computing. “You don’t have to build a server farm” the way internet startups used to. “You could say [that even before cloud computing,] we could have outsourced calculations to Sperry and IBM and Burroughs.” Pande diplomatically called those old-fashioned approaches “high-touch,” especially in contrast with the “low-touch” process of ordering up experimental results on a laptop, an approach, Pande said, that reduces “friction” and increases elasticity. “The friction part is not to be underestimated,” Pande says. “Think about Airbnb and Lyft and Uber. You could have done that before mobile. You could have just gone to your laptop to do those things. But mobile really reduces the friction to make it possible. Human beings are surprisingly lazy. A little bit of friction goes a long way to making something happen or not happen…Right now, removing the friction to be able to do things is very intriguing.” Due to those online platforms, Pande said, there are opportunities out there for a startup company that just did not exist before.

Why we should be cautious

To inform myself about the possible challenges for “cloud biology,” especially as it applies to drug discovery, I spoke to three experienced Boston-area experts, all of whom rose to become executives in companies like Alnylam, Biogen and Vertex and all of whom are now working with earlier-stage drug discovery companies. Their pushback fell into three main buckets:

  1. The best drug discoverers, those pioneering new platforms in hot spaces, are already using outsourcing to its best effect, thank you very much.
  1. Every great drug discovery researcher, like every great academic biologist, gets stuck on the inherent “messiness” of biology. No computer or data scientist will overcome that alone, cloud or no cloud.
  1. Automation has inherent limitations, a fact that should keep researchers cognizant of the advantages of the old-fashioned, “high-touch” way of doing things.

First, outsourcing. Old drug discovery hand Michael Gilman is a well-known serial outsourcer. He ran his last company, Stromedix, successfully sold to Biogen in 2012, out of a former bank in East Cambridge. There was never a lab. Still, Gilman came down squarely on the side of human-led drug discovery. Having gone as far as it could with “push-button” drug discovery that could be done at Evotec and other CROs, his latest company, Padlock Therapeutics, recently decided to “devirtualize,” a decision Gilman artfully described in a recent blog post. “We pushed this program [at Padlock] very, very far in two years on less than $12 million in capital,” Gilman told me. “There is no way we could have done that if we had hired a staff and if we had been paying rent in a lab.”

Gilman then went even further in support of human-run discovery with all its foibles. In a session he shared with Werner Lanthaler, the CEO of Evotec, at the Convergence Forum on Cape Cod in May, 2015, Gilman extolled the virtues of outsourcing to CROs like Evotec specifically because of the experience and insight of the people who were running the assays. “At the risk of sounding like a luddite,” he said, “I do believe that true discovery biology remains more of an art than a science. Hypothesis generation is a human act. It is pattern recognition; there is some synaptic spark.”

Second, biology is very messy indeed. Mark Murcko, former Chief Technology Officer of Vertex Pharmaceuticals and currently an advisor to a number of biotech startups as well as an executive at a leading life sciences software company, Schrödinger, where Pande is a scientific advisor, put it this way: “ Can you solve the biology problem with the latest technology out on the cloud? I have not seen that. Every day, every company I work with is struggling with target validation, biomarkers and patient selection. Questions come up such as “‘I have a hit from a screen and I do not know what it does’ and  ‘Which of these two targets (out of ten in total) do I pick for my next drug discovery project?’” Murcko said. All of this, Murcko said, gets into biology that is “half-right and half-wrong. For example, ‘I have to extrapolate from mouse data.’ Or ‘It is human genetic data but it’s from the germ line [e.g. from sperm or egg cells or their immediate progeny].'”  So the data do not necessarily teach you what will happen if you shut down 80% of the activity of the same target in a 50-year-old patient.

Third, automation has inherent limitations. The transition to robotic or virtualized labs might marginally improve efficiencies but not without tradeoffs. Another senior Boston-area drug developer, CEO and PhD neuroscientist Nagesh Mahanthappa of Scholar Rock, told me, “You can automate an assay and be in love with the output. But if you bother to look you can find out that you have been grossly misled…These days, so much equipment is automated or semi-automated. They give you results in thirty minutes. But the results often sound like this: 'The molecule inhibited signaling.' You have to remember to ask in that case, are you sure that the molecule did not just kill all the cells? Or that the cells were not washed off the plates during a washing step?”

Even assuming that robotic assays will have human checkers somewhere along the line, what happens, Mahanthappa asked, when there is an unexpected result? “I am not pooh-poohing that these [robotic or virtual] assays are available on a commercial scale at higher volumes than before and with perhaps a better user interface. But you will not be able to completely replicate the experience of doing the experiments yourself, with all the frustrations and insight that brings.” As usual when technology enters the picture, there will be tradeoffs and the part the experienced drug developers have trouble imagining is how big a benefit can counteract or even vastly outweigh these obvious flaws. When I was a venture capitalist, my motto was “it’s the biology, stupid.” In other words, I constantly reminded myself to focus on investing in fundamental advances in basic biology, the kind that would win Nobel Prizes. One of my successful investments was in RNA interference, a field that yielded the 2006 prize just five years after the field broke into the headlines; now my consulting practice is active in CRISPR/Cas9-based gene editing, which most observers expect to garner a Nobel in record time. I have to wonder how "cloud biology" could add to, or even create, one of these fields.

An instructive precedent

To imagine how biology might successfully be “clouded,” it is instructive to turn back to the project that put Pande’s work on the radar: the Folding@Home project he started at Stanford. Now in its fifteenth year, Folding@Home (F@H for short) has taken the apparently intractable problem of predicting the behavior, including the folding behavior, of wickedly complex protein molecules, including their mysterious and otherwise unpredictable interactions with would-be drug molecules, and broken that problem down into pieces small enough that they could be worked on by hordes of people with personal computers across the globe, hence “@home.” According to Wikipedia, the F@H project “has pioneered the use of graphics processors in PCs and Macs, PlayStation 3s … and some Sony Xperia smartphones” and “uses statistical simulation methodology that is a paradigm shift from traditional computational approaches.” F@H is “one of the world's fastest computing systems, with a speed of approximately 40 petaFLOPS,” wrote Wikipedia.

F@H is reported to have led researchers to some advances in drug discovery in Alzheimer’s disease, Huntington’s disease and cancer. The etiology of all of these diseases can involve the misfolding of proteins. In all, according to the F@H web site, 129 papers have been published in peer-reviewed literature based on F@H. As far as I can tell, none of these has led to a drug candidate in clinical testing. To be fair, F@H was set up as a non-profit and so the project team did not pursue its own drug discovery efforts.

Pande mentioned that F@H was included in the Guinness Book of World Records in 2007 for having “the world’s most powerful distributed computing network” for achieving a speed of one petaflop. Recreating that speed record, he said, now costs $400 at Amazon. “This,” he said, “is why there is an interesting opportunity now and why I am excited about moving to venture capital now: What used to be extraordinary and rare and record-breaking [is accessible to almost everyone.]”

One critical success factor for Folding@Home and other crowd-sourced computing challenges, such as the Sage Bionetworks-DREAM Breast Cancer Prognosis Challenge to predict survival among metastatic breast cancer patients based on large data sets, was that these challenges allowed non-biologists to contribute their intellectual energy and creativity along with compute cycles. Harnessing that sort of “expertise” runs counter to the traditions of academic biology and especially of pharmaceutical-industry-based drug discovery. But I could see how, if the challenges are framed correctly, framing being one of Pande’s specialties, such an approach might bring a breath of fresh air into the field. Murcko agreed, adding “I completely buy this idea of ‘democratizing’ science--getting diverse, globally distributed, very smart people to look at complex data and see what they can learn.”

The analogy to Folding@Home should be applied with caution, Pande added in our follow-up call. The concept of breaking down a biological challenge into components that can be tackled through computational approaches is useful, he said. But “the connection to Folding@Home in my venture capital work is not through cloud biology,” he said. Rather, Folding@Home just illustrates “that Moore’s law can have an impact on biology. But the impact is more on machine learning,” for example, than on cloud biology.

Even the senior researchers quoted earlier see some areas in which “cloud” or “virtual” approaches can lead to big advances. About Mousera, Murcko observed that “putting sensors on mice generates a lot of data. You need a way to analyze the data efficiently. It’s like Waze [the mobile traffic app acquired in 2013 by Google for $1.15 billion]. There is a ton of data coming in and you use the cloud to process the data. I am 100% convinced on that one,” he added.

In cases in which the hypotheses are clear and hence the questions are already well-defined, both Murcko and Mahanthappa agreed that virtualization could be a good thing . But, Mahanthappa went on, “I am questioning the hype: it is not that software ‘eats’ bio,” as a blog post on the Andreessen Horowitz web site claimed, but rather that “software can make certain experiments more feasible, increase the throughput of those experiments and make the quality of data obtained higher.” Those areas might grow, I would add, especially if the “clouding” begins to scale to larger CROs with more bandwidth or if the initial handful of newfangled CROs in California begin to grow.

Quite a bit of our podium discussion involved clusters. I asked Pande if Andreessen Horowitz will focus its investment geographically on Bay Area companies and he said it would. But doesn’t that ignore the massive and growing cluster of drug discovery expertise within a few blocks of Kendall Square in Cambridge, I asked. That is a different kind of cluster, he replied. Where to invest geographically in this emerging field, he explained, really depends on where you think the innovation will be coming from. “To the extent that software will play a key role,” Pande said, “the Silicon Valley network [and location] will be relevant. When it’s hardware, therapeutics and small molecules, Kendall Square will be more relevant. I am not saying that software will replace all of this. I think it will have a big, complementary impact.”

It’s a bit early to say if Pande is right or wrong. To me, the ecosystem growing up around Emerald Cloud Labs and the other companies Pande mentioned is a bit like a mobile platform for which app developers are just beginning to build apps. But there is one thing I noticed while speaking to the highly qualified sources I quoted earlier: one of the “network effects” of a cluster is that we are all reinforcing each other’s thinking. The same way that it was engineers and data scientists, not biologists by training, who submitted the winning entry in the breast cancer challenge, there could be some unexpected contributions coming to drug discovery from software companies unencumbered by the baggage of what everyone in Kendall Square “knows.” In that regard, I think Pande is onto something, possibly something big.

# # #

Disclosure: Evotec has been a client of CBT Advisors.