Thursday, September 5, 2019

Jorge Conde and Logistics/Strategy of KNOME (Bioinformatics for Whole Genome)

A17Z Podcast
Software Company in Healthcare

(Extracted)


https://en.wikipedia.org/wiki/Knome  






JORGE CONDE - KNOME
                             So in our experience and know me, it was interesting because here, this is a company with the sole purpose of the company was to provide software capability to analyze genomic information. And so, you know, when you launched that, your assumption is, well this could be used to power all kinds of applications. It could be used for research, either in academia and industry. It can be used for, you know, clinical diagnostics, flexible. We thought it was very flexible and so challenge one is, you know, a solution looking for a problem is always a very, very dangerous thing. I think that's universally true. It's especially true in the healthcare space. And challenge two was understanding exactly where in the case of the clinical setting where this technology would be used in the workflow. So here we want it to go after the clinical labs.
Speaker 2:          That was your initial hypothesis.

Our initial hypothesis for an application in a clinical setting, you have technicians and docs that are inside of the laboratory setting, receiving samples, running a test, analyzing the results of that test, generating a report that gets signed off by lab director, that goes back to a physician.

Usually it's in the form of a diagnosis, right? And it gets signed off and it goes to the physician. The physician now takes that report and basically decides what to do based on that information. So our assumption was, well, if you have the ability to sequence DNA now in a way that you couldn't before, before you'd have to do all of these specific tests, you have to know what the test, and then you'd test it and then you'd get a report. You have to know what street lamp the keys were under, right?
Speaker 2:          Like they're in that case. Whereas once you had the full genome, you would just sequence everything and just run a bunch of software [careers.] So our thought going into this was, well, that's an incredibly powerful tool for clinical labs because first of all, you can sequence just once and analyze over time, right? You can again get like a totally legitimate, right. And it turns out that there was a lot of challenges with that assumption. The first one is every lab is different.
A lot of them didn't have the budget or the willingness to basically pay the upfront piece to buy the capability to use this technology, or they didn't have the ability to sequence everything upfront, even of all of the subsequent queries would be technically free later the way they're reimbursed. Oh, how fascinating. Too expensive. Basically it's too expense. So even the, theoretically there's an ROI, a return on the investment of sequencing up front, just the way the industry is structured, the way reimbursement flows, the way payments flow.
Speaker 2:          It just didn't make sense for a lot of labs to do this.

So how is that not just a complete roadblock at that point it was a big roadblock. So that would, that required us to do was to then focus on clinical labs that had the ability to make certain investments in up front cost and those tended to be very sophisticated labs that do a lot of research work in addition to patient care. Then they tended to be on this sort of on the bleeding edge and they wanted to incorporate new technology and they were great partners and all of that. But then it goes back to your N of one problem.

So you sell something into that lab and you go next door and next door. It has a totally different set of capabilities, a totally different set of constraints, a totally different set of expectations.
Speaker 2:          And so therefore all of a sudden the solution you created for lab a is not relevant or unattainable for lab B.

 Now to just add to the stepping in it, you know, when you're analyzing genomic data, there's a massive amount of computation required. And so we went in there assuming, well this is easy, we're just gonna shoot all of this up to the cloud, we'll run the analysis, we'll send the data back to the lab to the lab, could verify it, generate a report, and off we go. It turns out labs weren't comfortable sending data up into the cloud full stop at that time, at that time, arguably even today, arguably even today in 2019 but definitely at that time we probably should have known that earlier. That would have changed how we thought about going into the clinical lab space. How would you have done your homework?
Speaker 2:          I mean, what would that have actually looked like? It was frankly, I think just defining the specs of what would be required to bring in our technology because I think people intuitively know that genomic data is massive. But you know, I don't think they know sort of the level of computation required to run the interpretation. Right? So like really running the numbers, running the numbers for them. And by the way, we tried everything. I mean we brought representatives from AWS that could show them that they had a HIPAA compliant cloud, that they had received all the certifications and it came back to risk aversion. So someone in the lab director saying like, look, I'm sure all of that's true, but I'm not going to risk sending all of this data up into the cloud. So that was a big, big challenge for us. And it ended up being a major limitation for our ability to expand into the clinical setting because of all of those barriers.
Speaker 2:          So what did you do? We have to do a plan a and a plan B. And so the plan a was we assumed that there would be a couple of forward looking labs or forward thinking labs that would be willing to work in a cloud environment much easier to deploy there. The plan B was we had to create a box, we had a grid of box and the box had to have essentially the competition. Yeah, we had a normal plan. Remember that? Oh my God. Because they didn't want the data to go outside and it's for the reasons that we'd expect, you know, there, there's regulatory, there's risk associated with that today in 2019 in fact, the companies that have managed to use this technology have taken the sort of full stack service approach. So that sort of high, low strategy became the approach is get folks to deploy into the cloud when they were willing to, and in the case where folks needed an appliance, we basically had to go to labs that had enough of a sample volume that an appliance made sense for them and make basically the case there from an investment stand.
Speaker 1:          So again, multiple choice variety and like addressing in different ways. Okay.
Speaker 2:          Pure software company in healthcare is a really hard thing to do because on the one side you have this challenge that it's hard to create a sort of a solution that's going to fit everyone and therefore you need to have some level of services around that software that's on one extreme. So when you need to have humans in the process or in the loop, and in the other extreme of it, if it is pure software, then it's considered the, it should be free. So it's very hard to abstract value.
Speaker 1:          That's so interesting. Do you think that's shifting at all with the kind of understanding of the importance of data and some other,
Speaker 2:          yeah, look, I would argue with shifting on a couple of of axes. The first one is, is data is becoming more and more valuable. Historically data was viewed as being either too small in terms of its impact, too narrow, too dirty, et Cetera, et cetera. Too difficult. Yeah, to unstructured, you know, so that historically had been the case. So if you have ways to ingest data and clean it and make it meaningful, then I think that is valued. Probably the most public one is would flat iron was able to do and ultimately getting acquired by Roche for $2 billion. And that's viewed as using an electronic medical record to capture patient experiences, take that information and give researchers the ability to drive valuable insights from that. That's a relatively new thing. So I think there is the ability to create value there. So I think that's one axis. I think there's a general shift in the model that having a tech enabled service can be a valuable thing and if done well can be a scalable business. In other words, if you know what you're trying to build and if the software layer reduces sufficient friction in the system and allows you to add people not linearly as you scale, right, but in a leverageable way, then all of a sudden you can have tech enabled services that can grow and become large businesses.
Speaker 1:          So leaning into what it is that makes it difficult almost. And then scaling that, leveraging that. Exactly. Finding ways to make that scalable. Yeah, no, that's not easy to do, but

No comments:

Post a Comment

Note: Only a member of this blog may post a comment.