A data moat is built from well planned experiments
*** A few quick plugs for things you should check out after you finish reading this week’s newsletter:
Please fill out the annual Bits in Bio member survey before November 22nd. We’re trying to make this community as useful and engaging as possible, and we can’t do that without your feedback. (And if you aren’t on BiB Slack yet, join here.)
Check out Deep Origin’s guide to the Life Science Software Landscape. This is the most thorough list of biotech software I’ve seen so far.
Read Kaleidoscope’s blog post about Biotech Milestones for Effective Fundraising. It has a ton of insights directly from investors in this space.
Through Merelogic, I’m offering a low-cost, packaged program/workshop on Data Reliability for Biotech Startups. The price is discounted for the next two weeks, so sign up today!
Now, on to the newsletter. ***
We’re getting towards the end of this exploration of the use cases that biotech software needs to support on the way to an IND. And since we’re taking this trip backwards, that means we’re near the beginning of the process. So this week’s newsletter is all about planning experiments. But it’s also not really, or not just, about planning experiments. Because experiments are just a means to get data, and data is what we really care about. So how do we get away from planning experiments and start planning data?
It’s becoming much more common for biotech founders to think about data as a differentiating asset. For example, this is mentioned multiple times in Kaleidoscope’s post I mentioned above. It’s different from thinking of data as a product, which is something for a future post (maybe). But the idea is that models are cheap and data is expensive. Most startups eventually realize that the main thing keeping a competitor with deep pockets from catching up is the time and trial and error it would take them to produce the same data. The data is a moat that will contribute to future valuations.
When bench teams plan experiments, they’re usually thinking about specific, narrowly scoped questions they want to answer - What does a specific compound or biologic do to a particular cell? Does a biological model have a particular response or behavior? These questions are part of a larger plan, but they’re very specific and very focused.
On the other hand, when a data-focused startup plans a data moat, they’re thinking about a much broader class of questions - the questions that they’ll need to answer today, in three months and in three years. This will probably include training ML models that require lots of both positive and negative examples, much more than is necessary to answer the narrow, immediate questions. Of course, they still need to answer those immediate questions, especially as investors become more skeptical of platform-heavy strategies. They’ll just want to cast a broader net in the process.
In some ways, this isn’t such a hard problem - there are usually reasonable ways to compromise between the specific and the broad. In other ways, it really is a hard problem, which is why many teams struggle to get it right. The hard part, it turns out, is identifying those compromises and implementing them. And this is where we get into shared mental models and team dynamics and all that stuff that isn’t just software.
You can think of it as a hammers and nails problem, but in the opposite direction from the problem of finding applications of AI/ML: Bench scientists know all the ways you can design experiments, and thus implicitly all the ways you can generate data. These are the hammers. Data scientists know what kind of data their models need, and they tend to be more willing to speculate about what data they’ll need in the future, i.e. what would contribute to an asset/moat. These are the nails.
As with the AI hammers and nails problem, when one group is familiar with the hammers and the other with the nails, it’s very hard to match them up. Data teams often don’t feel empowered to push back on experiment designs don’t suit their needs, they don’t know enough of the details to jump into planning, and wet lab teams who are used to planning experiments on their own often don’t invite them.
Ultimately, though, it isn’t an impossible problem. It just takes some careful thought and deliberate effort. Many biotech teams have managed to figure this out. You can too. And then you can shift from planning experiments to planning your moat.
Scaling Biotech is brought to you by Merelogic - an independent consulting firm that helps biotech startups design and deploy digital infrastructure to get their data under control so they can focus on the science (and build that moat).
Deep Origin’ are not knowledgeable in all software solutions of note. Anyone going there or taking this recommendation should be aware of that and do their own research.