I’ve decided to start interweaving my regular posts (about the challenges faced in building TechBio companies and ways to address them) with spotlights on TechBio and adjacent companies that I think are making exciting contributions to the field, and bringing unique value to the world through the meta-experiments they are conducting. My intent is three fold:
Showcase the diverse approaches being taken to advance the field of technology enabled drug discovery.
Articulate what I think makes for a compelling platform.
Draw some attention to the companies I spotlight.
Showcase the Diverse Approaches in TechBio
Something I think most humans suffer from is an over-binarization of the world. We reduce situations to yes or no, on or off, black or white. It’s a simplification that helps with rapid processing and decision making, but lacks the nuance required to really understand and optimize things. Related to over-binarization is acting as though situations are zero-sum when they are in fact not. It’s a “we win, they lose” binarization that is often fallacious.
One way I found myself incorrectly binarizing the world pertained to tech-enabled drug discovery. While Recursion was by no means the first player in TechBio, we were an early mover, and one that gained considerable traction ahead of many others. Most of the others playing in the field at the time were led by similarly unknown individuals - everybody in the field was a nobody at that point. And then some of the superstars appeared. Either the founder was a well-known and respected figure in drug discovery or AI, the investors were highly reputable in the space, or both.
At that moment, I thought binarily and zero-sum. I thought that with the rise of others we were in for a struggle to keep rising ourselves. I thought the world would think that an investment in one of these TechBio companies was effectively an investment in any of them (and in some ways, some investors do seem to think that way), and as a result competition would be stiffer. I momentarily naively thought that there must be one right way to do technology-enabled drug discovery, that only one of us could be right, and that we had to prove it was us.
This was over 6 years ago. Since then I’ve realized that there are far more shades of gray than I had appreciated. I’ve learned that biology and chemistry are both so complex and so ill understood that it will require a myriad of ways to measure and explore them if we are ever to decode them. I’ve come to understand that the diversity of tech-enabled drug discovery approaches are in many ways far more independent and thus quite non-zero-sum; multiple could, and will, work quite well and bring great therapeutic value to patients.
I hope in writing these spotlights that I will be able to demonstrate some of the diverse approaches being undertaken by these TechBio and adjacent companies that I know well.
Articulate What Makes a Compelling Platform
Another intent is to refine and articulate clearly what I think makes a compelling platform within a TechBio company. To be totally honest, this may be an exercise in me rubber ducking my readers, wherein I refine my own views by writing them down in clear prose. Currently my views center on a few principles:
It must be rooted in experimental data generation/collection
It must contain an element of scale with accompanying automated processing
It must be inherently
unless-biased in its hypothesis generation
Rooted in Experimental Data Generation/Collection
Technology is being developed and commoditized so quickly that developed models on static data are not a sufficient moat. Furthermore, publicly available datasets suffer dramatically from various biases, batch effects, design differences, inconsistent naming and underspecification that they are often untrustworthy for many tech-enabled drug discovery use cases. For a TechBio platform to be compelling, it must involve its own experimental data generation - to answer many of the outstanding questions facing any TechBio platform, one must be able to design and execute its own data-generating experiments. It must be able to dictate its own terms of the data being generated, which means either having full control over experiment design and specification, or collecting from such a broad and deep data stream that it can narrow down to the data necessary to answer its questions.
Element of Scale and Automated Processing
While I don’t believe unequivocally that all problems can be solved by just throwing more data at them, I do believe that all-else-equal, more data is better. And because biology and chemistry are so complex, and drug discovery is likened to finding a needle in a haystack, you’ve got to be able to sift through a lot of hay to find that needle. So there must be an element of scale in the data generation/collection.
To be able to effectively search through this level of scale, there must be accompanying automated processing. This may need to be both physical automation (think lab hardware, robotic arms, liquid handlers, sequencers, imagers, etc.) and digital automation (distributed compute systems, orchestration, data registries, etc.). But we can’t simply have more humans pipetting in the lab or processing data in Excel files. And the more complex the analysis is, the greater the requirement for well-engineered data systems to ensure quality, timely processing and accessibility of data.
Unless-biased Hypothesis Generation
I started writing “unbiased” but realized that no matter what we do, we have biases (without them, we actually can learn very little). But we should limit our biases as much as possible, especially in hypothesis generation. The most compelling platforms to me are those that enable the data to not just determine the outcomes of hypothesis testing, but also guide and shape hypothesis generation. Without this, we are limited to exploring the areas of biology and chemistry that are fairly well understood/established, rather than charting the unknown, which is likely where the greatest opportunities lie.
As a result of this belief, I’m naturally drawn towards companies with platforms that can aid in scaled target discovery (identifying novel disease-relevant biology), or are generating data specifically to shape their predictions rather than just confirm them.
Draw Attention to Companies I Like
Another intent is to bring some attention to the companies that I like. I’m not naive though - I recognize that my influence in this regard is limited, and probably actually even smaller than I think it is. But maybe by spotlighting a company I like, some talented engineer or ML scientist might apply that otherwise wouldn’t have even known to, or some investor might reach out for more information that could help the company move forward. Maybe. But this kind of maybe is worth it in the effort to find more effective ways to bring medicines to patients.
I’ve also got to share a disclaimer that many of the companies I will spotlight are companies I’m likely connected with in some capacity. I likely either advise them, consult for them, have invested in them, or personally know some founding members. Part of this is because as a result of these connections, they are companies I know best and about whom I can accurately articulate what makes them unique. Another part of this comes from the reality that if I see a company who is developing a platform I think is really compelling, I like to work with them in some way. So yeah, I’m biased. I’m well aware of my biases, and I’m sharing them here so that you are aware too.
Conclusion
Look out over the next few weeks for my first TechBio spotlights. The world of tech-enabled drug discovery is fascinating, and there are so many different interesting approaches being taken to improve our ability to bring value to patients. Hopefully we’ll all learn about these approaches together and maybe this may even lead to some useful connections for either these companies, my readers, or both.