Great post. I'm as excited about foundational models for biology as the next scientist, but to your point on translation, the lower hanging fruit feels like using AI to address problems that have previously been immune to the kinds of efficiency advancements you see outside of science.
To be specific, the real issue with building software for biotech companies historically has been that each one is a special snowflake in terms of data. Expanding out of tiny niches (from one type of chemistry to another) used to cause a crazy amount of scope expansion in the product, which also just made building venture scale software super hard. Now thanks to AI, interfaces can be a lot simpler because LLMs mean you can work with messier data under the hood.
It's not as sexy, but I see this as more of an arbitrage — find stuff that's working outside of science, bring it into a scientific context. Give scientific companies the same quality of tools in areas adjacent of the lab that other industries have.
Thanks Anna, this resonates since we're living through this struggle right now.
What are you thoughts on hub-and-spoke models like Nimbus or BridgeBio to potentially avoid these traps?
By tying the platform (hub) to clear therapeutic goals (spokes), they claim to accelerate translation and reduce risk. But does this approach inherently dilute focus, or is it a more pragmatic answer to the “myth of platform” problem?
If the hub gets too much attention or the spokes are too diverse, the same platform risks you highlight could emerge. Curious how you see these models fitting—or failing to fit—your framework.
I think some of those efforts will likely meet with some success, but I don't view it as being fundamentally different from more traditional drug discovery - the value is still derived from fully integrated drug discovery, they're just claiming (and perhaps rightly so) to be building and reusing some expertise in their stack (but pharma would make the same claim!)
Super well articulated, Anna. Having been in this intersection (biotech r&d x operations x finance) for a while, I feel you have beautifully formulated the core problem of lack (or illusion) of modularity and need for widely accepted/open experimental validations to fix the leaky bucket situation faced in biotech.
My 1st thoughts on this (which I will research on using your article as a “thought” substrate) - shouldn’t there be some good examples of this same problem within the tech or other industry? What can we learn from them and how can we translate it to Biotech ?
Thank you for a thought-provoking post...I read it multiple times and can't make complete sense of it. I vaguely sense that you're blending two issues: network effects (or lack thereof) and the rewards of specializing in a step of product development (or lack thereof).
Let's start with the fact that all of biopharma's effort is directed towards one thing only: selling human therapeutics. No matter how many scammy, derivative or parasitic side-taps there are in this edifice, the wellspring of money is ultimately to sell drugs.
There are no network effects in pharma: just because I sold some Humira doesn't make me more likely to discover the next Humira. Every decade's high-flying company is the next's struggling has-been. There are linear, but not exponential effects such as: the rheumatologist interacts more often with the Abbvie rep and is thus more likely to be reached by an Abbvie message. That's what pharma calls "a franchise" - not a network.
Can you make mad (tech) money becoming a specialized actor in drug development (aka a CRO?). So far, not really. You could have a buzzy tech and be highly regarded, but that's called "being good at what you do and getting customers" - so is my plumber, and he's not a platform. Again, you haven't made it more valuable for external parties to do business with you.
I don't have a good explanation myself...I just think biology and human health are harder than tech.
Thanks Anna Marie for the deep analysis and insightful comments. I was mapping the ag experience onto footnote 5. Its just like pharma except there’s 100x less capital and 100x fewer partners :-) Two comments: you have hit the David Foster Wallace ratio of footnotes to main body—no easy task! And check out the Two Frontiers Project—a bootstrapped not-for-profit run by Braden Tierney: https://twofrontiers.org/
Great post. I'm as excited about foundational models for biology as the next scientist, but to your point on translation, the lower hanging fruit feels like using AI to address problems that have previously been immune to the kinds of efficiency advancements you see outside of science.
To be specific, the real issue with building software for biotech companies historically has been that each one is a special snowflake in terms of data. Expanding out of tiny niches (from one type of chemistry to another) used to cause a crazy amount of scope expansion in the product, which also just made building venture scale software super hard. Now thanks to AI, interfaces can be a lot simpler because LLMs mean you can work with messier data under the hood.
It's not as sexy, but I see this as more of an arbitrage — find stuff that's working outside of science, bring it into a scientific context. Give scientific companies the same quality of tools in areas adjacent of the lab that other industries have.
Thanks Anna, this resonates since we're living through this struggle right now.
What are you thoughts on hub-and-spoke models like Nimbus or BridgeBio to potentially avoid these traps?
By tying the platform (hub) to clear therapeutic goals (spokes), they claim to accelerate translation and reduce risk. But does this approach inherently dilute focus, or is it a more pragmatic answer to the “myth of platform” problem?
If the hub gets too much attention or the spokes are too diverse, the same platform risks you highlight could emerge. Curious how you see these models fitting—or failing to fit—your framework.
I think some of those efforts will likely meet with some success, but I don't view it as being fundamentally different from more traditional drug discovery - the value is still derived from fully integrated drug discovery, they're just claiming (and perhaps rightly so) to be building and reusing some expertise in their stack (but pharma would make the same claim!)
This is an incredible post and deeply insightful!
As a SWE trying to contribute to bio, this post hits the mark.
it takes a great general to see the obvious molstream.com
Super well articulated, Anna. Having been in this intersection (biotech r&d x operations x finance) for a while, I feel you have beautifully formulated the core problem of lack (or illusion) of modularity and need for widely accepted/open experimental validations to fix the leaky bucket situation faced in biotech.
My 1st thoughts on this (which I will research on using your article as a “thought” substrate) - shouldn’t there be some good examples of this same problem within the tech or other industry? What can we learn from them and how can we translate it to Biotech ?
Keep up the good work!
Great post!
Thank you for a thought-provoking post...I read it multiple times and can't make complete sense of it. I vaguely sense that you're blending two issues: network effects (or lack thereof) and the rewards of specializing in a step of product development (or lack thereof).
Let's start with the fact that all of biopharma's effort is directed towards one thing only: selling human therapeutics. No matter how many scammy, derivative or parasitic side-taps there are in this edifice, the wellspring of money is ultimately to sell drugs.
There are no network effects in pharma: just because I sold some Humira doesn't make me more likely to discover the next Humira. Every decade's high-flying company is the next's struggling has-been. There are linear, but not exponential effects such as: the rheumatologist interacts more often with the Abbvie rep and is thus more likely to be reached by an Abbvie message. That's what pharma calls "a franchise" - not a network.
Can you make mad (tech) money becoming a specialized actor in drug development (aka a CRO?). So far, not really. You could have a buzzy tech and be highly regarded, but that's called "being good at what you do and getting customers" - so is my plumber, and he's not a platform. Again, you haven't made it more valuable for external parties to do business with you.
I don't have a good explanation myself...I just think biology and human health are harder than tech.
Very good post
Brilliant and through provoking post thank you! Reinforces our thoughts. The comment on the Recursion JPM event made me chuckle.
Thanks Anna Marie for the deep analysis and insightful comments. I was mapping the ag experience onto footnote 5. Its just like pharma except there’s 100x less capital and 100x fewer partners :-) Two comments: you have hit the David Foster Wallace ratio of footnotes to main body—no easy task! And check out the Two Frontiers Project—a bootstrapped not-for-profit run by Braden Tierney: https://twofrontiers.org/