Very thoughtful – thanks for sharing. When people ask why the biotech markets have been so stagnant, I reflexively say something about interesting rates. But I think that's largely cope and the real reasons are the ones you mention: our target discovery methods are no match for complex disease biology and we're running out of easy targets. Excited for your vision.
Jacob...am sorry to disagree. The problem is precisely the target-centric nature of drug discovery as envisioned by Pharma and Biotech and sadly, most if not all AIDD companies have followed the pied piper and like the proverbial sheep, are jumping off the proverbial cliff. You are right that complexity matching is the future but it is more than that. We are swimming in a sea of dark shit that we do not know. We only know 10% of proteins and ergo, only a smidgeon of available pharmacology. Biology as conveyed through Lehninger and Strickland to me as a Biochem student, is not at all 2D and linear, it is a network. I challenge all these Pharma/Biotech bozo's to show me how, for the most part, and particularly in complex multi-factorial disease conditions, how this can be reduced down to a single or even a few single targets. There are some valid targets of course but drugs are inherently multi-targeted so one always risks bringing in unintended biological consequences. It is time for the industry to stop deluding itself. The models used (as elegantly shown by my friend Jack Scannell) are, for the most part, useless at translating into humans and part of the problem is that we use them for the wrong purpose and to answer the wrong questions. The future is bright but it is not anchored in the past!!
Since I've started digging to genetics (as a complete outsider to biology) my intuition was that target-based approach and genetics are inherently one-dimensional while it biology by itself is extremely complex and messy - it make sense that given 50 years of research we are running out of "single switches".
It reminds me a bit of N-gram models in NLP back on 2010s - they were simple, intuitive but were hard to scale and produced diminishing returns in quality since language is a complex system. In this case, neural nets (as universal approximators) were able to advance the state of the art and it seems to me the same could happen in bio but the key questions are primarily about data:
1/ representation: to use DNA, RNA, proteins, something else?
2/ what's the objective function? ESM/Evo like next amino-acid / nucleotide prediction for pre-training or using some pertrubation data?
3/ scale: how / where to get large-scale of such data? it requires large CAPEX to achieve so and we have to have fast feedback loop to understand that data generation process works well on small scale, so we could collect more data or pivot completely.
Very thoughtful analysis, although the timelines for the adoption of a new “systems biology” approach regarding drug discovery are pretty depressing. I’d be curious if the author is aware of the prevalence of translational models that can accurately recapitulate human disease assets. It’s often been argued that one of the reasons for poor and stagnant clinical trial success is the lack of valid animal models in many therapeutic areas (cancer being a prime example here).
Very thoughtful – thanks for sharing. When people ask why the biotech markets have been so stagnant, I reflexively say something about interesting rates. But I think that's largely cope and the real reasons are the ones you mention: our target discovery methods are no match for complex disease biology and we're running out of easy targets. Excited for your vision.
what an absolutely fantastic post, thank you so much for writing it.
Thank you very much sincerely
Спасибо большое искренне ❤️ 👍
incredibly insightful
Jacob...am sorry to disagree. The problem is precisely the target-centric nature of drug discovery as envisioned by Pharma and Biotech and sadly, most if not all AIDD companies have followed the pied piper and like the proverbial sheep, are jumping off the proverbial cliff. You are right that complexity matching is the future but it is more than that. We are swimming in a sea of dark shit that we do not know. We only know 10% of proteins and ergo, only a smidgeon of available pharmacology. Biology as conveyed through Lehninger and Strickland to me as a Biochem student, is not at all 2D and linear, it is a network. I challenge all these Pharma/Biotech bozo's to show me how, for the most part, and particularly in complex multi-factorial disease conditions, how this can be reduced down to a single or even a few single targets. There are some valid targets of course but drugs are inherently multi-targeted so one always risks bringing in unintended biological consequences. It is time for the industry to stop deluding itself. The models used (as elegantly shown by my friend Jack Scannell) are, for the most part, useless at translating into humans and part of the problem is that we use them for the wrong purpose and to answer the wrong questions. The future is bright but it is not anchored in the past!!
Awesome essay Jacob!
Since I've started digging to genetics (as a complete outsider to biology) my intuition was that target-based approach and genetics are inherently one-dimensional while it biology by itself is extremely complex and messy - it make sense that given 50 years of research we are running out of "single switches".
It reminds me a bit of N-gram models in NLP back on 2010s - they were simple, intuitive but were hard to scale and produced diminishing returns in quality since language is a complex system. In this case, neural nets (as universal approximators) were able to advance the state of the art and it seems to me the same could happen in bio but the key questions are primarily about data:
1/ representation: to use DNA, RNA, proteins, something else?
2/ what's the objective function? ESM/Evo like next amino-acid / nucleotide prediction for pre-training or using some pertrubation data?
3/ scale: how / where to get large-scale of such data? it requires large CAPEX to achieve so and we have to have fast feedback loop to understand that data generation process works well on small scale, so we could collect more data or pivot completely.
Curious what your thoughts are on these :)
Great summary, totally agree! I've been building with this in mind for years.
Very thoughtful analysis, although the timelines for the adoption of a new “systems biology” approach regarding drug discovery are pretty depressing. I’d be curious if the author is aware of the prevalence of translational models that can accurately recapitulate human disease assets. It’s often been argued that one of the reasons for poor and stagnant clinical trial success is the lack of valid animal models in many therapeutic areas (cancer being a prime example here).