Great piece, thank you for sharing it. I certainly subscribe to the notion that faster feedback loops would accelerate biology dramatically and I loved the notion of moving fast and making things easier for everyone.
But I'm not sure that you're addressing the slowest parts of the process in your piece. Sure, simulation would help – but we need much more data for that. Sure, automation would help – but you first need to know how exactly you want to do your experiment (and even then it may not be necessarily automatable).
This leads me to the same observation that you make towards the end of your piece – that currently biology is hard to predict, especially because everything is so interlinked that it is hard to isolate root cause when things fail. In a way, I see a parallel to ML here where you may not necessarily know ahead of time which combination of hyperparameters will lead to the best validation score until you train your model – which may take just as much time and be just as costly as wet lab experiments. But in ML at least you can exactly replicate your experiments whereas in the wet lab your millage might vary.
Therefore, I believe our focus should be on reducing this unpredictability. If I come up with an experiment, I should be able to validate it rapidly without getting stuck in issues like "you should have used buffer X" or "try this other kit, it's been working great for me lately". I don't think biologists are good at documenting (or, rather, sharing) these intuitions but, on the other hand, people have done tons and tons of similar experiments over the years so perhaps by combing through the entire literature we could get AI to provide decent research plans and assist in troubleshooting. Tools like PaperQA and OpenAI's Deep Research are good glimpses into how we could perform assay development more efficiently. But much more work needs to be done before these tools can actually make a difference.
Once we know how to do experiments reliably, we can test ideas fast and move forward!
> Therefore, I believe our focus should be on reducing this unpredictability. If I come up with an experiment, I should be able to validate it rapidly without getting stuck in issues like "you should have used buffer X" or "try this other kit, it's been working great for me lately". I don't think biologists are good at documenting (or, rather, sharing) these intuitions but, on the other hand, people have done tons and tons of similar experiments over the years so perhaps by combing through the entire literature we could get AI to provide decent research plans and assist in troubleshooting. Tools like PaperQA and OpenAI's Deep Research are good glimpses into how we could perform assay development more efficiently. But much more work needs to be done before these tools can actually make a difference.
I agree that more predictable experiments are an important component that we only touched on briefly. Thanks for raising this! (As an aside, Noah Olsman (https://www.nolsman.com/) made related, very interesting arguments on this topic to me years ago when we discussed it.)
great essay, and a very enjoyable read. i think part of having faster feedback loops is also developing new informative modalities of measurement that are cell/species agnostic and enable high throughput screens.
1) Everyone talks about wanting biology to move faster but what is it specifically that should be done? And more importantly, why hasn't that thing been done? I really don't think experimental speed is the biggest roadblock to progress in biology. Can it help at the margins? Sure. But good science fundamentally takes time, patience, and most importantly a lot of thinking.
2) Feedback loops are important, but feedback loops coupled with abstractions is where the true magic happens. If a new technique comes out to run an experiment 1.5 times faster than standard, I can't just import it like I would if a new python library came out. In theory, CROs are that abstraction. Why CROs aren't that transparent is a great question because it seems they have every incentive to run their experiments as efficiently and accurately as possible. And then tell the whole world how great they are! The CRO market seems very competitive, so it may be more due to the general conservative nature of biotech.
Some other ways of accelerating those feedback loops (by no means novel ideas but maybe relevant here):
Open science / publish uninteresting and negative data, saves people doing same experiments. If people are worried of posting their work then make it an anonymised platform.
Make doing science more accessible - create an open research community centre with communal lab space and equipment - importantly this must not be unfeasibly expensive. Like community bio labs.
Thank you for this interesting piece, full of good and useful references!
Yes, biologists have a big capacity for suffering, but this does not mean they enjoy the slow progress. You speak about bacteria growing slowly? Try working with plants!
Some comments here point to first deciding what problem we want to solve, before speeding ahead very fast towards we are not sure where.
Recently i read a piece about the divide between those who want to prioritize doing ”many, fast” and those who want to do “very few, clever, fast”. Also interesting.
If we are going to invest in speeding up biology, do let us think what we want to solve!
Increase antibody production in CHO???? Why? CHO is the tool we have, not the fit for purpose antibody making platform we would develop, given a blank canvas. Rather instead develop the best antibody-making platform.
So do let us improve the goals of our biological experiments and speed up the processes, but let’s do them on meaningful goals.
Thanks for your comment. I think we largely agree, but lemme just repeat back where I land on this in case it provokes useful discussion. I agree that it's important to pick the right problems in general and that optimization can get locked into local minima (e.g. your CHO cell example). On the other hand, I think it can be easy to neglect the power of compounding benefits of continuous improvements.
Taking the Ab / CHO example, as I understand it, therapeutic Ab production has become orders of magnitude more reliable / cheaper as it's scaled up. Maybe we are hitting the limits of the current platform (I really don't know enough to say), but it seems like the effort to optimize that platform over the past few decades has paid off in spades. If that's true, I believe it would have been a mistake to not have done that optimization work because eventually a new paradigm would be needed.
Zooming out, what I'm really saying is that I think the "faster feedback loops" tent is big enough to hold both the people who want to tinker with current systems to speed them up and the pioneers who want to invent the next ones that get us order of magnitude gains. There is undoubtedly some optimal balance, but catalyzing the cultural change to get to the point where finding that balance is a top priority seems more pressing to me than finding that point right now. Does that make sense?
Yes, i also think we agree, also on “ it can be easy to neglect the power of compounding benefits of continuous improvements.” . I agree they have that power and i agree that it is often neglected.
My concern is, again using CHO and Ab production as example, to believe that those 30 years of concerted efforts to improve production that brought many benefits to the patients receiving the medicines and to research, have at best also brought us closer to understanding what are the basics of the biology for mAb production at scale so we can prepare the next step up, and at least has not stood in the way of potentially good alternatives. I am not sure of the first and i fear that we got the second. At what point in CHO improvement we collectively asked ourselves - hey, should we be applying this knowledge (and resources) to other potentially more flexible / [other benefits ] production systems? That is maybe the “point of balance” that you write?
I think you need to spend a year in a lab to gain some much needed perspective that you seem to entirely lack. I've never worked in a lab where urgency isn't the norm. Doing things in the physical world takes time. Whenever some breaks through a time barrier, they start a company and charge for that expedited timeline. Take plasmidasuarious for example. If Scientists could get results faster, or with less work, we would. To imply that we just don't think our work is important enough to warrant urgency is nothing short of naive and insulting.
Great piece, thank you for sharing it. I certainly subscribe to the notion that faster feedback loops would accelerate biology dramatically and I loved the notion of moving fast and making things easier for everyone.
But I'm not sure that you're addressing the slowest parts of the process in your piece. Sure, simulation would help – but we need much more data for that. Sure, automation would help – but you first need to know how exactly you want to do your experiment (and even then it may not be necessarily automatable).
This leads me to the same observation that you make towards the end of your piece – that currently biology is hard to predict, especially because everything is so interlinked that it is hard to isolate root cause when things fail. In a way, I see a parallel to ML here where you may not necessarily know ahead of time which combination of hyperparameters will lead to the best validation score until you train your model – which may take just as much time and be just as costly as wet lab experiments. But in ML at least you can exactly replicate your experiments whereas in the wet lab your millage might vary.
Therefore, I believe our focus should be on reducing this unpredictability. If I come up with an experiment, I should be able to validate it rapidly without getting stuck in issues like "you should have used buffer X" or "try this other kit, it's been working great for me lately". I don't think biologists are good at documenting (or, rather, sharing) these intuitions but, on the other hand, people have done tons and tons of similar experiments over the years so perhaps by combing through the entire literature we could get AI to provide decent research plans and assist in troubleshooting. Tools like PaperQA and OpenAI's Deep Research are good glimpses into how we could perform assay development more efficiently. But much more work needs to be done before these tools can actually make a difference.
Once we know how to do experiments reliably, we can test ideas fast and move forward!
> Therefore, I believe our focus should be on reducing this unpredictability. If I come up with an experiment, I should be able to validate it rapidly without getting stuck in issues like "you should have used buffer X" or "try this other kit, it's been working great for me lately". I don't think biologists are good at documenting (or, rather, sharing) these intuitions but, on the other hand, people have done tons and tons of similar experiments over the years so perhaps by combing through the entire literature we could get AI to provide decent research plans and assist in troubleshooting. Tools like PaperQA and OpenAI's Deep Research are good glimpses into how we could perform assay development more efficiently. But much more work needs to be done before these tools can actually make a difference.
I agree that more predictable experiments are an important component that we only touched on briefly. Thanks for raising this! (As an aside, Noah Olsman (https://www.nolsman.com/) made related, very interesting arguments on this topic to me years ago when we discussed it.)
great essay, and a very enjoyable read. i think part of having faster feedback loops is also developing new informative modalities of measurement that are cell/species agnostic and enable high throughput screens.
1) Everyone talks about wanting biology to move faster but what is it specifically that should be done? And more importantly, why hasn't that thing been done? I really don't think experimental speed is the biggest roadblock to progress in biology. Can it help at the margins? Sure. But good science fundamentally takes time, patience, and most importantly a lot of thinking.
2) Feedback loops are important, but feedback loops coupled with abstractions is where the true magic happens. If a new technique comes out to run an experiment 1.5 times faster than standard, I can't just import it like I would if a new python library came out. In theory, CROs are that abstraction. Why CROs aren't that transparent is a great question because it seems they have every incentive to run their experiments as efficiently and accurately as possible. And then tell the whole world how great they are! The CRO market seems very competitive, so it may be more due to the general conservative nature of biotech.
Great post, thanks, got the cogs turning!
Some other ways of accelerating those feedback loops (by no means novel ideas but maybe relevant here):
Open science / publish uninteresting and negative data, saves people doing same experiments. If people are worried of posting their work then make it an anonymised platform.
Make doing science more accessible - create an open research community centre with communal lab space and equipment - importantly this must not be unfeasibly expensive. Like community bio labs.
Most excellent read and call to action! Biologists need to stop settling for the slow iterative status quo.
Thank you for this interesting piece, full of good and useful references!
Yes, biologists have a big capacity for suffering, but this does not mean they enjoy the slow progress. You speak about bacteria growing slowly? Try working with plants!
Some comments here point to first deciding what problem we want to solve, before speeding ahead very fast towards we are not sure where.
Recently i read a piece about the divide between those who want to prioritize doing ”many, fast” and those who want to do “very few, clever, fast”. Also interesting.
If we are going to invest in speeding up biology, do let us think what we want to solve!
Increase antibody production in CHO???? Why? CHO is the tool we have, not the fit for purpose antibody making platform we would develop, given a blank canvas. Rather instead develop the best antibody-making platform.
So do let us improve the goals of our biological experiments and speed up the processes, but let’s do them on meaningful goals.
Thanks for your comment. I think we largely agree, but lemme just repeat back where I land on this in case it provokes useful discussion. I agree that it's important to pick the right problems in general and that optimization can get locked into local minima (e.g. your CHO cell example). On the other hand, I think it can be easy to neglect the power of compounding benefits of continuous improvements.
Taking the Ab / CHO example, as I understand it, therapeutic Ab production has become orders of magnitude more reliable / cheaper as it's scaled up. Maybe we are hitting the limits of the current platform (I really don't know enough to say), but it seems like the effort to optimize that platform over the past few decades has paid off in spades. If that's true, I believe it would have been a mistake to not have done that optimization work because eventually a new paradigm would be needed.
Zooming out, what I'm really saying is that I think the "faster feedback loops" tent is big enough to hold both the people who want to tinker with current systems to speed them up and the pioneers who want to invent the next ones that get us order of magnitude gains. There is undoubtedly some optimal balance, but catalyzing the cultural change to get to the point where finding that balance is a top priority seems more pressing to me than finding that point right now. Does that make sense?
Yes, i also think we agree, also on “ it can be easy to neglect the power of compounding benefits of continuous improvements.” . I agree they have that power and i agree that it is often neglected.
My concern is, again using CHO and Ab production as example, to believe that those 30 years of concerted efforts to improve production that brought many benefits to the patients receiving the medicines and to research, have at best also brought us closer to understanding what are the basics of the biology for mAb production at scale so we can prepare the next step up, and at least has not stood in the way of potentially good alternatives. I am not sure of the first and i fear that we got the second. At what point in CHO improvement we collectively asked ourselves - hey, should we be applying this knowledge (and resources) to other potentially more flexible / [other benefits ] production systems? That is maybe the “point of balance” that you write?
Loved this deep dive into how to shorten feedback loops in biology. Appreciate the shoutout at the end, too :)
Great read. Thought provoking. Thank you.
I think you need to spend a year in a lab to gain some much needed perspective that you seem to entirely lack. I've never worked in a lab where urgency isn't the norm. Doing things in the physical world takes time. Whenever some breaks through a time barrier, they start a company and charge for that expedited timeline. Take plasmidasuarious for example. If Scientists could get results faster, or with less work, we would. To imply that we just don't think our work is important enough to warrant urgency is nothing short of naive and insulting.