CRISPR is more like a machine to make large-node-size integrated circuits.
Somebody else has to design the circuit, make it manufacturable, and integrate the circuit with a whole bunch of other hardware.
I work in biotech at a company that is one of the few golden geese that lays 2-3 successful drugs with no competitors every few years. I have 30+ years of experience (deep experience) in machine learning, biology, and computer science.
We are so far behind where we could be, in terms of turning biology into technology, that's almost shameful. Every day I see another system that says it can generate 10 times the data of the previous machine, but the actual amount of knowledge we are extracting for all that data collection is growing logarithmically. This is because for a long time biology has greatly underfunded computing and data.
The one great shining light is AlphaFold. AF2 finally demonstrated to a wide range of scientists across many domains that a really great team using techniques that are barely known outside of FAAMG can work with some long-term experts to move a metric (quality of predicted protein structures compared to golden data) substantially further and faster than even the most wildly optimistic predicted. Not only that, some of the techniques they used didn't even exist several years ago (transformers, jax, various graph learning systems), and the work was replicated externally once the leading academic team had a hint of the direction to go in.
To me, nothing about what I said is surprising to me; I predicted these outcomes a long time ago. Most of the reasons that it comes slower than it could are combinations of culture, incentive, morals/ethics, politics, innovator's dilemmas and a hundred different bottlenecks. Recently, the challenge has been that most of the really smart computational biologists disappear into FAAMG and don't contribute back the things they learn there to research.
We're all waiting for that next moment when the cross product of Genentech and Isomorphic Labs announces that they have a computational model that can do end to end prediction of drug, from initial disease target to FDA approval post-phase III trial. That's been the dream for some time but we're nowhere near it still, and it remains to be seen whether some group can conjure all the necessary bits to solve the remaining underlying problems associated with that "far beyond NP-hard problem"
I work in biotech at a company that is one of the few golden geese that lays 2-3 successful drugs with no competitors every few years. I have 30+ years of experience (deep experience) in machine learning, biology, and computer science.
We are so far behind where we could be, in terms of turning biology into technology, that's almost shameful. Every day I see another system that says it can generate 10 times the data of the previous machine, but the actual amount of knowledge we are extracting for all that data collection is growing logarithmically. This is because for a long time biology has greatly underfunded computing and data.
The one great shining light is AlphaFold. AF2 finally demonstrated to a wide range of scientists across many domains that a really great team using techniques that are barely known outside of FAAMG can work with some long-term experts to move a metric (quality of predicted protein structures compared to golden data) substantially further and faster than even the most wildly optimistic predicted. Not only that, some of the techniques they used didn't even exist several years ago (transformers, jax, various graph learning systems), and the work was replicated externally once the leading academic team had a hint of the direction to go in.
To me, nothing about what I said is surprising to me; I predicted these outcomes a long time ago. Most of the reasons that it comes slower than it could are combinations of culture, incentive, morals/ethics, politics, innovator's dilemmas and a hundred different bottlenecks. Recently, the challenge has been that most of the really smart computational biologists disappear into FAAMG and don't contribute back the things they learn there to research.
We're all waiting for that next moment when the cross product of Genentech and Isomorphic Labs announces that they have a computational model that can do end to end prediction of drug, from initial disease target to FDA approval post-phase III trial. That's been the dream for some time but we're nowhere near it still, and it remains to be seen whether some group can conjure all the necessary bits to solve the remaining underlying problems associated with that "far beyond NP-hard problem"