ALSO: Enveda, Anew Labs and Varda raise over $850M between them; Genentech signs a $1.5B collaboration with Earendil; Neuralink pretrains on 50,000 hours of brain data and reports a BCI record, and mo
Really enjoyed this, Andrii. The Biossil section makes you wonder how many potentially useful treatments we’ve shelved because we couldn’t identify who would benefit. I’d love to see AI get good enough at that to help select patients before a trial begins, as well as revisit drugs afterwards. There’s something quite hopeful about the possibility that a treatment someone needs has already been developed and is sitting on a shelf. Looking forward to seeing how these programs progress.
I agree with you. Indeed, one of the non-very-obvious ways AI can make an impact on science is being able to revisit old data/experiments, give second life to old attempts. It is a fruitful ground scientifically speaking, but crucially, it can also be a great business model.
The interesting variable is the purchase price. A failed asset is sold by a company that has already written it off, so the buyer acquires the molecule, the safety data and years of development at the price of a disappointment. Even if the subgroup thesis is often wrong, the cost basis carries the trade. The seller is not parting with a drug. It is parting with a line it no longer wants to explain to its board.
The failed-drug angle is interesting because the sunk cost may finally become useful data instead of a dead end. But rescuing a molecule still has to survive the same biology and trial economics that killed it the first time. The model can find the signal. The company still has to prove it.
Rescuing a failed drug is compelling because the asset already carries real-world evidence about what did not work. The hard part is proving the AI found a biologically meaningful second chance, not simply a better narrative for an old molecule.
Really enjoyed this, Andrii. The Biossil section makes you wonder how many potentially useful treatments we’ve shelved because we couldn’t identify who would benefit. I’d love to see AI get good enough at that to help select patients before a trial begins, as well as revisit drugs afterwards. There’s something quite hopeful about the possibility that a treatment someone needs has already been developed and is sitting on a shelf. Looking forward to seeing how these programs progress.
I agree with you. Indeed, one of the non-very-obvious ways AI can make an impact on science is being able to revisit old data/experiments, give second life to old attempts. It is a fruitful ground scientifically speaking, but crucially, it can also be a great business model.
The interesting variable is the purchase price. A failed asset is sold by a company that has already written it off, so the buyer acquires the molecule, the safety data and years of development at the price of a disappointment. Even if the subgroup thesis is often wrong, the cost basis carries the trade. The seller is not parting with a drug. It is parting with a line it no longer wants to explain to its board.
The failed-drug angle is interesting because the sunk cost may finally become useful data instead of a dead end. But rescuing a molecule still has to survive the same biology and trial economics that killed it the first time. The model can find the signal. The company still has to prove it.
I mean, yeah, OK model and great business execution still beat an OK execution and the best model in the world.
Fair. And execution here means years of trials and a lot of unglamorous paperwork, which is probably why the model ends up with the headline.
Rescuing a failed drug is compelling because the asset already carries real-world evidence about what did not work. The hard part is proving the AI found a biologically meaningful second chance, not simply a better narrative for an old molecule.
Pretty much.