Hmm, the disclaimer at the bottom definitely made something click for me regarding the hype-y tone..
It would've been nice to see some comparisons with competitors: what are e.g. Cellarity, or Emulate Liver-Chip doing? They're closing higher accuracy on DILIrank than Axiom, for example
Nice article. I assume the next step for Axiom (or a similar company) is to develop models that have some predictive validity for clinical efficacy. That would be a more challenging project given the multiplicity of organ types and effector systems to mimic. But the POS stats are pretty brutal and the trajectory is not going in the right direction.
The comparison to early quant finance is sharp. What Karmali's pointing at is how aggregatign weak signals can flip an entire industry's approach, and that same logic applied to biological phenotypes could be transformative. The vacuole detection example stands out because it's showing a failure mode current assays miss entirely, cells dunno they're about to die. If Axiom's models can consistently spot these latent signals before standard tests, that's genuinely a step change in predictive capability.
Hmm, the disclaimer at the bottom definitely made something click for me regarding the hype-y tone..
It would've been nice to see some comparisons with competitors: what are e.g. Cellarity, or Emulate Liver-Chip doing? They're closing higher accuracy on DILIrank than Axiom, for example
(just to be clear: did "commissioned" mean they paid you for it?)
Yes, they paid me to write it - that's why there's a disclosure at the bottom.
Fair point on competitors, but Emulate and Cellarity are solving different problems with
different methods. A landscape comparison would be a different article. This one is about what Axiom specifically is building and why.
Nice article. I assume the next step for Axiom (or a similar company) is to develop models that have some predictive validity for clinical efficacy. That would be a more challenging project given the multiplicity of organ types and effector systems to mimic. But the POS stats are pretty brutal and the trajectory is not going in the right direction.
Super insightful!
Disclaimer should go in the beginning. A little annoying to only realize at the end I read an advertisement.
Thanks for the feedback! Fixed.
The comparison to early quant finance is sharp. What Karmali's pointing at is how aggregatign weak signals can flip an entire industry's approach, and that same logic applied to biological phenotypes could be transformative. The vacuole detection example stands out because it's showing a failure mode current assays miss entirely, cells dunno they're about to die. If Axiom's models can consistently spot these latent signals before standard tests, that's genuinely a step change in predictive capability.