This vial contains a new drug called PAC-832, which I recently invented to treat Alzheimer’s disease. It is the world’s first selective GalR1 antagonist.
I designed and synthesized PAC-832 in a chemistry lab I built in my garage. (1/16) https://t.co/J72K43yvlX
Getting to IND-enabling studies requires a huge breadth of expertise. You need a lot of people because the breadth of experience required is beyond what one person can do. He’s substituted that with the plagiarism robot, and even if it worked, he’s in no position to be able to verify the output. He has to trust it, and the plagiarism robot is very confident but often wrong.
The whole “in his garage” thing is also ridiculous. The hard part of drug discovery isn’t making molecules, it’s making molecules that pass ADME and tox and can be formulated and show exposure and efficacy in non-rodent species. You can’t do all those things in a garage. Either he’s farmed that out to CRO’s or (worse) just showed his structure to the LLM and had it predict the test results.
But the main thing is that even the best machine learning models tuned specifically to predict protein binding and ADME properties and all that are still in their infancy. One guy in a garage cannot possibly have the resources to build them himself. So he’s just asking publicly available models, and they’ll confidently give you very bad information that looks plausible to an idiot.
Everything about that thread demonstrates unearned confidence from a person who is probably pretty good at one of the things necessary to discover and develop a drug but cannot possibly be good at them all.
Well, unfortunately the only real test to treat AD is in a phase 3 human trial. We have animal models but the correlation to humans is, so far, very bad. Every time we’ve cured it in a mouse it hasn’t worked in people.
Can you expand on that in a way a person not in the field is able to understand?
Getting to IND-enabling studies requires a huge breadth of expertise. You need a lot of people because the breadth of experience required is beyond what one person can do. He’s substituted that with the plagiarism robot, and even if it worked, he’s in no position to be able to verify the output. He has to trust it, and the plagiarism robot is very confident but often wrong.
The whole “in his garage” thing is also ridiculous. The hard part of drug discovery isn’t making molecules, it’s making molecules that pass ADME and tox and can be formulated and show exposure and efficacy in non-rodent species. You can’t do all those things in a garage. Either he’s farmed that out to CRO’s or (worse) just showed his structure to the LLM and had it predict the test results.
But the main thing is that even the best machine learning models tuned specifically to predict protein binding and ADME properties and all that are still in their infancy. One guy in a garage cannot possibly have the resources to build them himself. So he’s just asking publicly available models, and they’ll confidently give you very bad information that looks plausible to an idiot.
Everything about that thread demonstrates unearned confidence from a person who is probably pretty good at one of the things necessary to discover and develop a drug but cannot possibly be good at them all.
Thank you for taking the time to explain it. It sounds like they need to publish so that their work can be checked.
Well, unfortunately the only real test to treat AD is in a phase 3 human trial. We have animal models but the correlation to humans is, so far, very bad. Every time we’ve cured it in a mouse it hasn’t worked in people.