3 ms·

I wonder if it's possible to use machine learning to speed up this process. Rather than doing computational biology, can we use a hybrid system? For instance, c

I wonder if it's possible to use machine learning to speed up this process. Rather than doing computational biology, can we use a hybrid system? For instance, can we test this on mice and then feed some sort of parameter into a model to see the probability it working on humans rather than testing it directly on humans. I'm sure this won't be a perfect model but I wonder if it can raise the 14% to 20% while using less resources.

aaavl2821 8y ago
ML (or anything really) that could predict which molecules would work in humans without testing them in humans would probably be on of the most important advances in medicine ever, and would massively lower drug costs and open up new areas for drug research Phase 2 failure -- ie stuff failing in humans after working in mice -- is the biggest cost driver in drug dev. You spend $50-100M+ to get there and 35% of things fail. If you could eliminate you'd cut a massive chunk of the cost and risk of drug dev However I don't think ML is the thing to study if you want to create ML to predict whether drugs will work in people. There is not enough data about how the human body works to develop a good model. Better to work in biology to develop good models of disease, or in developing tools to better measure the molecular biology of the living human body. It's unclear if predicting drug effectiveness with ML will be possible in our generation For more near term applications of ML in drug discovery and development see here: https://www.getrevue.co/profile/nathanbenaich/issues/6-impactful-applications-of-ai-to-the-life-sciences-new-essay-150757 https://www.getrevue.co/profile/nathanbenaich/issues/6-impac...
zinclozenge 8y ago
The short answer is "probably". The use of AI in drug design is only just now starting, we'll see the results in about 10 years. I optimistically expect AI to be able to shave off ~2-3 years of the drug lifecycle. edit: but it won't do much to optimize P2 trial success specifically, it'll simply help iterate substantially faster in the pre-P2 trial phases.

aaavl2821 8y ago

ML (or anything really) that could predict which molecules would work in humans without testing them in humans would probably be on of the most important advances in medicine ever, and would massively lower drug costs and open up new areas for drug research Phase 2 failure -- ie stuff failing in humans after working in mice -- is the biggest cost driver in drug dev. You spend $50-100M+ to get there and 35% of things fail. If you could eliminate you'd cut a massive chunk of the cost and risk of drug dev However I don't think ML is the thing to study if you want to create ML to predict whether drugs will work in people. There is not enough data about how the human body works to develop a good model. Better to work in biology to develop good models of disease, or in developing tools to better measure the molecular biology of the living human body. It's unclear if predicting drug effectiveness with ML will be possible in our generation For more near term applications of ML in drug discovery and development see here: https://www.getrevue.co/profile/nathanbenaich/issues/6-impactful-applications-of-ai-to-the-life-sciences-new-essay-150757 https://www.getrevue.co/profile/nathanbenaich/issues/6-impac...

zinclozenge 8y ago

The short answer is "probably". The use of AI in drug design is only just now starting, we'll see the results in about 10 years. I optimistically expect AI to be able to shave off ~2-3 years of the drug lifecycle. edit: but it won't do much to optimize P2 trial success specifically, it'll simply help iterate substantially faster in the pre-P2 trial phases.