3 ms·

The pharmaceutical industry is in a drug-discovery slump. How much can AI help?

9 points by vo2maxer 7y ago

cde-v 7y ago
I wasn't aware the pharmaceutical industry was still trying to discover new drugs. Seems like they are more focused on marketing and keeping their drugs (or their drugs, but with slight modifications) under patent.
- mft_ 7y ago
  Interesting; can I ask what your opinion was based on? A quick Google gives a couple of sources; admittedly, the devil's usually in the details, but they should be enough to convince that there is significant amount being spent on research: https://www.bbc.com/news/business-28212223 https://www.bbc.com/news/business-28212223 https://www.raps.org/news-and-articles/news-articles/2019/7/do-biopharma-companies-really-spend-more-on-market https://www.raps.org/news-and-articles/news-articles/2019/7/...
Der_Einzige 7y ago
Better dimensionality reduction techniques are high up there. How do we get better than UMAP?
aaavl2821 7y ago
The oft cited $2.6B cost to develop a drug is derived from data from big pharma drug development programs. There is some evidence that startups are 5x more efficient at developing drugs [0] The last several years have seen large increases in the number of new FDA approved drugs. The number of FDA approved drugs developed by big pharma is roughly constant, while the number being developed by startups is increasing AI can certainly help improve R&D productivity, but it is far from the only lever. Better risk management, clinical trial design, advances in protein engineering, larger genomic datasets analyzed with "traditional" statistical methods, better disease models, new therapeutic modalities, etc have all contributed to more efficient R&D productivity over the last decade and will continue to do so. But these don't get much attention outside of the pharma industry [0] https://www.baybridgebio.com/blog/rd_bigpharma_startup.html https://www.baybridgebio.com/blog/rd_bigpharma_startup.html
- denzil_correa 7y ago
  > The oft cited $2.6B cost to develop a drug is derived from data from big pharma drug development programs. This number can not be validated as most of the information is not public. It's a self-reported number and should be treated as such. https://www.reuters.com/article/pharmaceuticals-tufts/corrected-tufts-says-average-new-drug-costs-2-6-bln-to-develop-critics-wary-idUSL2N0T82C620141118 https://www.reuters.com/article/pharmaceuticals-tufts/correc...
  - aaavl2821 7y ago
    I don't think that number is too crazy. I did my own back of the envelope analysis and got $1.5B in early stage r&d spend per approved drug discovered by big pharma. If you add a few hundred million for late stage r&d (big pharma spends about the same on late and early stage r&d) and then include cost of capital its in the ballpark of that number
denzil_correa 7y ago
> “In some cases, we might be allowed to pass over animal models and go straight to human testing once we show these drugs can hit their targets with no toxicity.” But those changes are probably many years away, he admits. He adds that it is wrong to imply that AI replaces scientists and conventional research—whereas AI supports and amplifies human efforts, it still depends on humans to generate novel biological insights, set research directions and priorities, guide and validate results, and produce needed data. This paragraph succinctly summarizes the current situation, future hope and effort required.

9 points by vo2maxer 7y ago

I wasn't aware the pharmaceutical industry was still trying to discover new drugs. Seems like they are more focused on marketing and keeping their drugs (or their drugs, but with slight modifications) under patent.

mft_ 7y ago
Interesting; can I ask what your opinion was based on? A quick Google gives a couple of sources; admittedly, the devil's usually in the details, but they should be enough to convince that there is significant amount being spent on research: https://www.bbc.com/news/business-28212223 https://www.bbc.com/news/business-28212223 https://www.raps.org/news-and-articles/news-articles/2019/7/do-biopharma-companies-really-spend-more-on-market https://www.raps.org/news-and-articles/news-articles/2019/7/...

Der_Einzige 7y ago

Better dimensionality reduction techniques are high up there. How do we get better than UMAP?

aaavl2821 7y ago

The oft cited $2.6B cost to develop a drug is derived from data from big pharma drug development programs. There is some evidence that startups are 5x more efficient at developing drugs [0] The last several years have seen large increases in the number of new FDA approved drugs. The number of FDA approved drugs developed by big pharma is roughly constant, while the number being developed by startups is increasing AI can certainly help improve R&D productivity, but it is far from the only lever. Better risk management, clinical trial design, advances in protein engineering, larger genomic datasets analyzed with "traditional" statistical methods, better disease models, new therapeutic modalities, etc have all contributed to more efficient R&D productivity over the last decade and will continue to do so. But these don't get much attention outside of the pharma industry [0] https://www.baybridgebio.com/blog/rd_bigpharma_startup.html https://www.baybridgebio.com/blog/rd_bigpharma_startup.html

denzil_correa 7y ago
> The oft cited $2.6B cost to develop a drug is derived from data from big pharma drug development programs. This number can not be validated as most of the information is not public. It's a self-reported number and should be treated as such. https://www.reuters.com/article/pharmaceuticals-tufts/corrected-tufts-says-average-new-drug-costs-2-6-bln-to-develop-critics-wary-idUSL2N0T82C620141118 https://www.reuters.com/article/pharmaceuticals-tufts/correc...
- aaavl2821 7y ago
  I don't think that number is too crazy. I did my own back of the envelope analysis and got $1.5B in early stage r&d spend per approved drug discovered by big pharma. If you add a few hundred million for late stage r&d (big pharma spends about the same on late and early stage r&d) and then include cost of capital its in the ballpark of that number

denzil_correa 7y ago

> “In some cases, we might be allowed to pass over animal models and go straight to human testing once we show these drugs can hit their targets with no toxicity.” But those changes are probably many years away, he admits. He adds that it is wrong to imply that AI replaces scientists and conventional research—whereas AI supports and amplifies human efforts, it still depends on humans to generate novel biological insights, set research directions and priorities, guide and validate results, and produce needed data. This paragraph succinctly summarizes the current situation, future hope and effort required.