class="knockout" style="margin-top: 0px;margin-bottom: 1px;font-size: 14px;line-height: 1.2;padding: 5px;padding-left: 20px;padding-right: 20px;font-weight: normal;">THANK YOU FOR SUBSCRIBING
First, there are about a hundred companies who claim they are utilizing AI for drug discovery. About a dozen, including Cloud Pharmaceuticals, are established companies. Upon close examination, all of these companies are doing something different; either they are focusing on different drugs or are applying AI to different aspects of drug discovery. Many of them are “giving away” their services in an effort to establish themselves that has resulted in what appears to be “tire kicking” by big pharma companies.
Before answering the question posed above, let us address a few important points. First, the product of the pharmaceutical industry is still a drug. AI is a tool, not a product. For the most part, AI is not proprietary. Some things work, and many do not. What is proprietary about AI is some knowledge about things that have worked in the past.
There is not “AI Drug Discovery” business or market. There is a pharmaceutical and biotechnology market that AI can make more efficient. The market for “services,” that is small companies giving AI discovery services to pharma leaders is limited and almost nonexistent. The deals that have been done are one-offs, often big pharma companies trying to learn. We have observed about a dozen big pharma companies starting to build 100+ people AI groups internally now. The value of AI in pharma lies in the data, not in the AI, and big pharma owns that data if it exists, and they know it.
So what works and what does not for small AI drug discovery companies? The companies that have the most funding are those with drugs in the clinic. They are getting funded for their drugs. All of the dozen established companies have raised seed and startup money from angels and VCs, but few have received large amounts unless they are in the clinic. One exception is a company, designing pesticides for Monsanto, that obtained a corporate partnership with some capital and in-kind investment.
And here are a few things that do not work. Using pure machine learning to design molecules is a “hit or miss” proposition at best. No data set spans molecular space. The claims that have been made and the papers in this area are anecdotal and not rigorous evidence.
"What is proprietary about AI is some knowledge about things that have worked in the past"
Another thing that does not work or add credibility to the AI discovery firm is to announce they can get “billions of dollars in milestones” when they have not even started discovery yet. Maybe 1 in 100 discovery programs makes it, and that requires raising significant capital and diluting the milestones. And even when pharma companies promise such, the probability that they go forward with a program is low. These future milestones are years away and elusive at best, and not a credible statement to current valuation, nor a sign that the small company knows what they are doing. And yet another thing that does not work is replacing drug discovery scientists with PhD-holder computer scientists (CS). The AI is a tool in support of discovery, and the CS majors are not taking over the industry.
So what does work? Using AI to statistically improve the odds of a project can work. Using AI to add knowledge to an already existing body of knowledge can improve odds. Using AI to predict failures can reduce ensemble cost. Using AI to discover drugs works as long as you do not throw out traditional research and methods to do so.
Cloud Pharmaceuticals has moved in two directions with its AI drug discovery platform. First, its discovery platform uses AI to add value to traditional advanced computational chemistry, not replace it. We have some 25 target lead pairs ready for funding. We have chosen to fund 5 of them, including a rare disease project ready for the clinic and four candidate stage preclinical projects. This is a lot, and they are being done with partners to increase volume and mitigate discovery and development risk.
The second thing we are doing is partnering with a major corporation, not ready to be announced, to build a massive database of inhibitors and agonist candidates for the entire druggable genome with plans to offer it for free for royalties only, as a means of displacing the entire HTS and screening markets.
There are no magic bullets. Success requires good data, good algorithms, good engineers and scientists, and a lot of common sense, and usually humility and not arrogance. AI for drug discovery is in its infancy, but it is a tool and not a market, and that will not change.
See Also :