DeepMind Spin-off Aims To Halve Drug Discovery Times Following Big Pharma Deals 25
The head of Google DeepMind believes its drug discovery spinout will halve the time taken to find new medicines, attracting the attention of the world's biggest pharmaceutical companies which are looking to artificial intelligence to revolutionise the lengthy process. From a report: Speaking to the Financial Times, Demis Hassabis, who co-founded Google's AI unit and also leads the drugs offshoot Isomorphic Labs, said the goal was to reduce the discovery stage -- when potential drugs are identified before clinical trials -- from the average of five years to two. "I think that would be success for us and be very meaningful," he said.
Hassabis stated the goal days after announcing Isomorphic Lab's first two pharmaceutical partnerships with Eli Lilly and Novartis, which came to a combined value of up to $3bn, in deals set to transform the finances of the unprofitable group. Isomorphic Labs uses an AI platform to predict biochemical structures, which aids the creation of new drugs by recommending which potential compounds will have the desired impact in the body. Including clinical trials, it often takes up to a decade to discover and develop a new drug, costing on average about $2.7bn, according to research by the Tufts Center for the Study of Drug Development.
Large drugmakers, under pressure to fill their pipelines with new potential medicines while existing ones face patent cliffs, when they will face far cheaper generic competition, are eager for new ways to shorten the process. As healthcare systems around the world put pressure on drug prices, pharma companies are also looking for ways to cut costs in research and development. Hassabis said that many drugmakers had also been eager to partner with Isomorphic but the company wanted to focus on collaborations that could improve its technology.
Hassabis stated the goal days after announcing Isomorphic Lab's first two pharmaceutical partnerships with Eli Lilly and Novartis, which came to a combined value of up to $3bn, in deals set to transform the finances of the unprofitable group. Isomorphic Labs uses an AI platform to predict biochemical structures, which aids the creation of new drugs by recommending which potential compounds will have the desired impact in the body. Including clinical trials, it often takes up to a decade to discover and develop a new drug, costing on average about $2.7bn, according to research by the Tufts Center for the Study of Drug Development.
Large drugmakers, under pressure to fill their pipelines with new potential medicines while existing ones face patent cliffs, when they will face far cheaper generic competition, are eager for new ways to shorten the process. As healthcare systems around the world put pressure on drug prices, pharma companies are also looking for ways to cut costs in research and development. Hassabis said that many drugmakers had also been eager to partner with Isomorphic but the company wanted to focus on collaborations that could improve its technology.
Concerning (Score:1)
Re: (Score:2)
Tell us you don't know how drug trials work without telling us you don't know how drug trials work.
Re: (Score:3)
Re: (Score:2)
Re: (Score:2)
Re: (Score:2)
Drug discovery is heuristic search. AI is a powerful heuristic. Of course most of what it thinks 'might work' will not, but it will often be better than the status quo of throwing everything at the wall and seeing what sticks.
Re: (Score:2)
Re: (Score:2)
" I fully expect we will hear about AI hallucinating a completely ineffective drug and it getting railroaded through trial "
Natural Intelligence has been doing this very successfully for years.
Re: (Score:2)
Considering that supposedly independent labs running pharmaceutical trials are fully captured by the Big Pharma, considering that the revolving door made regulators ineffective, the AI 'research' is yet another step toward removing checks and controls. I fully expect we will hear about AI hallucinating a completely ineffective drug and it getting railroaded through trial based on hype very shortly.
I just don't understand the basis for this assumption. The nice thing about AI even if it is sometimes or even mostly wrong is you can just sit it in a corner and let it bang its head against reality all day long until it gets lucky.
Drug candidates are increasingly evaluated 'in silico' early in the process. Whether it is something a flawed ignorant human dreams up or a flawed ignorant AI... does it really matter? Passing through automated and human evaluation steps before animal and clinical trials is t
Re: (Score:2)
I just don't see why the modality of discovery matters.
In research your reputation matters a lot. As such, a lot of early "no go" decisions are made by people that have a decision making horizon longer than the next quarter. This is also why a lot of whistle-blowing happens. Considering how captured everything else is, this might be the last effective check in place.
pink unikornz (Score:1)
Masking symptoms (Score:1)
So far as I can tell, most of our medical research doesn't address fundamental causes and fixes, it focuses on masking symptoms.
Not completely. Chemotherapy comes to mind, where any bits of cancer in the body are killed off and the person will be cancer-free afterwards.
Just about all medical studies I've seen are phrased in terms of "reducing symptoms", as in "after 7 weeks the test subjects reported a significant reduction in symptoms compared to the control group".
Most of the people I know are on medicati
Re: (Score:3)
Re: Masking symptoms (Score:2)
Check out gene therapy drugs. For example the drug Zolgensma has a price tag of $2.1 million. For *one* IV treatment. But, it literally cures the disease SMA which was previously impossible. The previous "treatment" cost $350,000/year for the rest of your life.
While the costs are insane (Hemgenix is $3.5 million) gene therapy feels like absolute magic. One treatment and you're done. Much better than needing periodic blood transfusions for the rest of your life.
reducing failure rates? (Score:2)
A non-paywall version of the article; https://archive.ph/l4LU3 [archive.ph]
Isomorphic Labs appears to be aimed at accelerating discovery of drug candidates which is great, but I'm wondering if it produces better quality candidates.
"Ninety percent of clinical drug development fails despite implementation of many successful strategies"
"40%–50% of clinical failure of drug development is due to lack of clinical efficacy"
https://www.ncbi.nlm.nih.gov/p... [nih.gov]
Drug that are effective have to find the right balance between eff
Re: (Score:1)
This article is not surprisingly light on details. It suggests an effort to expedite the process for a potential drug to progress from an initial screening hit, through the stages of refining medicinal chemistry properties during lead optimization, and ultimately becoming a drug candidate ready for preclinical trials. However, this accelerated process does not eliminate the necessity for preclinical trials to demonstrate safety and efficacy before proceeding to clinical trials.
Re: (Score:1)
I didn't realize you can't edit posts anymore. Regardless, you hit the main question that hasn't been answered. Will this produce better canidates. Drugs have a binding site. Can these AI methods determine the exact amino acid residues that are critical in the binding site and provide good suggestion on what to modify. Currently it is a semi-manual and computationally brute force method. Could AI take all the existing structure activity relationships we generated to better design a better drug. In theory ye
Re: (Score:2)
Very interesting, thanks! It sounds like you actually know something about this.
"Could AI take all the existing structure activity relationships we generated to better design a better drug"
It seems likely that they ingested everything they could get their hands on. In principle they can do a better and quicker job of finding matches between substances and known binding sites. And maybe designing novel compounds that bind well. After all, they did already "predict the 3D structures of nearly every catalogued
AI instructions (Score:2)
AI: flag all of our drugs that will go out of patent in the next year, and give us a list of drugs that do the same, not better, that will be under patent for at least five more years.
You think I'm making this up? A few years ago India refused a patent from a major drug company, and in the decision said that it was absolutely proven to be no better than the one going out of patent, and therefore was not a new and improved drug, suitable for patent.
Newer, cheaper drugs! (Score:2)
And then no patents awarded to AI (Score:2)
So... then all these drug discoveries will be patent free? Because other areas say that AI isn't patentable/copyrightable. So....
Embarrasing mistakes (Score:2)
Look at the contacts section on their website. ALL of their email contacts are invalid. Was the website created by AI as well?
Need more duplicate drugs! (Score:2)
Most "new" drugs are minor tweaks to existing drugs which are created so that they can extend patents and bilk patients through obscenely high prices.