How long would you last in school if you consistently received 12 percent out of 100s on your exams?
Not very long.
If you were a major league baseball player and only knocked out 1.2 hits for every ten times at bat, how long would you stay in the majors? Because the average batting average in the major leagues is about twice that, you might want to seriously consider getting into sports commentating.
However, if you are in charge of a pharmaceutical company you might be happy -- or, at least, satisfied -- with a 12 percent success rate. Currently, the overall clinical success of a drug is about 12 percent. For every 100 drugs that enter the process, only 12 make it to market.
What might not be evident in this figure, though, is the time -- in most cases, decades -- wasted on drugs that never come into fruition, as patients wait for help, and the money -- in the billions of dollars -- thrown at unsuccessful treatments.
In a study that was published in Nature Biotechnology, a group of scientists, led by Insilico Medicine, enlisted a robot ally to improve that drug discovery average, designing six novel inhibitors of DDR1, a kinase target implicated in fibrosis and other diseases, in 21 days, according to a news release. Four compounds were active in biochemical assays, and two were validated in cell-based assays.
The team added that the lead candidate was tested in mice with favorable results.
“This paper is a significant milestone in our journey towards AI-driven drug discovery. We work in generative chemistry since 2015 and when Insilico's and Alán's papers were published in 2016 everyone was very skeptical. Now, this technology is going mainstream and we are happy to see the models developed a few years ago and producing molecules against simpler targets being validated experimentally in animals. When integrated into comprehensive drug discovery pipelines, these models work for many target classes and we work with the leading biotechnology companies to push the limits of generative chemistry and generative biology even further,” said Alex Zhavoronkov, the founder and CEO of Insilico Medicine, said in the release.
Jürgen Schmidhuber, a professor at IDSIA, co-founder of NNAISENSE, and the original inventor of many core techniques and initial concepts in the field of artificial intelligence, was one of the many scientists who weighed in on the study. He suggests that this research is just the beginning of the massive transformation that AI will have on healthcare.
“This technology builds on our early work on adversarial and generative neural networks since 1990. Insilico has been working on generative models for drug discovery since 2015, and I am happy to see that their GENTRL system produced molecules that were experimentally validated in cells and in mice. AI will have a transformative effect on the pharmaceutical industry, and we need more experimental validation results to accelerate progress," said Schmidhuber.
"AI will have a transformative effect on the pharmaceutical industry, and we need more experimental validation results to accelerate progress." Jürgen Schmidhuber
To give an idea on just how transformative the team’s 21-day turnaround time may be, consider that the drug discovery process typically takes over 5 years and costs more than a billion dollars. That’s just the start. Companies then face another five years of clinical trials and heap another $1.5 billion on the pile for investment. The failure rates in drug discovery exceed 99 percent and in drug development, after the molecule passes animal testing, the failure rates exceed 90 percent.
If research like this can continue to make the drug discovery process not only more efficient, but more accurate, treatment costs would fall -- and more effective treatments would make it to patients much faster than the years it currently takes. With improved return-on-investments, pharmaceutical companies might be more willing to invest even more in drug discovery, creating a beneficial cycle of better treatments, more profitability and increased investments.
The team used Generative Adversarial Networks (GANs), which Zhavoronkov describes as a type of AI imagination. The technique is relied on a lot for image generation and recognition.
The researchers used six datasets: a large set of molecules derived from ZINC dataset; a dataset of known DDR1 kinase inhibitors; a dataset of common kinase inhibitors (positive set); a collection of molecules acting on non-kinase targets (negative set); patent data for biologically active molecules that have been claimed by pharmaceutical companies and 3D structures for DDR1 inhibitors. Before the data was used to train the AI, the researchers excluded gross outliers and reduced the number of compounds containing similar structures.