Credit: Google News
Finding a new drug or innovative material traditionally has involved sophisticated guesswork, trying out many possible combinations and seeing what works—a cumbersome, time-consuming exercise of trial and error. Machine-learning algorithms are set to change that, drastically speeding up the development process for new pharmaceuticals.
The World Economic Forum included this use of artificial intelligence last year in its annual list of the top 10 emerging technologies that are set to be potentially disruptive over the next three to five years while providing significant benefits to economies and societies.
Artificial intelligence has the potential to reshape the nature of innovation and R&D, according to a 2017 paper by researchers at the National Bureau of Economic Research. Machine learning may be able to expand the set of problems that can be feasibly addressed through automation, lowering the costs of discovery across broad set of domains where classification and predictions play a major role. Molecular design is one such domain.
“AI is starting to increase the efficiency of both design and synthesis [in molecular design], making the enterprise faster, easier and cheaper while reducing chemical waste,” according to the WEF’s top 10 list for 2018.
Machine-learning algorithms can analyze past experiments that attempted to find new drugs and materials—whether or not they worked—and predict which new molecular structures are likely to fare best. The selected predictions can then be further validated by the researchers involved.
Pharmaceutical companies generally store millions of compounds to be screened as potential new drugs, a process that’s slow and yields relatively few hits. Moreover, these libraries collectively include a tiny fraction of the more than 10^30 theoretically possible molecules.
Machine-learning tools can not only rapidly search these libraries, but can also generate virtual repositories of new compounds that have similar properties to the most promising molecules. In addition, machine learning can flag those drugs and materials that look promising but may come with potentially harmful risks and side effects.
Protein structure prediction, which tries to project the three-dimensional shape of proteins in the human body, is one of the most important problems in drug design. Every two years, scientists enter a global competition to work on this puzzle.
At the latest gathering in Cancún, Mexico, in December, the contest wasn’t won by any of the academic research teams, but by DeepMind, the AI startup founded in 2010 and acquired by Google in 2014. DeepMind made news globally in 2016 after its AlphaGo program, based on advanced deep-learning algorithms, defeated one of the top professional players of Go, an ancient board game.
AI techniques can significantly speed up drug discovery by performing some of the tasks typically handled by scientists. “It is not that machines are going to replace chemists,” said drug discovery researcher and blogger Derek Lowe. “It’s that the chemists who use machines will replace those that don’t.”
Irving Wladawsky-Berger worked at
for 37 years, has been a strategic adviser to Citigroup, HBO and Mastercard and adjunct professor at Imperial College. He’s currently visiting lecturer and research affiliate at MIT, and is a regular contributor to CIO Journal.
Credit: Google News