AI in Medicine
One of the promises of artificial intelligence, both in science fiction and in the real world, has always been the promise of better health, or at least leaning us more towards prevention on the treatment versus prevention scale that is the core of our health care system. Though the latest breakthrough in AI falls short of that progressive promise in terms of prevention, it does much to advance the treatment side of things. Researchers at the Massachusetts Institute of Technology have taken advantage of a type of artificial intelligence, known as deep-learning, to create a designer anti-biotic that appears to be able to kill many strains of human-toxic and drug-resistant bacteria. With any luck, using AI to model and manufacture medicines, though not fully a practicable reality, could end up fulfilling a little bit of that promise of prevention.
For their study, researchers Stokes, Barzilay and Collins utilized an entirely digital approach, preferring to go “in silico” for the design and testing phases.i The Computer Science and Artificial Intelligence Laboratory at MIT, or CSAIL, was the perfect setting for these faculty and post-doc students. In an eternal search for novel compounds, the researchers turned to a not-so-new method of predictive modeling, only this time, they let artificial neurological networks do the processing. Once programmed, they set the neural net to work on examples of drugs that are known to be effective at killing various types of bacteria and for the dataset, they set the digital brain to consider 2,500 different molecular combinations. About 1,700 of the combinations consisted of federally approved drugs with another 800 consisting of natural products with a wide spectrum of biological interactions.
After feeding the system the initial dataset of drugs and chemicals known to work, they introduced a library of about 6,000 compounds and let the AI’s digital mind start to wander. After the immense number-crunching session was over, the deep learning machine came back with one molecule it thought would be good at killing bacteria. The researchers ran the results through another computer model that tests for human toxicities and it passed a second time. The drug was initially developed as a diabetes medication, but now has been found to kill all but a hard-to-treat lung bacteria that likes standing water, named Pseudomonas aeruginosa. The unique shape of the molecule lends itself to disrupting the outer electrochemical balance of a cell’s membrane, breaking down essential cellular functions and destroying the cell. But this was not the end of their deep-learning experience.
After their success with what they came to name halicin, after the AI in Stanley Kubrick’s 2001 A Space Odyssey, HAL, the researchers set their still-hungry digital brain on a much larger library of molecules. The ZINC15 database is an online collection of more than a billion chemical compounds. The artificial intelligence behind their work digested the new input in only three days, identifying a possible 23 non-toxic anti-biotic-like chemicals. Taking those 23 to task, resulted in eight compounds that treated antibiotic resistant bacterial infections, with two of the eight registering as highly effective. What’s more, their system also works to optimize existing compounds.
Their discovery is so new, it’s still being reviewed, and funded, by institutions from around the world. These kinds of breakthroughs and more are what the SUNY Polytechnic Institute hopes to bring to the table with their innovative Nanobioscience Graduate Programsii iii at their Albany CNSE Campus. With a robust research and innovation community in the SUNY System supporting technology transfer,iv the next greatest breakthrough in prevention and treatment could be just around the corner.
i http://imes.mit.edu/artificial-intelligence-yields-new-antibiotic/
ii https://sunypoly.edu/academics/majors-and-programs/nanobioscience-ms.html
iii https://sunypoly.edu/academics/majors-and-programs/phd-nanobioscience.html
