The Ar­ti­fi­cial Mir­a­cle

Newsweek - - NEWS - ARVIND DILAWAR

four out of 10—that’s how

many Amer­i­cans the Na­tional Cancer In­sti­tute es­ti­mates will be di­ag­nosed with cancer at some point. While 33 per­cent of those pa­tients won’t live longer than five years, giv­ing them pre­cious lit­tle time to find ef­fec­tive treat­ments, it takes over a decade to bring new cancer drugs to mar­ket. The process in­volves an­i­mal test­ing, hu­man tri­als and reg­u­la­tory re­view—a gant­let through which less than 7 per­cent of ex­per­i­men­tal medicines suc­cess­fully pass. Is it any won­der, then, that there are less than 2,000 Food and Drug Ad­min­is­tra­tion-ap­proved phar­ma­ceu­ti­cals on the mar­ket? Not 2,000 cancer treat­ments—2,000 drugs for all dis­eases.

In­sil­ico Medicine, a Bal­ti­more­based biotech re­search com­pany, hopes to rev­o­lu­tion­ize drug de­vel­op­ment by slash­ing the time nec­es­sary for re­search with the help of ar­ti­fi­cial in­tel­li­gence (AI). In a study pub­lished in the med­i­cal jour­nal

On­co­tar­get, a team led by In­sil­ico Medicine de­tails their ap­proach. Es­sen­tially, re­searchers built two com­puter net­works (to­gether known as gen­er­a­tive ad­ver­sar­ial net­works, or GANS). One sug­gests new mol­e­cules that may have cancer-fight­ing prop­er­ties; the other elim­i­nates those sug­ges­tions based on known treat­ments. “It’s bet­ter to ex­plain with an anal­ogy from art,” says Polina Mamoshina, a re­search sci­en­tist at In­sil­ico Medicine. If cancer drugs were works of art, she says, the first net­work would be an art stu­dent at­tempt­ing to copy them, and the sec­ond net­work would be an art ex­pert flag­ging forg­eries. Each time the stu­dent’s work gets called out as a forgery, the stu­dent must get bet­ter at copy­ing the orig­i­nal; each time the stu­dent’s work gets bet­ter, the ex­pert must work harder at spot­ting forg­eries.

Re­lat­ing that back to GANS, as the first net­work keeps try­ing to “trick” the lat­ter into ac­cept­ing new mol­e­cules as le­git­i­mate drugs, both bet­ter learn what cancer treat­ments should look like. Once they’re through test­ing each other, the net­works can be used to vet com­pounds for their cancer-fight­ing po­ten­tials. In this way, the team from In­sil­ico Medicine screened 72 mil­lion chem­i­cals from a pub­lic data­base. Among the com­pounds se­lected by the GANS were 60 patented cancer treat­ments—mean­ing that the net­works were able to ac­cu­rately iden­tify th­ese medicines and that the other com­pounds they se­lected were likely wor­thy of fur­ther study.

Com­pared with stan­dard in vitro (test tube) ex­per­i­men­ta­tion, this

in sil­ico (com­puter-tested) method is ex­po­nen­tially faster. In­stead of be­gin­ning the search for a new cancer treat­ment with a mil­lion com­pounds that have po­ten­tial, re­searchers could, in just one month, nar­row the pool of can­di­dates to the 100 most promis­ing leads.

This ap­proach not only fos­ters faster drug de­vel­op­ment but is more cost-ef­fec­tive re­search too. Each ex­per­i­men­tal medicine that fails to make it through the de­vel­op­ment process is a loss of sev­eral mil­lion dol­lars’ worth of la­bor and re­sources. A study from the Jour­nal of Health

Eco­nomics es­ti­mates that the costs as­so­ci­ated with failed drugs adds more than $1.6 bil­lion to the cost of each suc­cess­ful one. With fewer, more promis­ing leads, re­searchers could save mil­lions, per­haps bil­lions.

But not ev­ery­one is con­fi­dent about the ap­pli­ca­tions of in sil­ico test­ing. Mamoshina ac­knowl­edges that many cancer re­searchers who work with more tra­di­tional bi­o­log­i­cal and chem­i­cal meth­ods are un­fa­mil­iar with AI, which can breed doubt. “To them, it’s a black box,” she says. “It’s re­ally chal­leng­ing to un­der­stand, which is why they’re re­ally skep­ti­cal.”

As with other cut­ting-edge tech­nolo­gies, hype may also be fu­el­ing the progress in—and set­ting the po­ten­tial pit­falls for—in­sil­ico Medicine. Olexandr Isayev, an as­sis­tant pro­fes­sor at the Univer­sity of North Carolina whose lab fo­cuses on de­vel­op­ing meth­ods of Ai-as­sisted drug dis­cov­ery, ac­knowl­edges that there may be too much ex­cite­ment over a tech­nol­ogy that has yet to pro­vide any ma­te­rial re­sults. “Most pub­lished pa­pers, in­clud­ing this one, are purely com­pu­ta­tional,” he says. “Un­for­tu­nately, some pre­dic­tions could be wrong. I re­ally would like to see the first ex­per­i­men­tal con­fir­ma­tion of ‘Ai-dis­cov­ered’ mol­e­cules.”

As would In­sil­ico Medicine, which con­tin­ues to de­velop GANS. Rather than li­cens­ing out the tech­nol­ogy in a sort of soft­ware-as-a-ser­vice model, the com­pany is ex­pand­ing re­search into the mol­e­cules the net­works have iden­ti­fied as hav­ing cancer-fight­ing po­ten­tial. Once th­ese com­pounds pass through tra­di­tional in vitro test­ing, they will be li­censed out to phar­ma­ceu­ti­cal com­pa­nies for fur­ther reg­u­la­tory re­view and, if all goes well, mar­ket­ing. This past Au­gust, it was an­nounced that In­sil­ico Medicine is part­ner­ing with phar­ma­ceu­ti­cal giant Glax­o­smithk­line to be­gin im­ple­ment­ing some of its new re­search tech­niques.

In­sil­ico Medicine’s be­lief in this ap­proach is re­flected in its de­ci­sion to li­cense the drugs it dis­cov­ers, rather than the tools of dis­cov­ery them­selves. Yet for the com­pany to prove that AI can in fact elim­i­nate the guess­work in­volved in early drug dis­cov­ery, In­sil­ico Medicine will have to head back to the lab—and the test tubes.

This ap­proach not only fos­ters faster drug de­vel­op­ment but is more cost­ef­fec­tive re­search too.

FIGHT­ING SMART In­sil­ico Medicine is ex­pand­ing its re­search into the olec les its I net or s ha e identi ed as ha ing cancer ght­ing po­ten­tial

Newspapers in English

Newspapers from USA

© PressReader. All rights reserved.