Ge­nius Drugs From Dumb Sil­i­con

Can a gi­ant pile of data beat hu­man ex­per­tise in the de­sign of mir­a­cle drugs? Daphne Koller may come up with a sur­pris­ing an­swer.

Forbes - - INSIDE - By Jil­lian D’On­fro

Can a gi­ant pile of data beat hu­man ex­per­tise in the de­sign of mir­a­cle drugs? Daphne Koller may come up with a sur­pris­ing an­swer.

NNot many sci­en­tists get so­licited for photo ops, but for Daphne Koller it’s a reg­u­lar oc­cur­rence. “It hap­pens at pretty much any event that has tech peo­ple,” Koller says when asked about one re­cent snap­shot. “It’s a lit­tle awk­ward. It’s not like I feel like this is some­thing I de­serve.”

Selfie re­quests are just one sign of Koller’s star­dom, earned from more than 20 years bridg­ing com­puter sci­ence, bi­ol­ogy and ed­u­ca­tion. She chalked up a string of ac­co­lades along the way: get­ting a mas­ter’s de­gree from Jerusalem’s He­brew Univer­sity at 18; be­com­ing a Stan­ford Univer­sity pro­fes­sor fo­cused on ma­chine learn­ing at 26; win­ning, nearly a decade later, a MacArthur “ge­nius grant” for re­search that com­bined ar­ti­fi­cial in­tel­li­gence and ge­nomics; co­found­ing $1 bil­lion (val­u­a­tion) Cours­era, an early plat­form to let peo­ple around the world take univer­sity classes for free.

The next act for this 51-year-old in­no­va­tor: In­sitro, a firm in South San Fran­cisco that aims to find new drugs by sort­ing through masses of data. If it suc­ceeds, it will have over­turned how drugs get dis­cov­ered.

Lab bi­ol­o­gists typ­i­cally fo­cus on a few spe­cific pro­teins as drug tar­gets. If those fail, data sci­en­tists make sug­ges­tions for oth­ers to try. In­sitro, on the other hand, wants to col­lect much more data be­fore the bi­ol­o­gists go off on their hunt. It will lever­age ad­vances in bio­engi­neer­ing (such as Crispr gene edit­ing) and in soft­ware that en­ables com­put­ers to see things that es­cape hu­mans.

Koller de­scribes her aha mo­ment this way: “Ma­chine learn­ing is now do­ing amaz­ing things if you give it enough data. We finally have the op­por­tu­nity to cre­ate bi­o­log­i­cal data at scale.”

In­sitro’s com­pu­ta­tional ex­perts and bi­ol­o­gists work to­gether to cre­ate lab ex­per­i­ments to pro­duce mas­sive cus­tom data sets. Ma­chine learn­ing mod­els then find pat­terns to sug­gest new tests and po­ten­tial ther­a­pies. Ro­bot­ics like au­to­mated pipet­ting ma­chines re­duce hu­man er­ror. With all this, In­sitro can do “ex­per­i­ments in a mat­ter of weeks in­stead of years,” Koller says.

AI plus bi­ol­ogy, her back­ground, was a “mar­riage made in heaven” for in­vestors, she says. Within six months Koller raised $100 mil­lion from ARCH Ven­tures, An­dreessen Horowitz, Fore­site Cap­i­tal, Al­pha­bet’s ven­ture fund GV and Third Rock, with Jeff Be­zos and oth­ers join­ing later. In April, she landed a deal with Gilead Sciences that gives In­sitro $15 mil­lion now with $1 bil­lion to fol­low if it helps find a treat­ment for a deadly form of non­al­co­holic fatty liver dis­ease. The dis­ease is ex­pected to soon be­come the lead­ing cause of liver trans­plants.

“There are very few in­di­vid­u­als who un­der­stand both sides of the beast,” says Mani Subra­ma­nian, who heads liver dis­ease clin­i­cal re­search at Gilead. “The bi­ol­ogy as well as the deep learn­ing.”

In­sitro’s fu­ture pay­outs from Gilead hang on whether it can iden­tify five pro­teins that could be tar­gets for drugs and then whether tar­get­ing those pro­teins leads to ap­proved ther­a­pies for the liver dis­ease. The con­tin­gent pay­ments, which in­clude rev­enue shar­ing from suc­cess­ful drugs, helped In­sitro earn a spot on Forbes’ in­au­gu­ral AI 50 list of the most promis­ing ar­ti­fi­cial in­tel­li­gence com­pa­nies.

More than 20 other star­tups are chas­ing the dream of faster, cheaper drug dis­cov­ery through AI. Among them are No­table Labs, with $55 mil­lion of ven­ture cap­i­tal, and Verge Ge­nomics, with $36 mil­lion. No­var­tis has an­nounced a five-year AI col­lab­o­ra­tion with Mi­crosoft, and Merck and GSK have startup part­ner­ships as well.

Ar­ti­fi­cial in­tel­li­gence does not make bi­ol­ogy easy. “I don’t think the plat­form can be magic,” Koller says.

Be­fore In­sitro can reap re­wards, a few hun­dred thou­sand lab tests need to hap­pen. Koller has the en­ergy. Bounc­ing around In­sitro’s of­fice—she gave away her desk chair to one of her 53 em­ploy­ees be­cause she never used it—she moves from a room named Macrophage (a white blood cell) to one named Elas­tic Net (a data-mod­el­ing tech­nique) to show off the lat­est lab equip­ment.

Big Pharma’s in­ter­est would seem to make In­sitro a likely ac­qui­si­tion tar­get if it hits pay dirt. But Koller says she doesn’t want to see In­sitro “swal­lowed into the maw” of a larger or­ga­ni­za­tion. She wants it to make its own branded drugs.

The ul­ti­mate goal is that the peo­ple ask­ing for photos ops will be health­ier thanks to In­sitro. Koller says she hopes they come up to her and say, “Be­cause of you, I have my life back.”

“DATA TRUMPS EV­ERY­THING.” —Josh Estelle, a lead en­gi­neer for Google Trans­late

Newspapers in English

Newspapers from USA

© PressReader. All rights reserved.