AI- A prom­ise for im­prov­ing ef­fi­ciency and ac­cu­racy in drug dis­cov­ery

In a highly com­pet­i­tive busi­ness en­vi­ron­ment, the first com­pany to patent a new chem­i­cal en­tity has all the ad­van­tage, and hence, the idea of fast track­ing the drug dis­cov­ery process us­ing artificial in­tel­li­gence (AI) seems promis­ing. While adop­tion of AI

BioSpectrum (Asia) - - Bio Content - Piyush Bansal, Se­nior In­dus­try An­a­lyst, Trans­for­ma­tional Health, Frost & Sul­li­van

Apast few years have been rev­o­lu­tion­ary for phar­ma­ceu­ti­cal in­dus­try. We have seen a sig­nif­i­cant trans­for­ma­tion in phar­ma­ceu­ti­cal re­search and de­vel­op­ment prac­tice, which is largely driven by in­creas­ing im­por­tance of pa­tient cen­tric­ity, value based care and fo­cus on re­duc­ing drug de­vel­op­ment cost.

10 years back, with lim­ited dig­i­tal­iza­tion, data avail­abil­ity was a chal­lenge. How­ever, in to­day’s world, we see that the health­care sec­tor is sit­ting on a data gold­mine and con­se­quently we are see­ing an up­trend of uti­liz­ing big data sci­ence in drug dis­cov­ery and clin­i­cal trial ac­tiv­i­ties. As per IBM, health­care In­dus­try gen­er­ates 2.5 Quin­til­lion bytes of data ev­ery­day in form of pa­tient records, di­ag­nos­tics records, health­care records and other sci­en­tific data­bases. How­ever, the im­por­tant part is—How do we uti­lize this in­for­ma­tion? So far we have seen that not ev­ery pa­tient is same; not ev­ery ther­apy is be­hav­ing the same way for ev­ery pa­tient, and hence com­pa­nies are look­ing to uti­lize pa­tient fo­cused in­sights to de­velop drug. How­ever, it is al­most im­pos­si­ble to han­dle, man­age and an­a­lyse this vol­ume of data man­u­ally, and that’s where the con­cept of Artificial In­tel­li­gence (AI) comes into pic­ture.

What is artificial in­tel­li­gence (AI)?

AI is not a new con­cept, in fact, a large part of the­o­ret­i­cal and tech­no­log­i­cal con­cepts was de­vel­oped over the past 60 years. How­ever re­cent ad­vance­ments

in ma­chine learn­ing and deep learn­ing tech­nolo­gies have de­vel­oped the prac­ti­cal ap­pli­ca­tion as­pects and there­fore, com­pa­nies are in­creas­ingly look­ing to uti­lize AI in their op­er­a­tions.

AI refers to a set of tech­nolo­gies which are con­verged to sense, un­der­stand and act with the abil­ity to learn from ex­pe­ri­ence and adapt over time. AI in­cludes big data an­a­lyt­ics, nat­u­ral lan­guage pro­cess­ing (for sup­port­ing trans­la­tion, pat­tern recog­ni­tion, vis­ual per­cep­tion and de­ci­sion mak­ing), and ma­chine learn­ing (for de­vel­op­ing com­pu­ta­tional ap­proaches to au­to­mat­i­cally make sense of data).

Need for in­volv­ing AI in drug dis­cov­ery

The ef­fi­ciency of drug dis­cov­ery and trial process— de­fined as num­ber of drugs ap­proved vs. to­tal R&D bud­get—has con­tin­u­ously de­clined over the past few years. The av­er­age cost of drug dis­cov­ery and de­vel­op­ment comes out to be more than $2.5 bil­lion and it may take more than 10 years for com­pa­nies to come up with suc­cess­ful can­di­dates. In process of rel­e­vant com­pound iden­ti­fi­ca­tion, sci­en­tists and re­searchers scroll through thou­sands of re­search pa­pers, pub­lished lit­er­a­ture, patents, and disease data­bases, to un­der­stand re­la­tion­ships be­tween bi­o­log­i­cal en­ti­ties such as genes, symp­toms, dis­eases, pro­teins, tis­sues, species and can­di­date drugs, which is a time con­sum­ing process. On top of it, a large chunk of in­for­ma­tion is be­ing up­dated on a daily ba­sis. In fact, more than 10,000 med­i­cal re­search pa­pers and lit­er­a­ture get pub­lished ev­ery­day. Con­sid­er­ing the ef­fort, it is a hu­mon­gous task, and de­spite this, there is no guar­an­tee that this iden­ti­fied and short­listed com­pound will be suc­cess­ful in later stages. In fact, 90 per cent of these short­listed com­pounds fail to reach the ap­proval stage, and spend on these fail­ure com­pounds ac­count for about 70–75 per cent of the to­tal drug de­vel­op­ment cost.

Given this busi­ness sce­nario and in­creas­ing cost pres­sure on phar­ma­ceu­ti­cal com­pa­nies, re­searchers and sci­en­tist com­mu­nity is hop­ing to com­bine the ex­ist­ing knowl­edge from pre­vi­ous drug dis­cov­ery projects with new and ex­ist­ing ex­per­i­men­tal data, to drive AI-based drug dis­cov­ery and de­sign process.

AI is ex­pected to stream­line and speed-up some key drug dis­cov­ery ac­tiv­i­ties in­clud­ing iden­tify new drug com­pounds for screen­ing dur­ing the early stages of drug dis­cov­ery; un­der­stand and study new ther­a­peu­tics ap­pli­ca­tions for pre­vi­ously tested com­pounds, im­prove new com­pound de­sign process and sup­port de­vel­op­ment of pa­tient-spe­cific ther­a­peu­tics. Fur­ther, an ex­pected higher ac­cu­racy from au­to­mated sys­tems will likely re­duce the late stage fail­ure rate. Con­sid­er­ing the strong prom­ises and ex­per­i­men­tal suc­cess in these ap­pli­ca­tions, AI has gained strong trac­tion from phar­ma­ceu­ti­cal and life sci­ences IT com­pa­nies to cash on the po­ten­tial op­por­tu­nity. In fact, all the lead­ing phar­ma­ceu­ti­cal com­pa­nies, in­clud­ing J&J, No­var­tis, Merck, Pfizer, Roche, and Astel­las, are pur­su­ing AI ca­pa­bil­ity and ap­pli­ca­tions through ex­ter­nal part­ner­ships or in-house ca­pa­bil­ity de­vel­op­ment. As per Frost & Sul­li­van’s re­cent es­ti­mates, AI-based drug dis­cov­ery mar­ket will ex­ceed $3 bil­lion by 2022.

Life Sci­ence IT Start-ups tak­ing lead in the busi­ness, with can­di­date-as-a-ser­vice to be­come re­al­ity in pharma in­dus­try

As pref­er­ence to­wards data-driven de­ci­sion mak­ing in

drug dis­cov­ery and de­vel­op­ment grows, we are look­ing at a sce­nario, wherein a large part of drug dis­cov­ery and trial ac­tiv­i­ties will shift from labs to com­put­ers to bring the much-needed ef­fi­ciency. In the nut­shell, we are look­ing at the con­ver­gence of data sci­ence and medicine sci­ence.

When we talk about adopt­ing ma­chine learn­ing and big data an­a­lyt­ics in drug dis­cov­ery, we look at uti­liz­ing petabytes of data—sourced from dif­fer­ent pa­tient data­bases, di­ag­nos­tics records, health­care records and other sci­en­tific data­bases. About 90 per cent top sell­ing block­buster medicines only work for 30–50 per cent of the pa­tients, and hence pa­tient cen­tric­ity is an­other key trend and chal­lenge, ex­pected to be solved by AI. How­ever one im­por­tant ques­tion which is yet to be an­swered is—how do we find ways to ef­fec­tively col­lect dif­fer­ent types of data for bet­ter anal­y­sis; and more im­por­tantly how do we use avail­able in­for­ma­tion to get the de­sired re­sults? While strong pro­gresses have been made in this di­rec­tion, the in­dus­try still awaits a per­ma­nent and uni­ver­sal so­lu­tion to this chal­lenge. An­other core ques­tion is that—is pub­li­cally avail­able data is good enough to over­come these chal­lenges? Or do we need to do more with data col­lec­tion? Pro­vid­ing right set of data to AI and ma­chine learn­ing plat­forms is still a work-in-progress and this is an ex­ist­ing chal­lenge for com­pa­nies for im­ple­ment­ing AI and pa­tient cen­tric­ity con­cepts in drug dis­cov­ery prac­tices.

So far we have seen a num­ber of start-ups ven­tur­ing in this seg­ment, es­pe­cially in the US and Europe, and can­di­date-as-a-ser­vice is a con­cept in de­vel­op­ment in the in­dus­try. In near fu­ture, we ex­pect a num­ber of start-ups and re­search fo­cused com­pa­nies of­fer­ing this so­lu­tion us­ing their in-house AI and ma­chine learn­ing plat­forms. Ac­cord­ingly, we ex­pect num­ber of part­ner­ships and M&As ac­tiv­i­ties to in­crease in the next 2–3 years. Com­pa­nies, with ex­clu­sive ac­cess to spe­cific pa­tient and health­care data, will get higher trac­tion from clients and in­vestors.

Fur­ther, with in­vent and con­ver­gence of ge­nomics and next gen-com­pu­ta­tional chem­istry tech­niques, the scope of ap­ply­ing artificial in­tel­li­gence in drug dis­cov­ery will in­crease, which in turn will re­sult in the de­vel­op­ment of per­son­al­ized and pre­ci­sion medicines.

Asia – A slow starter in AI adop­tion race

Tra­di­tion­ally the con­ver­gence of data an­a­lyt­ics, AI, ma­chine learn­ing and medicine sci­ences in Asia Pa­cific (APAC) mar­ket has been slower than the US and Euro­pean mar­ket. The fun­da­men­tal rea­son be­hind this is that most APAC mar­kets are not R&D in­ten­sive and the in­dus­try revolves around generics and biosimilars. In re­search fo­cused mar­kets, in­clud­ing Ja­pan and China, avail­abil­ity of data­bases and lan­guage bar­rier have been key chal­lenges. How­ever, some large Asian phar­ma­ceu­ti­cal com­pa­nies are in­creas­ingly adopt­ing AI-based drug dis­cov­ery prac­tices by en­gag­ing ex­ter­nal ven­dors. For ex­am­ple, Dainip­pon Su­mit­omo Pharma has en­gaged Ex­sci­en­tia to work on small mol­e­cules drugs. Sim­i­larly, Astel­las Pharma has signed a part­ner­ship with big datadriven bioin­for­mat­ics com­pany NuMedii to con­duct drug re­pur­pos­ing projects us­ing ma­chine learn­ing plat­form. In ad­di­tion, the Ja­panese gov­ern­ment has also ini­ti­ated a project in col­lab­o­ra­tion with some phar­ma­ceu­ti­cal and tech­nol­ogy com­pa­nies to de­velop AI-based drug dis­cov­ery so­lu­tions. How­ever most of these projects are in ini­tial phase and we are yet to see a wide­spread adop­tion and im­ple­men­ta­tion of ma­chine learn­ing and AI con­cepts in APAC mar­ket.

Con­clu­sion

AI of­fers a lot of prom­ises for im­prov­ing ef­fi­ciency and ac­cu­racy in drug dis­cov­ery. In a highly com­pet­i­tive “win­ner takes all” busi­ness en­vi­ron­ment, the first com­pany to patent a new chem­i­cal en­tity has all the ad­van­tage, and hence, the idea of fast track­ing the drug dis­cov­ery process us­ing AI seems promis­ing. While adop­tion of artificial in­tel­li­gence pro­vides a num­ber of busi­ness growth and cost sav­ing op­por­tu­ni­ties for phar­ma­ceu­ti­cal com­pa­nies and re­searcher com­mu­ni­ties, creat­ing and col­lab­o­rat­ing with right set of data­base in­fra­struc­ture will be crit­i­cal for the de­sired suc­cess.

TRA­DI­TION­ALLY THE CON­VER­GENCE OF DATA AN­A­LYT­ICS, AI, MA­CHINE LEARN­ING AND MEDICINE SCI­ENCES IN APAC MAR­KET HAS BEEN SLOWER THAN THE US AND EURO­PEAN MAR­KET. THE FUN­DA­MEN­TAL REA­SON BE­HIND THIS IS THAT MOST APAC MAR­KETS ARE NOT R&D IN­TEN­SIVE AND THE IN­DUS­TRY REVOLVES AROUND GENERICS AND BIOSIMILARS. IN RE­SEARCH FO­CUSED MAR­KETS, IN­CLUD­ING JA­PAN AND CHINA, AVAIL­ABIL­ITY OF DATA­BASES AND LAN­GUAGE BAR­RIER HAVE BEEN KEY CHAL­LENGES. HOW­EVER, SOME LARGE ASIAN PHAR­MA­CEU­TI­CAL COM­PA­NIES ARE IN­CREAS­INGLY ADOPT­ING AI-BASED

DRUG DIS­COV­ERY PRAC­TICES BY EN­GAG­ING EX­TER­NAL VEN­DORS.

Newspapers in English

Newspapers from India

© PressReader. All rights reserved.