$48 Bil­lion Is Lost to Avoid­able Ex­per­i­ment Ex­pen­di­ture Every Year

Phar­ma’s pro­duc­tiv­i­ty chal­lenges are wide­ly known. Over the last decade, phar­ma R&D costs have been steadi­ly in­creas­ing, while the rate at which drugs are ap­proved has re­mained rough­ly con­stant. As of 2018, it cost ap­prox­i­mate­ly $2.17 bil­lion (all fig­ures USD) to bring a new drug to mar­ket, al­most dou­ble the $1.18 bil­lion cost re­port­ed in 2010.¹ In the same times­pan, an­nu­al fore­cast peak sales were halved, from $816 mil­lion in 2010 to $407 mil­lion in 2018.¹ This has re­sult­ed in a de­crease in re­turns of 8.2 per­cent­age points, from 10.1% in 2010, to 1.9% in 2018.¹

Drug de­vel­op­ment isn’t on­ly cost­ing more, it’s al­so tak­ing longer. The av­er­age project length in­creased from 9.7 years in the 1990s to 10-15 years through the 2000s and 2010s.²,³ Fac­tors com­mon­ly cit­ed for these trends in­clude clin­i­cal tri­al in­ef­fi­cien­cies, such as pa­tient en­rol­ment com­pe­ti­tion and more strin­gent se­lec­tion cri­te­ria, and the in­dus­try’s shift in­to more dif­fi­cult, high-risk, high-re­ward re­search ar­eas like on­col­o­gy.

But there’s an­oth­er, less­er known cul­prit. Our da­ta sug­gests that Avoid­able Ex­per­i­ment Ex­pen­di­ture (AEE) is an over­looked source of un­nec­es­sary spend, sig­nif­i­cant­ly con­tribut­ing to the in­crease in project length and fall in re­turns. Con­sid­er­ing that ~42.9% of over­all spend on drug de­vel­op­ment goes to­ward pre­clin­i­cal R&D, this is­sue de­serves more at­ten­tion.⁴ Avoid­able Ex­per­i­ment Ex­pen­di­ture refers to all in­ef­fi­cien­cies and pro­duc­tiv­i­ty chal­lenges in de­sign­ing and car­ry­ing out pre­clin­i­cal ex­per­i­ments. Ex­per­i­ments are the foun­da­tion of pre­clin­i­cal re­search and de­vel­op­ment, how­ev­er, ir­re­pro­ducibil­i­ty rates in pre­clin­i­cal ex­per­i­ments ex­ceed 50%, cost­ing the in­dus­try near­ly $48 bil­lion an­nu­al­ly.⁵

In­ap­pro­pri­ate reagents ac­count for $17 bil­lion in Avoid­able Ex­per­i­ment Ex­pen­di­ture an­nu­al­ly

There are mul­ti­ple fac­tors that con­tribute to AEE, and are in­her­ent chal­lenges of any drug dis­cov­ery ef­fort. These in­clude in­ap­pro­pri­ate reagents, poor ex­per­i­men­tal de­sign, un­re­li­able or vari­able pro­to­cols, lack of de­tails in re­port­ing, and lack of trans­paren­cy in in­ter­nal da­ta. How­ev­er, in­ap­pro­pri­ate or faulty reagents are one of the most sig­nif­i­cant in­di­vid­ual con­trib­u­tors.

Over a third of AEE, or more than $17 bil­lion, can be at­trib­uted to the in­ef­fec­tive­ness of bi­o­log­i­cal reagents or ref­er­ence ma­te­ri­als.⁵ By re­solv­ing reagent-re­lat­ed AEE, life sci­ence or­ga­ni­za­tions could po­ten­tial­ly re­coup $17 bil­lion in un­nec­es­sary spend while stream­lin­ing op­er­a­tions, im­prov­ing R&D ef­fi­cien­cy, bring­ing as­sets to clin­i­cal tri­al faster, and gen­er­al­ly ac­cel­er­at­ing pipeline progress.

Com­mon­ly used bi­o­log­i­cal reagents, in­clud­ing an­ti­bod­ies, re­com­bi­nant pro­teins, RNAi, and CRISPR, as well as mod­el sys­tems like cell lines, are sourced from bi­o­log­i­cal ori­gins, such as an­i­mals and pa­tient sam­ples. As a re­sult, the per­for­mance of bi­o­log­i­cal reagents can vary de­pend­ing on the bi­o­log­i­cal sys­tems in which they are both cre­at­ed and used. Reagent se­lec­tion is chal­leng­ing—ex­per­i­men­tal con­text must be tak­en in­to ac­count, da­ta for which is scat­tered through­out many sources. This is ex­ac­er­bat­ed by un­re­li­a­bil­i­ty in reagent da­ta, as well as in reagents them­selves; it is es­ti­mat­ed up to 50% of reagents do not work as in­tend­ed.⁶

In fact, over 36% of all un­pro­duc­tive ex­per­i­ments in pre­clin­i­cal R&D can be at­trib­uted to in­ef­fec­tive reagents.⁵ Cur­rent stan­dard process­es for se­lect­ing reagents are flawed and un­re­li­able at best. How­ev­er, the gen­er­al per­cep­tion is that the ex­plorato­ry na­ture of pre­clin­i­cal R&D makes it in­her­ent­ly in­ef­fi­cient, thus the se­lec­tion and use of in­ap­pro­pri­ate reagents is deemed an un­avoid­able and nec­es­sary part of the sci­en­tif­ic process. In fact, how­ev­er, reagent se­lec­tion can be sig­nif­i­cant­ly im­proved with emerg­ing tech­nol­o­gy.

Ma­chine learn­ing of­fers new hope for im­proved reagent se­lec­tion and re­duced Avoid­able Ex­per­i­ment Ex­pen­di­ture

The in­ef­fi­cien­cies in pre­clin­i­cal ex­per­i­ments have a di­rect and mea­sur­able im­pact on an or­ga­ni­za­tion’s pro­duc­tiv­i­ty, re­sult­ing in longer re­search times, more spend on drug de­vel­op­ment, and in­creased down­stream risk. How­ev­er, with ac­cess to bet­ter in­for­ma­tion, im­proved reagent se­lec­tion, more ef­fec­tive ex­per­i­men­tal pro­to­cols, and more in­formed ex­per­i­men­tal de­sign, is­sues such as un­pro­duc­tive ex­per­i­ment rates and pro­por­tions of ma­te­r­i­al waste could be re­duced.

The first step to solv­ing a prob­lem is find­ing a way to quan­ti­fy it. As of now, very lit­tle is be­ing done to track pro­duc­tiv­i­ty in pre­clin­i­cal ex­per­i­ments. We sug­gest it would be wise for the in­dus­try to align on stan­dards to mea­sure this, as well as the im­pacts of AEE, be­gin­ning with reagent-re­lat­ed AEE. Once we know the scope and scale of the prob­lem, we can mea­sure the im­pact of so­lu­tions that ad­dress it by help­ing sci­en­tists make more in­formed de­ci­sions.

The sec­ond step is to em­pow­er sci­en­tists with in­for­ma­tion to de­sign more suc­cess­ful ex­per­i­ments. Da­ta to guide de­ci­sions about ex­per­i­men­tal de­sign, such as se­lect­ing reagents, ex­ists but is buried in dis­parate sources of pub­lic and pro­pri­etary in­for­ma­tion. Sci­en­tists of­ten spend days search­ing through lit­er­a­ture for reagent ref­er­ences, then weeks to months val­i­dat­ing mul­ti­ple reagents for their ex­per­i­ments. Yet still, at least half of the reagents they pur­chase can fail them. It’s im­pos­si­ble for sci­en­tists to sort through and or­ga­nize all of the da­ta to in­form their reagent se­lec­tion man­u­al­ly.

Ma­chine learn­ing can en­able both the quan­tifi­ca­tion and re­duc­tion of reagent-re­lat­ed AEE. When trained to an­a­lyze sci­en­tif­ic da­ta as a hu­man re­searcher would, ma­chine learn­ing al­go­rithms can find pat­terns at scale that no hu­man could iden­ti­fy. For quan­tifi­ca­tion of reagent-re­lat­ed AEE, we can use the al­go­rithms to an­a­lyze an or­ga­ni­za­tion’s reagent da­ta to mea­sure the util­i­ty and cost-ef­fec­tive­ness of reagent pur­chas­ing. For re­duc­tion of reagent-re­lat­ed AEE, we can use these al­go­rithms to help sci­en­tists se­lect the most ef­fec­tive reagents for their spe­cif­ic ex­per­i­ments in a frac­tion of the time it would nor­mal­ly take.

Bench­Sci is the on­ly com­pa­ny in the world cur­rent­ly pur­su­ing this av­enue. Our dis­cov­ery of AEE was in­formed by our work with over 40,000 sci­en­tists at more than 4,300 in­sti­tu­tions, as well as 15 of the top 20 phar­ma com­pa­nies. To learn more about Avoid­able Ex­per­i­ment Ex­pen­di­ture, ac­cess the full whitepa­per here.


1. “Ten years on, Mea­sur­ing the re­turn from phar­ma­ceu­ti­cal in­no­va­tion 2019.” De­loitte.

2. Nor­man, Gail A. Van. “Drugs, De­vices, and the FDA: Part 1: An Overview of Ap­proval
Process­es for Drugs.” JACC: Ba­sic to Trans­la­tion­al Sci­ence. El­se­vi­er, April 25, 2016.

3. Pam­mol­li, Fabio, Lau­ra Mag­a­zz­i­ni, and Mas­si­mo Ric­caboni. “The Pro­duc­tiv­i­ty Cri­sis in
Phar­ma­ceu­ti­cal R&D.” Na­ture News. Na­ture Pub­lish­ing Group, June 1, 2011.

4. Di­Masi, Joseph A., Hen­ry G. Grabows­ki, and Ronald W. Hansen. “In­no­va­tion in the
Phar­ma­ceu­ti­cal In­dus­try: New Es­ti­mates of R&D Costs.” Jour­nal of Health Eco­nom­ics. North-Hol­land, Feb­ru­ary 12, 2016.

5. Freed­man, Leonard P., Iain M. Cock­burn, and Tim­o­thy S. Sim­coe. “The Eco­nom­ics of
Re­pro­ducibil­i­ty in Pre­clin­i­cal Re­search.” PLOS Bi­ol­o­gy 13, no. 6 (Sep­tem­ber 2015).

*Freed­man et al. re­port fig­ures for the US on­ly. We have ex­trap­o­lat­ed their da­ta to re­flect the fact that the US ac­counts for 58% of glob­al phar­ma­ceu­ti­cal R&D. See:

6. Bak­er, Monya. “Re­pro­ducibil­i­ty Cri­sis: Blame It on the An­ti­bod­ies.” Na­ture News.
Na­ture Pub­lish­ing Group. May 19, 2015.