Can AI replace animal testing?

Arti­fi­cial intel­li­gence (AI) algo­rithms have proved to be high­ly effec­tive when it comes to sim­u­lat­ing human voic­es or the pro­duc­tion of images. But are they also capa­ble of sim­u­lat­ing liv­ing beings well enough to make it pos­si­ble to dis­pense with ani­mal test­ing? The ques­tion has emerged in the face of grow­ing con­cern around ani­mal wel­fare. “We no longer think of ani­mals as machines, and our soci­ety believes that humans are respon­si­ble for pro­tect­ing them. In addi­tion advances in ethol­o­gy and ani­mal psy­chol­o­gy, and the emer­gence of con­cepts such as ani­mal cul­ture. All these fac­tors are call­ing ani­mal test­ing into ques­tion,” explains Jean-Michel Besnier, philoso­pher of sci­ence and pro­fes­sor emer­i­tus of phi­los­o­phy at Sor­bonne Uni­ver­sité. This soci­etal con­cern coin­cides with a grow­ing aware­ness of anoth­er issue. “It is not that easy to draw con­clu­sions about humans from mice… What is the point of mak­ing ani­mals suf­fer for results that are open to question?”

Ani­mals are used in research for sev­er­al pur­pos­es: to study and under­stand bio­log­i­cal phe­nom­e­na as part of fun­da­men­tal research, and to check the safe­ty and effi­ca­cy of a prod­uct or drug as part of reg­u­la­to­ry and tox­i­co­log­i­cal research. This issue has not been over­looked by the sci­en­tif­ic com­mu­ni­ty. The 2010 Euro­pean Direc­tive (2010/63/EU) pro­vides a frame­work for ani­mal test­ing through the three Rs: Replace­ment, Reduc­tion and Refine­ment, which aims to reduce the suf­fer­ing inflicted.

Can arti­fi­cial intel­li­gence sys­tems replace ani­mal test­ing? This is the aim of so-called “dig­i­tal twin” sys­tems. These are sim­u­la­tion pro­grammes that, for exam­ple, mim­ic the bio­chem­i­cal or bio­phys­i­cal prop­er­ties of human tis­sue. “This resource is increas­ing­ly used in the med­ical field, par­tic­u­lar­ly in surgery, to give prac­ti­tion­ers the oppor­tu­ni­ty to train before the pro­ce­dure, on a sim­u­la­tion of an organ resem­bling the patient. This avatar reduces the risks involved in surgery,” con­tin­ues Jean-Michel Besnier.

Tox­i­col­o­gy is explor­ing anoth­er alter­na­tive avenue, that of organoids. These are three-dimen­sion­al cul­tures that are intend­ed to rep­re­sent an organ. “They are bio­log­i­cal objects resem­bling an organ and pro­duced in the lab­o­ra­to­ry from stem cells. How­ev­er, it is not a real organ, and here­in lie the lim­i­ta­tions. There is no guar­an­tee that the response of an organoid will be iden­ti­cal to that of a real organ,” explains Nico­las David, a devel­op­men­tal biol­o­gist at the Optics and Bio­sciences Lab­o­ra­to­ry (IP Paris), which is devel­op­ing this approach(1)¹. Sim­i­lar sys­tems are also envis­aged for per­son­alised med­i­cine, to test the response of a patient’s cells before an anti-can­cer pre­scrip­tion, for example.

Fun­da­men­tal research is cer­tain­ly the field in which it is most dif­fi­cult to replace ani­mal test­ing. Recent­ly, a team from Wash­ing­ton Uni­ver­si­ty School of Med­i­cine in St Louis² pre­sent­ed a machine learn­ing algo­rithm capa­ble of pre­dict­ing how a net­work of genes and the reg­u­la­tion of their expres­sion inter­act to con­struct the iden­ti­ty of a cell dur­ing devel­op­ment. The sys­tem pre­dicts what dri­ves a cell to become a mus­cle, skin or nerve cell, depend­ing on the genet­ic levers acti­vat­ed. Called Cel­lOr­acle, it com­piles decades of research from around the world, draw­ing on pub­lic data­bas­es that cat­a­logue known genet­ic inter­ac­tions. For exam­ple, it can be asked what effect the dis­ap­pear­ance of a gene will have in one of the mod­el organ­isms that the soft­ware inte­grates. This saves researchers the trou­ble of design­ing ani­mals that car­ry the genet­ic anom­aly. “It’s an in-sil­i­co sim­u­la­tion of knock-outs,” explains Nico­las David, “ani­mals whose genome con­tains a muta­tion pre­vent­ing the expres­sion of a cer­tain gene.”

Cer­tain mol­e­c­u­lar biol­o­gy tech­niques have been repro­duced so many times that they can be processed by a machine learn­ing algo­rithm. “But if we are able to sim­u­late a sys­tem, then we have under­stood it,” com­ments Nico­las David. Such sys­tems save explorato­ry work or point to unex­ploit­ed avenues of research, but their results are not infal­li­ble. Before embark­ing on an avenue and enter­ing a phase of applied research, they need to be verified.

The sys­tems also need to be ver­i­fied using oth­er math­e­mat­i­cal approach­es. “There are now bio­sta­tis­ti­cal sys­tems for antic­i­pat­ing the min­i­mum num­ber of ani­mals that need to be used in research to answer a giv­en ques­tion. This approach is deployed at the Insti­tut Pas­teur and helps us to reduce the vol­ume of ani­mal exper­i­ments through opti­mi­sa­tion,” explains Jean-Bap­tiste Mas­son, a sta­tis­ti­cal physi­cist at the Insti­tut Pas­teur in Paris.

In recent years, new replace­ment approach­es have emerged. Nico­las David explains: “Not all ani­mals are recog­nised as sen­tient. We can try to use sub­sti­tute species.” This may involve study­ing the ear­li­est stages of devel­op­ment, before the organ­ism is cov­ered by the law on the pro­tec­tion of ani­mals in sci­en­tif­ic research³. “In mam­mals, this depends on the total ges­ta­tion peri­od. They are only pro­tect­ed after two thirds of nor­mal devel­op­ment. The sit­u­a­tion is more com­pli­cat­ed when it comes to non-mam­mals. The law refers to an autonomous lar­val form. In fish, auton­o­my is inter­pret­ed as the abil­i­ty to feed by itself, and there­fore at the moment when the mouth opens,” explains biol­o­gist Nico­las David.

Final­ly, some teams are turn­ing to inver­te­brate ani­mal mod­els, with one obvi­ous con­straint: the fur­ther one moves away from the human species in the evo­lu­tion­ary tree, the greater the risk that the con­clu­sions will not be applic­a­ble to clin­i­cal research. This trend is also being chal­lenged by the sci­en­tif­ic com­mu­ni­ty itself. Last April, 287 philoso­phers, ethi­cists, ethol­o­gists and neu­ro­bi­ol­o­gists spe­cial­is­ing in ani­mal con­scious­ness signed the New York Dec­la­ra­tion on Ani­mal Con­scious­ness⁴. This states that it is “irre­spon­si­ble” to ignore the pos­si­bil­i­ty that all ver­te­brates and sev­er­al inver­te­brate species (such as cephalopods, insects and crus­taceans of the crab and shrimp fam­i­ly) have a “con­scious expe­ri­ence” in the light of a grow­ing body of sci­en­tif­ic research. The authors refer to ground­break­ing work on bees⁵, octo­pus­es⁶ and two species of snake⁷. This dec­la­ra­tion there­fore con­sti­tutes a pre­cau­tion­ary prin­ci­ple in the face of the pos­si­bil­i­ty that species used for sci­en­tif­ic exper­i­ments may become con­scious. All the more rea­son to resume the search for alternatives.

Agnès Vernet