Anne-Vir­ginie Sal­sac is a research direc­tor at the CNRS, in the Inter­ac­tions Flu­ides Struc­tures Biologiques (IFSB) team of the UTC-CNRS-BMBI lab­o­ra­to­ry. With Flo­ri­an De Vuyst, she is devel­op­ing “dig­i­tal twins” of micro­cap­sules under flow conditions.

In con­crete terms? “A micro­cap­sule con­sists of a mem­brane pro­tect­ing a drop of an active flu­id, like a drug. Like red blood cells, it enables the flu­id to be deliv­ered direct­ly to the tis­sues, while avoid­ing side effects for the patient. Cur­rent­ly, ther­a­peu­tic solu­tions on the mar­ket are nanoscale: their advan­tage (and dis­ad­van­tage) is that they can cross all bar­ri­ers, but the quan­ti­ty of drug encap­su­lat­ed is infin­i­tes­i­mal”, she explains.

Hence the idea, as part of the Mul­ti­phys Micro­caps project fund­ed by the Euro­pean Research Coun­cil, to devel­op larg­er, microme­tre-sized vec­tors. “How­ev­er, to ensure that the vec­tors are safe and can pass through cap­il­lar­ies or even small­er pores, we need to be sure of their deforma­bil­i­ty and resis­tance to blood flow. We are there­fore devel­op­ing numer­i­cal mod­els that enable us to study their behav­iour under flow and iden­ti­fy their mechan­i­cal prop­er­ties by cou­pling them to micro-exper­i­ments. One of the quan­ti­ties we need to esti­mate is their risk of rup­ture”, she explains. This is a mul­ti-physics prob­lem requir­ing com­plex sim­u­la­tions. “We need to mod­el the dynam­ics of the vec­tor with its liq­uid core and thin mem­brane with non-lin­ear mechan­i­cal prop­er­ties, all inter­act­ing with the exter­nal flu­id. The only way to do this is to devel­op our own high-fideli­ty dig­i­tal codes. Being explic­it, sim­u­la­tions are time-con­sum­ing, which led us to cre­ate dig­i­tal twins. Hav­ing a large amount of sim­u­la­tion data at our dis­pos­al, we chose the “reduced-order mod­els” approach, which can be seen as physics-based AI. Their inter­est lies in pro­vid­ing us with small-scale alge­bra­ic sys­tems, which can reduce part of the high-fideli­ty codes, or indeed the codes in their entire­ty. Their oth­er inter­ests are to reduce cal­cu­la­tion times and improve under­stand­ing of the ele­ments dri­ving flu­id-struc­ture cou­pling”, stress­es Anne-Vir­ginie Salsac.

What is the next step? “We’re going to bring our mod­el reduc­tion tools into a form of dia­logue with more tra­di­tion­al AI tools. This is the aim of Lucas Wicher’s the­sis, which I am co-direct­ing with Flo­ri­an De Vuyst, co-financed by the Hauts­de- France Region and the UTC Safe-IA Chair. The aim is to deter­mine which AI tools guar­an­tee robust, safe and reli­able dig­i­tal twins”, she concludes.

MSD