Background and objectives. The nutritional quality of proteins depends on their indispensable amino acid profile and their digestibility. Accurate measurement of protein digestibility requires in vivo experimentation, which presents several constraints. In vitro digestion models have been developed, but presently, none of them can provide a totally reliable assessment of protein digestibility. An alternative approach, the subject of this study, could be to model protein digestibility on the basis of in vitro digestion data and/or readily available physicochemical characteristics of protein sources.Methods. a dataset of 47 individuals, each corresponding to a specific plant protein ingredient in a given form (seed, powder, or isolate, with or without processing), was created. Each individual was characterized by variables measured after in vitro digestion using the Infogest protocol (proteolysis degree, % soluble nitrogen), compositional traits (content of protein, lipid, carbohydrate, starch, fiber, and amino acid profile), and in vivo true N digestibility (ileal or fecal). PCA and hierarchical clustering (HAC) were applied, followed by multiple linear regression to model in vivo digestibility, with missing data handled.Results. The PCA did not reveal significant correlations between in vivo digestibility and characteristics derived from in vitro digestibility, nor with compositional traits, except for fiber content, which showed an inverse correlation (r² = -0.73) with in vivo digestibility. The HAC identified three clusters of individuals: those with low in vivo and in vitro digestibility, those poorly digestible in vivo, and those highly digestible in vivo and in vitro. The latter cluster was characterized by the presence of ingredients that had undergone significant processing, while the former included various forms of chia. Several highly significant (p-value<10 -6 ) regression models were established, with the best performing (r²=0.777, p-value=4.8e-08) modeling in vivo digestibility from the degree of proteolysis measured in vitro, content of fibers and protein content in Arg, His, Ile, Lys and Pro.Conclusions. This preliminary study shows potential for predicting in vivo digestibility of plant protein ingredients using easily accessible measurements, which could reduce the need for in vivo experiments. However, further validation is required, both by expanding the dataset with more diverse, well-characterized ingredients and by using cross-validation.These approaches are currently being explored.