Fig. 1

Modelling overview. a Schematic metabolic network of fibre degradation in the gut. The metabolic network used to model fibre degradation in the gut is represented from complex dietary and host-derived fibres to terminal metabolites. Dashed boxes in the upper part represent fibre pools that are linked to fibre-derived sugars by GH and PL. Intra- and extra-cellular metabolites are respectively represented by gray and black boxes. Metabolic pathways linking metabolites are numbered from 1 to 68 (see Table 2): representative KOs are selected for each pathway, checking for specificity (KO are not involved in other metabolic reactions) and essentialness (essential reactions for the completion of the pathway). Functional blocks are represented by colored shapes. GH_Fucose and GH_galactose, complex carbohydrate involving respectively fucose and galactose; ED, Entner-Doudoroff; SP-ED, semi-phosphorylative Entner-Doudoroff; EMP, Embden-Meyerhoff-Parnas; Bif. shunt, Bifidobacterium shunt; WL, Wood-Ljundhal. Complete list of reactions and abbreviations can be found in the Additional file 11—Dataset count matrices, Profile decomposition and metadata. b Gene count aggregation pipelines. The pipelines used to build the count matrices are sketched. To build \(X^{(AFT)}\), KO, GH and PL are first selected according to the metabolic network in a, leading to a list of selected genes (SG) that are annotated in the 9.9M gene catalog and pooled into their respective KO, GH or PL. Some KOs are gathered according to functional proximity, leading to aggregated functional trait (AFT). This aggregation scheme allows to transform sample gene frequencies into AFT frequencies in \(X^{(AFT)}\) by pooling SG counts. For prevalent genome (PG) counts, taxonomic marker genes (TMG) are extracted from the genomes with FetchMg and annotated in the 9.9M catalog: the aggregated TMG are next counted in the samples to build \(X^{(PG)}\). MGS are reconstructed from the metagenomes, directly counted in the samples and pooled by genus to build \(X^{(mgs)}\). A NMF is performed on \(X^{(AFT)}\) to obtain \(W^{(AFT)}\) (weights) and \(H^{(AFT)}\) (functional profiles). Then, nonnegative least square inference is conducted on \(X^{(PG)}\) and \(X^{(mgs)}\) using \(W^{(AFT)}\) as regressor to obtain \(H^{(PG)}\) and \(H^{(mgs)}\) (PG and MGS taxonomic profiles)