Reverse prediction of carbohydrate esterase polysaccharide targets

Carbohydrate esterases (CEs) catalyze the selective removal of ester-linked substituents from complex polysaccharides, influencing biomass bioprocessing. Predicting CE substrate specificity remains challenging due to the functional diversity and limited experimental characterization of CEs. Classic full sequence-based alignment approaches often fail to capture functional nuances across distant homologs. Here, we introduce a reverse prediction framework that leverages genomic context, specifically polysaccharide utilization loci (PULs), to infer natural polysaccharide targets of CE families. By integrating motif-based functional groups with large-scale co-occurrence analysis across Bacteroidota genomes, we identify substrate preferences at the clade level for 20 CE families. Subdivision of families into clades mitigated any polyspecificity observed when families were treated as a whole, highlighted unexplored regions within CE1, CE2, CE3, CE6, CE7, CE14, CE15, CE19, CE20, and partly within CE8 and CE12, and expanded functional coverage by up to 50% compared to characterized members alone. The approach and the data reveal substantial functional subdivision of the CE families, enabling prediction of specific targets including arabinoxylan, β-mannan, pectin sub-structures, and glycosaminoglycans. Reverse prediction thus offers a powerful tool for guiding enzyme discovery and designing tailored enzyme cocktails for biomass valorization.