Complex polyketides such as erythromycin, epothilone, and FK506 greatly contribute to human health and are synthesized by modular polyketide synthases (PKSs). A novel, evolutionarily distinct group of trans-AT PKSs—using free-standing acyltransferases (AT) instead of integrated AT domains—has been linked to diverse bioactive polyketides but was long overlooked. Building on a predictive approach for trans-AT PKSs based on ketosynthase (KS) domain phylogeny (correlating KS position in phylogenetic trees with substrate specificity), we applied trans-AT genome mining to Chitinophaga pinensis DSM 2588, a member of the chemically poorly studied Bacteroidetes phylum. We identified a 75.1 kb trans-AT PKS system (els PKS) with 18 genes: six multimodular PKS proteins lack integrated AT domains, while free-standing ATs (ElsA, ElsB) are present (ElsA also harbors a C-terminal oxidoreductase domain for trans enoyl reduction). To predict the polyketide structure, a phylogenetic tree was constructed from 413 KS sequences (catalyzing polyketide chain extensions). Most els KSs fell into functional clades with known substrate specificities, enabling high-confidence deduction of the polyketide structure despite aberrant PKS features (e.g., split modules, mutated HGTGT motifs essential for elongation). A conserved sequence motif (CGVYLGIMS/N/GN/SE) specific to a clade with β-methyl groups on E double bonds was also identified, facilitating rapid function prediction. Our study links the els PKS to the biosynthesis of elansolids—antibacterial and cytotoxic agents with unusual post-PKS modifications—extends trans-AT genome mining to the Bacteroidetes phylum, and validates the predictive power of KS phylogenetic analysis for non-canonical PKS pathways.