Iterative and data-driven ortholog mining enables reliable discovery of stereoselective ketoreductases

Ketoreductases (KREDs) enable stereoselective alcohol synthesis and are widely used industrially. Yet identifying process-compatible and selective enzymes for non-native substrates remains challenging, requiring extensive screening. Here, we present an iterative ortholog mining strategy to explore functional space. Starting from a known KRED, we sample its orthogroup across 48 evolutionarily distant variants, identifying five enzymes producing pharmaceutically relevant alcohols (de/ee 42.8% to >99%). Refined sampling of 60 phylogenetically close orthologs yields further improved ketoreductases (de/ee 98% to >99%). Because ortholog mining is genome-dependent, it overlooks enzymes from unsequenced organisms. To address this, we use our dataset to explore homologous KREDs beyond the sampled orthogroup, expanding the search space fifteenfold. To this end, we develop the automated pipeline HomoLogic, whose functional descriptors, fed into interpretable models, predict KRED performance accurately (R² up to 0.76). Overall, we show that evolutionarily balanced ortholog panels combined with data-driven modeling enable efficient discovery of stereoselective biocatalysts.