Fig. 4: Substrate environment residues targeted in the hot spot engineering and machine learning approaches.
From: Effective engineering of a ketoreductase for the biocatalytic synthesis of an ipatasertib precursor
a Model of Ssal-KRED binding 1a (blue) and NADPH (yellow). The targeted keto moiety in 1a and the hydride-donor C4 of the nicotinamide ring of the cofactor are circled in red. Catalytic residues are depicted as purple sticks. The six positions targeted in SSM libraries L2 are indicated in bold letters: F97, L174, A238, L241, M242, Q245. b Schematic representation of hot spot grouping for CSM libraries. While keeping mutation F97W fixed, five selected residues were grouped into four 3-site CSM libraries designated as L3 (174-242-245), L4 (241-242-245), L5 (238-241-242) and L6 (174-238-241). In addition, all five residues were simultaneously saturated in a 5-site CSM library (L7). c Surface representation depicting neutral to beneficial mutations (color-coded according to their FIOWT values) in the context of the wild-type enzyme. Linearly combining these mutations would have led to a theoretical library size of 29,400 variants. d Sequence-activity data of L3–L7 were used to train a ML algorithm. Predicted variants were combined in a small library of 75 variants (L10). Substitutions, which had not been found beneficial in context of the wild-type enzyme (L2) but were predicted to perform well in the ML-filtered library L10, are highlighted in red.
