Exploiting sequence to function relationships is fundamental to the methodology of directed evolution in the context of biocatalyst discovery and design. However, the establishment of frameworks that provide robust and reliable paths to the desired function via these approaches remains labor intensive and costly. Using an integrative approach, in which we combine the techniques of ancestral sequence reconstruction and resurrection, modeling and bioinformatics, and machine learning tools to yield structurally annotated sequence to function landscapes, we illustrate the exploration and refinement of biocatalysts directed toward stereo-specific azaphilone natural product synthesis through the enzymatic oxidative dearomatization of substrates utilizing a family of fungal-derived flavin-dependent mono-oxygenases. I will present a novel application of large-scale structure prediction through a pipeline based on AlphaFold2 together with cofactor modeling, substrate docking and machine learning to identify key residues controlling site-specific and stereo-selective asymmetric dearomatization reactions. I will discuss the pipeline we established and illustrate how we are using the structurally annotated ancestral tree to explore and inform mutational studies and prioritization of ancestral sequence resurrection for purposes of establishing a broadly substrate scoped biocatalyst in an in silico directed evolution workflow. I will also illustrate our use of machine learning applications of variational autoencoders to visualize and systematically explore low-dimensional (latent space) representations of sequence-function relationships to elucidate the transformation between enzymes that carry-out oxidative dearomatization and those that perform decarboxylative hydroxylation.