Fig. 5. (A) Reaction similarity vs. overall molecular similarity of 38 126 reaction pairs associated with homologs versus evolutionarily distant enzymes. Red corresponds to reaction pairs of homologous enzymes (total = 18 809 pairs). Blue corresponds to reaction pairs of evolutionarily distant enzymes (total = 19 317 pairs). (B) ‘Reaction similarity’ and ‘Overall molecular similarity’ of a pair of reactions was used to discriminate between homologs vs. evolutionarily distant enzymes. (C) The output of the neural network (‘Evolution Score’) is plotted as a function of reaction- and overall molecular-similarity feature values. The decision boundary (at evolution score = 0.5) is shown in red. The model has learned that reaction pairs with high reaction- and overall molecular-similarity scores are likely evolvable.