Objective

Natural products continue to play a vital role in modern drug discovery. Highly diverse secondary metabolites are produced via enzyme systems commonly encoded in biosynthetic gene clusters (BGC). Significant advances in genomics, structural biology and bioengineering enabled rational modification of such natural product pathways to produce novel molecules - an endeavor expected to be further accelerated by CRISPR. A notable success story is production of an artemisinin precursor in heterologous hosts, promising low-cost production of this antimalarial drug in LIC countries.
Experimental advances were accompanied by software developments (e.g. antiSMASH) and a recent data standardization process (MIBiG). Yet, a key problem remains in connecting metabolic engineering to drug discovery: Given the vast chemical space that natural products occupy, which areas should be explored by costly bioengineering efforts? What projects should be prioritized?
Our anticipation is that cheminformatics and virtual screening will contribute majorly to an answer. In this work, polyketide and nonribosomal peptide biosynthesis are modeled with a graph rewriting framework as a proof-of-concept prototype. It is demonstrated that a) controlled generation of virtual subspace libraries and b) computational evolution towards virtual screening objectives is feasible. This allows to identify novel natural products which are both pharmaceutically relevant and reachable via metabolic engineering.