Machine-Based Learning for Rational Engineering of Biomolecular Libraries

Ribonucleic acids (RNAs) have been become highly relevant to our understanding of cellular regulation. Although in highly diverse ways, RNAs effect cellular changes by altering recognition patterns with their targets.. Recent work in our laboratory has focused on the discovery of non-coding RNAs of relevance to various bacterial stress responses. An important part of characterizing these RNAs has been understanding their structural rearrangements in the native context of the cellular milieu. Although RNA structure has been extensively studied using a variety of in vitro and in silico techniques, to date, only a few RNA structural characterization techniques have been used intracellularly. In this work, we are taking advantage of principles derived from synthetic biology and from rational biomolecular engineering to design fluorescence-based libraries of probes capable of in vivo detection of structural perturbations in target RNAs. While we have been experimentally successful in using flow cytometry data to perceive structural differences between RNA conformers, a major challenge has been the co-engineering of optimal probe libraries that, in an interdependent and combinatorial way, provide the most structural information. We will discuss the application of machine-based learning strategies for automating the engineering of these larger molecular libraries that are coupled by functionality. These strategies have been largely successful in the rational design of non-biological materials. Our preliminary results demonstrate that a strong connection between machine learning and molecular engineering could be designed based on imposing mechanistic descriptions of the functionality of the library. We will also discuss potential implications of this approach for engineering molecules in a way that optimizes multiple traits.