Chemical-scale force responses across G-quadruplex topologies via machine learning.

G-quadruplexes (GQs) are four-stranded DNA structures formed by guanine-rich sequences and can act as regulatory impediments in gene expression. The core of GQ consists of stacked layers of Hoogsteen-paired guanine bases (G-tetrads), which can arise from various topologies adopted by the backbone strands of guanine tracts (G-tracts). The polymorphic nature of GQs presents significant challenges for therapeutic targeting. An as-yet-addressed question is how, if at all, chemical moieties of different G-tetrads and G-tracts mediate distinct mechanical strengths such that the strands and layers display nonuniform mechanical resistance. As a first attempt toward a quantitative understanding of this fundamental problem, we generalize the recently developed structure-mechanics statistical learning framework. The method transforms an all-atom molecular dynamics trajectory into Hookean spring coefficients between base, ribose, backbone, and metal-ion moieties, which allows a chemical-scale comparison of mechanical heterogeneity across GQ topologies. A recurring theme that emerges from this analysis is the cross-influence among chemical moieties. Contacts with loops may cooperatively strengthen base stacking in one strand while leading to compensatory weakening in another. Importantly, end-capping of intra-molecular GQs with small-molecule stabilizers is shown to induce system-wide and topology-specific variations in chemical-scale mechanical properties. Strong stabilizer-GQ interactions may compete with base-metal ion coordination or base stacking and reduce their strengths. However, a weakly bound stabilizer can trigger loop reorganization, which in turn leads to new interactions that enhance structural stability. Our analysis of chemical-scale mechanical heterogeneity thus provides a generalizable framework that could be applicable to understanding more complicated inter-molecular GQs and their responses to small-molecule modulations.