Search for a command to run...
Developing approaches to link structure and function is an ongoing challenge in computational and structural biology. Using a systems-level framework, we present here an analysis pipeline in a Python package, mdsa-tools, that constructs network representations of structures in a time series of trajectory frames from molecular dynamics (MD) simulations. Here, we demonstrate its use on a ribosomal subsystem. The subsystem is centered on the CAR interaction surface, a "brake pad" adjacent to the aminoacyl (A-site) decoding center that tunes protein translation rates. We leverage unsupervised learning algorithms to explore the conformational landscape of behaviors visited by two versions of the subsystem (brake-on and brake-off) that differ at the codon 3' adjacent to the A-site codon. Our network representations of MD frames embody H-bond interactions between all pairwise combinations of residues in the system. By utilizing per-frame vector representations of network edges, we can apply standard clustering and dimensionality reduction methods to explore behavioral differences between the brake-on and brake-off versions of the system. K-means clustering of frame vectors revealed a striking separation of the two system versions, consistent with principal components analysis (PCA) embeddings and Uniform Manifold Approximation and Projection (UMAP) embeddings. Dissection of K-means centroids and PCA loadings highlighted H-bond interactions between residue pairs in the ribosome's peptidyl site (P site), suggesting potential allosteric signaling across the subsystem.