Automating Model Building for SPM Images of Biomolecules Using MISO

Abstract

Direct imaging by scanning probe microscopy (SPM) at cryogenic temperatures in ultrahigh vacuum (UHV) has enabled structural characterization of individual biomolecules in heterogeneous systems, including glycans and glycan-decorated biomolecules (also known as glycoconjugates). However, interpreting SPM images into molecular structures remains a major bottleneck, particularly for flexible biomolecules that can adopt multiple adsorption geometries. Currently, hypothesizing, constructing, and testing these geometries are largely performed manually, thus limiting the data throughput of SPM-based structural analysis and the scale of molecular systems addressable by SPM imaging technologies. Here, we present MISO (Model building from Identity, Sequence, and Observed location) as a workflow for generating three-dimensional biomolecular models from user-defined hypotheses about subunit identity, connectivity, and observed locations in SPM images. MISO produces structures consistent with user-defined hypotheses by implementing the quaternion estimator algorithm (QUEST) followed by biased classical molecular dynamics (MD). When applied to existing SPM data of glycans and glycoconjugates, MISO generated structures that were qualitatively consistent with structures previously validated by density functional theory (DFT) calculations. In addition, we show the use of MISO in filtering user hypotheses as well as the scale-up potential of MISO in building a 3D model of an unfolded glycoprotein on the surface. These results establish MISO as a practical tool for automating molecular model building, which helps to improve the overall throughput of SPM image interpretation of flexible biomolecules.