Machine learning deciphers structural features of RNA duplexes measured with solution X-ray scattering

Abstract

Macromolecular structures can be determined from solution X-ray scattering. Small-angle X-ray scattering (SAXS) provides global structural information on length scales of 10s to 100s of Ångstroms, and many algorithms are available to convert SAXS data into low-resolution structural envelopes. Extension of measurements to wider scattering angles (WAXS or wide-angle X-ray scattering) can sharpen the resolution to below 10 Å, filling in structural details that can be critical for biological function. These WAXS profiles are especially challenging to interpret because of the significant contribution of solvent in addition to solute on these smaller length scales. Based on training with molecular dynamics generated models, the application of extreme gradient boosting (XGBoost) is discussed, which is a supervised machine learning (ML) approach to interpret features in solution scattering profiles. These ML methods are applied to predict key structural parameters of double-stranded ribonucleic acid (dsRNA) duplexes. Duplex conformations vary with salt and sequence and directly impact the foldability of functional RNA molecules. The strong structural periodicities in these duplexes yield scattering profiles with rich sets of features at intermediate-to-wide scattering angles. In the ML models, these profiles are treated as 1D images or features. These ML models identify specific scattering angles, or regions of scattering angles, which correspond with and successfully predict distinct structural parameters. Thus, this work demonstrates that ML strategies can integrate theoretical molecular models with experimental solution scattering data, providing a new framework for extracting highly relevant structural information from solution experiments on biological macromolecules.

RNA triplex structures revealed by WAXS-driven MD simulations

“…Solution X-ray scattering provides both global and local information about the structure(s) of RNA. While small-angle X-ray scattering (SAXS) reveals the overall size and shape of the molecule, wide-angle X-ray scattering (WAXS) is now a proven tool for discerning variations in the solution structure(s) of RNA molecules, 36, 55, 56 importantly reporting higher-resolution information about the distribution of distances present in the molecule in vitro . We apply WAXS to understand the structure(s) of motifs with tertiary structures: triple stranded RNAs.…”

In Vitro Electron Density Refinement from Solution X-ray Scattering in the Wide-Angle Regime

“…Moving to the second structure, the hairpin triplex, we rely on intuition gained from previous work using machine learning methods to interpret the WAXS features. For an RNA duplex [15], we 'learned' that the second peak at q ≈ 0.7Å −1 reflects the major groove. It is known that the third RNA strand involved in creating a triplex (three stranded structure) from a duplex (two stranded structure), the so called triplex forming oligo or TFO, occupies the RNA major groove via stabilization from consecutive base triples.…”

“…This length scale is ideally matched to structural features of nucleic acid systems and can be readily interpreted using the FM algorithm introduced here. Prior work, yielding quantitative interpretation of WAXS profiles required incorporation of molecular dynamics (MD) simulation to either fit the measurements [6], select important features for machine learning algorithms [15]. However, many biological systems and phenomena may prove challenging to model by MD simulations, due to long simulation times or artifacts from force fields.…”

Automated matching of two-time X-ray photon correlation maps from protein dynamics with Cahn-Hilliard type simulations using autoencoder networks

“…Machine learning (ML) has already proven to be useful in the analysis of various structural X-ray data from small angle X-ray scattering (SAXS) [29][30][31][32] and wide angle X-ray scattering (WAXS) [30,32], diffraction [33][34][35] and reflectometry [36] experiments. The algorithms provide real time feedback [32,36,37] during the experiment or sort out bad images to reduce the stored data volume [32,37].…”

“…The algorithms provide real time feedback [32,36,37] during the experiment or sort out bad images to reduce the stored data volume [32,37]. In many cases the machine learning algorithms were trained with synthetical structural data generated from existing data bases on proteins [29,31], RNA [30] or crystal structures [34,35], which enables them to classify the results obtained in experiments without an expert spending time on labeling large amounts of data for training [33].…”