Prediction of hydrogen and carbon chemical shifts from RNA using database mining and support vector regression
 

Prediction of hydrogen and carbon chemical shifts from RNA using database mining and support vector regression

Abstract

The Biological Magnetic Resonance Data Bank (BMRB) contains NMR chemical shift depositions for over 200 RNAs and RNA-containing complexes. We have analyzed the 1H NMR and 13C chemical shifts reported for non-exchangeable protons of 187 of these RNAs. Software was developed that downloads BMRB datasets and corresponding PDB structure files, and then generates residue-specific attributes based on the calculated secondary structure. Attributes represent properties present in each sequential stretch of five adjacent residues and include variables such as nucleotide type, base-pair presence and type, and tetraloop types. Attributes and 1H and 13C NMR chemical shifts of the central nucleotide are then used as input to train a predictive model using support vector regression. These models can then be used to predict shifts for new sequences. The new software tools, available as stand-alone scripts or integrated into the NMR visualization and analysis program NMRViewJ, should facilitate NMR assignment and/or validation of RNA 1H and 13C chemical shifts. In addition, our findings enabled the re-calibration a ring-current shift model using published NMR chemical shifts and high-resolution X-ray structural data as guides.



Source: Journal of Biomolecular NMR