Statistical Analysis of SHAPE-Directed RNA Secondary Structure Modeling

The ability to predict RNA secondary structure is fundamental for understanding and manipulating RNA function. The information obtained from selective 2'-hydroxyl acylation analyzed by primer extension (SHAPE) experiments greatly improves the accuracy of RNA secondary structure prediction. Recently, Das and colleagues [Kladwang, W., et al. (2011) Biochemistry 50, 8049-8056] proposed a "bootstrapping" approach for estimating the variance and helix-by-helix confidence levels of predicted secondary structures based on resampling (randomizing and summing) the measured SHAPE data. We show that the specific resampling approach described by Kladwang et al. introduces systematic errors and underestimates confidence in secondary structure prediction using SHAPE data. Instead, a leave-data-out jackknife approach better estimates the influence of a given experimental data set on SHAPE-directed secondary structure modeling. Even when 35% of the data were left out in the jackknife approach, the confidence levels of SHAPE-directed secondary structure prediction were significantly higher than those calculated by Das and colleagues using bootstrapping. Helix confidence levels were thus underestimated in the recent study, and the resampling approach implemented by Kladwang et al. is not an appropriate metric for evaluating SHAPE-directed secondary structure modeling.