FASTmC: A Suite of Predictive Models for Nonreference-Based Estimations of DNA Methylation

Abstract

We describe a suite of predictive models, coined FAST(m)C, for nonreference, cost-effective exploration and comparative analysis of context-specific DNA methylation levels. Accurate estimations of true DNA methylation levels can be obtained from as few as several thousand short-reads generated from whole-genome bisulfite sequencing. These models make high-resolution time course or developmental and large diversity studies practical regardless of species, genome size, and availability of a reference genome.

Keywords: DNA methylation; epigenetics; methylome; modeling; whole-genome bisulfite sequencing.

Figure 1

Detection of intraspecific DNA methylation levels by FAST^mC. (A) Linear models (LMs) for estimated levels of methylation, i.e., $\ddot{\widehat{F}}/\widehat{p}$, (Y-axis) vs. actual levels (X-axis) determined by reference mapping of WGBS reads. Estimated levels of methylation were based on 10,000 random WGBS reads. DNA methylation differences between A. thaliana mutants, mouse mutants/cell-types/tissues, and increasing CH methylation throughout brain development are captured with FAST^mC. Shaded area represents the 95% confidence interval. (B) Environmental (temperature) effects on CHH DNA methylation in A. thaliana is also recapitulated using FAST^mC. Left panel (“FAST^mC”) represents FAST^mC methylation estimates for individual lines based solely on the WGBS reads using the “plant” model from http://fastmc.genetics.uga.edu. Right panel (“True”) represents the methylation values from standard WGBS read alignment to the A. thaliana reference genome. Red lines are averages of all lines. Data from Dubin et al. 2015.

Figure 2

Detection of interspecific DNA methylation levels by FAST^mC. Linear models (LMs) for estimator ($\widehat{F}$) vs. target (m) CpG, CHG, CHH, and CH DNA methylation levels using 10,000 reads corrected for estimated GC content ($\widehat{p}$).