deletions | additions       diff --git a/methods-array.md b/methods-array.md  index 223c231..effa032 100644  --- a/methods-array.md  +++ b/methods-array.md 
...
Samples were labeled using the NimbleGen Dual-Color DNA Labeling Kit and the arrays were processed according to the manufacturer’s recommendations (Roche NimbleGen, Madison, Wisconsin) and imaged at 5um using a GenePix 4000B microarray scanner (Molecular Devices, Sunnyvale, CA).  Fluorescent intensities from the Cy3 (input) and Cy5 (IP) channels were extracted from each array using DEVA 1.2.1 (Roche NimbleGen; Madison, WI) and the output was processed using the Bioconductor package, Ringo [\cite{Toedling_2007}\cite{Toedling_2007}]. \cite{Toedling_2007}.  Control probes were discarded and the data from each array was initially normalized using the Tukey biweight mean method. Paired pre- and post-heat shock log2(IP/Input) values were adjusted using a linear regression fit (setting m = 1 and b = 0) to compensate for variation in data compression between arrays. To determine differential methylation, the input channels from the paired datasets were loess normalized and the results were used to calculate a threshold value, T, equivalent to 3 s.d from the mean. Applying this back to each pairwise comparison of pre- and post-heat shock log2(IP/Input) ratios, differential methylation was assigned to a given probe when the absolute value of the paired ratio exceeded T. For our studies, we concentrated on runs of at least 3 adjacent probes identified as differentially hyper- or hypo-methylated and localized to a gene body (plus 1000bp upstream). The R programing language [2] was used for data processing, including the generation of bedGraph and GFF tracks for visualization in IGV \cite{Robinson_2011}\cite{Robinson_2011}\cite{Robinson_2011}]. \cite{Robinson_2011}\.