# build dataframe for doing aov: signal = numeric values log2-transformed, group = subtypes
    # no need to check column ordering vs metadata AK ID ordering as they're both ordered by AK ID already
    data <- data.frame( signal=t(as.matrix( mcols(t)[,grep(x=colnames(mcols(t)), pattern="_H3K27ac")] )), group=factor(metadata$subtype_final) )
    data$signal[ data$signal <= 0 ] <- 1 # set minimum value of signal to 1
data$signal <- log2(data$signal) # log2 transform signal
    fit <- aov( signal ~ group, data ) # fit model
    active.anova[r,"pvalue"] <- unlist(anova(fit)[5])[1] # get unadjusted anova pvalue
    # calculate log2FC for each subtype in turn
    # these are log2 values so just subtract:
    active.anova[r,"log2FC_IDH"] <- mean( data$signal[ data$group=="IDH" ] ) - mean( data$signal[ data$group!="IDH" ] )
    active.anova[r,"log2FC_MES"] <- mean( data$signal[ data$group=="MES" ] ) - mean( data$signal[ data$group!="MES" ] )
    active.anova[r,"log2FC_RTK_I"] <- mean( data$signal[ data$group=="RTK_I" ] ) - mean( data$signal[ data$group!="RTK_I" ] )
    active.anova[r,"log2FC_RTK_II"] <- mean( data$signal[ data$group=="RTK_II" ] ) - mean( data$signal[ data$group!="RTK_II" ] )

findMotifsGenome.pl input.bed hg19 output_dir -size 2000 -len 8 -S 100