Convolution filter for smoothing transcript coverages

Filter out transcripts with sharp drops / rises in coverage, to be used in filter_coverage to remove transcripts with potential misalignments / internal priming etc. Filtering is done by convolving the coverage with a kernal of 1s and -1s (e.g. c(1, 1, -1, -1), where the width of the 1s and -1s are determined by the width parameter), and check if the maximum absolute value of the convolution is below a threshold. If the convolution is below the threshold, TRUE is returned, otherwise FALSE.

Usage

convolution_filter(x, threshold = 0.15, width = 2, trim = 0.05)

Arguments

x

numeric vector of coverage values

threshold

numeric, the threshold for the maximum absolute value of the convolution

width

numeric, the width of the 1s and -1s in the kernal. E.g. width = 2 will result in a kernal of c(1, 1, -1, -1)

trim

numeric, the proportion of the coverage values to ignore at both ends before convolution.

Value

logical, TRUE if the transcript passes the filter, FALSE otherwise

Examples

# A >30% drop in coverage will fail the filter with threshold = 0.3
convolution_filter(c(1, 1, 1, 0.69, 0.69, 0.69), threshold = 0.3)
#> [1] FALSE
convolution_filter(c(1, 1, 1, 0.71, 0.7, 0.7), threshold = 0.3)
#> [1] TRUE