change count thresholds to be fractions instead of integers

[kelly-sovacool]

previously, ccbrpipeliner used fractions which was more portable across groups of different sizes.

current code straight from nidap:

reneeTools/R/filter.R

Lines 247 to 300 in 79e612e

	remove_low_count_genes <- function(counts_matrix, sample_metadata,
	gene_names_column,
	groups_column,
	use_cpm_counts_to_filter = TRUE,
	Use_Group_Based_Filtering = FALSE,
	Minimum_Count_Value_to_be_Considered_Nonzero = 8,
	Minimum_Number_of_Samples_with_Nonzero_Counts_in_Total = 7,
	Minimum_Number_of_Samples_with_Nonzero_Counts_in_a_Group = 3) {
	value <- NULL
	df <- counts_matrix
	df <- df[stats::complete.cases(df), ]
	## duplicate Rows should be removed in Clean_Raw_Counts template
	# df %>% dplyr::group_by(.data[[gene_names_column]]) %>% summarise_all(sum) %>% as.data.frame() -> df
	# print(paste0("Number of features before filtering: ", nrow(df)))
	## USE CPM Transformation
	if (use_cpm_counts_to_filter == TRUE) {
	trans.df <- df
	trans.df[, -1] <- edgeR::cpm(as.matrix(df[, -1]))
	counts_label <- "Filtered Counts (CPM)"
	} else {
	trans.df <- df
	counts_label <- "Filtered Counts"
	}
	if (Use_Group_Based_Filtering == TRUE) {
	rownames(trans.df) <- trans.df[, gene_names_column]
	trans.df[, gene_names_column] <- NULL
	counts <- trans.df >= Minimum_Count_Value_to_be_Considered_Nonzero # boolean matrix
	tcounts <- as.data.frame(t(counts))
	colnum <- dim(counts)[1] # number of genes
	tcounts <- merge(sample_metadata[groups_column], tcounts, by = "row.names")
	tcounts$Row.names <- NULL
	melted <- reshape2::melt(tcounts, id.vars = groups_column)
	tcounts.tot <- dplyr::summarise(dplyr::group_by_at(melted, c(groups_column, "variable")), sum = sum(value))
	tcounts.group <- tcounts.tot %>%
	tidyr::pivot_wider(names_from = "variable", values_from = "sum")
	colSums(tcounts.group[(1:colnum + 1)] >= Minimum_Number_of_Samples_with_Nonzero_Counts_in_a_Group) >= 1 -> tcounts.keep
	df.filt <- trans.df[tcounts.keep, ]
	df.filt %>% tibble::rownames_to_column(gene_names_column) -> df.filt
	} else {
	trans.df$isexpr1 <- rowSums(as.matrix(trans.df[, -1]) > Minimum_Count_Value_to_be_Considered_Nonzero) >= Minimum_Number_of_Samples_with_Nonzero_Counts_in_Total
	df.filt <- as.data.frame(trans.df[trans.df$isexpr1, ])
	}
	# colnames(df.filt)[colnames(df.filt)==gene_names_column] <- "Gene"
	# print(paste0("Number of features after filtering: ", nrow(df.filt)))
	return(df.filt)
	}

[kelly-sovacool]

@phoman14 do you have any thoughts on this?

[phoman14]

do you mean that if a dataframe has 14 samples then Minimum_Number_of_Samples_with_Nonzero_Counts_in_Total = 0.5 instead of Minimum_Number_of_Samples_with_Nonzero_Counts_in_Total = 7?

[kelly-sovacool]

do you mean that if a dataframe has 14 samples then Minimum_Number_of_Samples_with_Nonzero_Counts_in_Total = 0.5 instead of Minimum_Number_of_Samples_with_Nonzero_Counts_in_Total = 7?

Yes exactly. This is @kopardev's suggestion.

[phoman14]

I think this is a fine way to do it.
In my head It is easier to specify the exact number but if we make the input a fraction we could always calculate the fraction upstream from the exact number.
The other consideration is the input format. We should include an error check to make sure the input is in the correct format. I could see a user not understanding the format in enter any of the following 0.5, 50% or 50