Nothing
#' Predict TSSs of miRNA
#'
#' Search for putative TSSs of miRNA, together with integrating available data such
#' as H3K4me3 data, Pol II data, miRNA expression data, and protein-coding gene
#' data, as well as provide the transcriptional regulation relationship between TF
#' and miRNA.
#'
#' @param bed_merged Peaks from ChIP-seq data to be provided for analysis can be
#' H3K4me3 peaks, Pol II peaks or both. Notice that peaks are supposed to be
#' merged(see also \code{\link{peak_merge}}) before \code{find_TSS} if using
#' only one kind of peak data, while peaks should be firstly merged and then
#' join together(see also \code{\link{peak_join}}) if both H3K4me3 data and
#' Pol II are input.
#' @param expressed_mir This parameter allows users to specify certain miRNAs,
#' the TSSs of which they want to search for by providing a list of
#' miRNAs(e.g., expressed miRNAs in a certain cell-line). If
#' \code{expressed_mir} is not specified, the default value of the parameter
#' is "\code{all}" and the function will acquiescently employ all the miRNAs
#' currently listed on "\code{miRbase}" database.
#' @param flanking_num A parameter in Eponine model to detect TSSs. It is
#' concluded that a peak signal with flanking regions of C-G enrichment are
#' important to mark TSSs. The default value is 1000.
#' @param threshold The threshold for candidate TSSs scored with Eponine method. The
#' default value is 0.7.
#' @param ignore_DHS_check The process of DHS_check further assists to filter
#' putative TSSs. When there is a DHS peak that locates within 1 kb upstream
#' of a putative TSS, this predicted TSS will be retained for its character is
#' consistent with that of an authentic TSS. Or the TSSs with no DHSs locating
#' within 1 kb upstream of them would be discarded.
#' @param DHS ChIP-seq data of DNase I hypersensitive sites(DHSs).
#' @param allmirdhs_byforce When we use DHS data to check the validity of TSSs,
#' there is a possibility where no DHSs locates within 1 kb upstream of all
#' putative TSSs and all these putative TSSs might be filtered out by our
#' method resulting no outputs. While "\code{allmirdhs_byforce} = TRUE", it
#' ensures to output at least 1 most possible TSS even if the nearest DHS
#' signal locates more than 1 kb upstream of this TSS.
#' @param expressed_gene Users can specify genes expressed in certain
#' cell-lines that are analyzed. Or the default value is "\code{all}", which
#' means all the expressed genes annotated on Ensemble will be employed.
#' @param allmirgene_byforce While integrating expressed_gene data to improve
#' prediction, there might be a circumstance where all the putative TSS are
#' discarded. To prevent this condition, users are allowed to use
#' "\code{allmirgene_byforce = TRUE}" to ensure at least 1 putative TSS for
#' each miRNA will be output.
#' @param seek_tf With the result of predicted TSSs, seek_tf provides users with
#' an option to predict related TFs for miRNA. The data of transcription
#' factors refer to \code{JASPAR2018} database.
#' @param tf_n TFBS locates on the upstream of the TSS of a certain TF, which is
#' considered as the promoter region. \code{tf_n} set the length of promoter
#' region for predicting transcription regulation between miRNAs and TFs.
#' @param min.score The threshold for scoring transcription factor binding sites. A
#' single absolute value between 0 and 1.
#'
#'
#'
#' @return The first part of the result returns details of predicted TSSs,
#' composed of seven columns: \code{mir_name, chrom, stem_loop_p1,
#' stem_loop_p2, strand mir_context, tss_type gene and predicted_tss}:
#'
#' \code{mir_name}: Name of miRNA.
#'
#' \code{chrom}: Chromosome.
#'
#' \code{stem_loop_p1}: The start site of a stem-loop.
#'
#' \code{stem_loop_p2}: The end site of a stem-loop.
#'
#' \code{strand}: Polynucleotide strands. (\code{+/-})
#'
#' \code{mir_context}: The relative positon relationship between stem-loop and
#' protein-coding gene. (\code{intra/inter})
#'
#' \code{tss_type}: Four types of predicted TSSs. See the section below
#' \code{TSS types} for details.
#' (\code{host_TSS/intra_TSS/overlap_inter_TSS/inter_TSS})
#'
#' \code{gene}: Ensembl gene ID
#'
#' \code{predicted_tss}: Predicted transcription start sites(TSSs).
#'
#' \code{pri_tss_distance}: The distance between a predicted TSS and the start
#' site of the stem-loop.
#'
#'
#' @section TSS types: TSSs are catalogued into 4 types as below.
#'
#' \code{host_TSS} The TSSs of miRNA that are close to the TSS of
#' protein-coding gene implying they may share the same TSS, on the condition
#' where \code{mir_context} is "\code{intra}". (See above:
#' \code{Value-mir_context})
#'
#' \code{intra_TSS} The TSSs of miRNA that are NOT close to the TSS of
#' the protein-coding gene, on the condition where \code{mir_context} is
#' "\code{intra}".
#'
#' \code{overlap_inter_TSS} The TSSs of miRNA are catalogued as
#' "\code{overlap_inter_TSS}" when the pri-miRNA gene overlaps with Ensembl
#' gene, on the condition where "\code{mir_context}" is "\code{inter}".
#'
#' \code{inter_inter_TSS} The TSSs of miRNA are catalogued as
#' "\code{inter_inter_TSS}" when the miRNA gene does NOT overlap with Ensembl
#' gene, on the condition where "\code{mir_context}" is "\code{inter}".
#'
#' (See Xu HUA et al 2016 for more details)
#'
#'
#'
#' @section Log:
#'
#' The second part of the result returns logs during the process of
#' prediction: \code{find_nearest_peak_log} If no peaks locate in the
#' upstream of a stem-loop to help determine putative TSSs of miRNA, we will
#' fail to find the nearest peak and this miRNA will be logged in
#' \code{find_nearest_peak_log}.
#'
#' \code{eponine_score_log} For a certain miRNA, if none of the candidate
#' TSSs scored with \code{Eponine method} meet the threshold we set, we will
#' fail to get an eponine score and this miRNA will be logged in
#' \code{eponine_score_log}.
#'
#' \code{DHS_check_log} For a certain miRNA, if no DHS signals locate
#' within 1 kb upstream of each putative TSSs, these putative TSSs will be
#' filtered out and this miRNA will be logged in \code{DHS_check_log}.
#'
#' \code{gene_filter_log} For a certain miRNA, when integrating
#' expressed_gene data to improve prediction, if no putative TSSs are
#' confirmed after considering the relative position relationship among TSSs,
#' stem-loops and expressed genes, this miRNA will be filtered out and logged
#' in \code{gene_filter_log}.
#'
#' @section Reference: Xu Hua, Luxiao Chen, Jin Wang*, Jie Li* and Edgar
#' Wingender*, Identifying cell-specific microRNA transcriptional start sites.
#' Bioinformatics 2016, 32(16), 2403-10.
#'
#'
#' @examples
#'
#' bed_merged <- data.frame(
#' chrom = c("chr1", "chr1", "chr1", "chr1", "chr2"),
#' start = c(9910686, 9942202, 9996940, 10032962, 9830615),
#' end = c(9911113, 9944469, 9998065, 10035458, 9917994),
#' stringsAsFactors = FALSE)
#' bed_merged <- as(bed_merged, "GRanges")
#'
#' \dontrun{
#' ownmiRNA <- find_tss(bed_merged, expressed_mir = "hsa-mir-5697",
#' ignore_DHS_check = TRUE,
#' expressed_gene = "all",
#' allmirgene_byforce = TRUE)
#' }
#'
#' @export
find_tss <- function(bed_merged, expressed_mir = "all",
flanking_num = 1000, threshold = 0.7,
ignore_DHS_check = TRUE,
DHS, allmirdhs_byforce = TRUE,
expressed_gene = "all", allmirgene_byforce = TRUE,
seek_tf = FALSE, tf_n = 1000, min.score = 0.8) {
h3_mir_flank <- find_nearest_peak(bed_merged, expressed_mir)
mir_eponine_score <- eponine_score(h3_mir_flank$success$mir_name,
h3_mir_flank$success$chrom,
h3_mir_flank$success$stem_loop_p1,
h3_mir_flank$success$stem_loop_p2,
h3_mir_flank$success$strand,
h3_mir_flank$success$peak_p1,
h3_mir_flank$success$peak_p2,
flanking_num, threshold)
candidate_tss <- find_candidate_tss(mir_eponine_score$success$mir_name,
mir_eponine_score$success$chrom,
mir_eponine_score$success$stem_loop_p1,
mir_eponine_score$success$stem_loop_p2,
mir_eponine_score$success$strand,
mir_eponine_score$success$tss_p1,
mir_eponine_score$success$tss_p2,
mir_eponine_score$success$eponine_score,
ignore_DHS_check,
DHS, allmirdhs_byforce)
gene_fliter_tss <- tss_filter(candidate_tss$mir_df$mir_name,
candidate_tss$mir_df$chrom,
candidate_tss$mir_df$stem_loop_p1,
candidate_tss$mir_df$stem_loop_p2,
candidate_tss$mir_df$strand,
candidate_tss$mir_df$tss_p1,
candidate_tss$mir_df$tss_p2,
expressed_gene, allmirgene_byforce)
if (any(h3_mir_flank$fail_nearpeak !=
"No one miRNA fail to find its nearest peaks")) {
if (length(h3_mir_flank$fail_nearpeak) == 1) {
log_1 <- paste(h3_mir_flank$fail_nearpeak,
"fail to find its nearest peaks", " ")
} else {
log_1 <- paste(h3_mir_flank$fail_nearpeak, collapse = ", ")
log_1 <- paste(log_1, "fail to find their nearest peaks", " ")
}
} else {
log_1 <- h3_mir_flank$fail_nearpeak
}
if (any(mir_eponine_score$fail_eponine != "All miRNAs have eponine socres")) {
if (length(mir_eponine_score$fail_eponine) == 1) {
log_2 <- paste(mir_eponine_score$fail_eponine,
"fail to have eponine socres", " ")
} else {
log_2 <- paste(mir_eponine_score$fail_eponine, collapse = ", ")
log_2 <- paste(log_2, "fail to have eponine socres", " ")
}
} else {
log_2 <- mir_eponine_score$fail_eponine
}
if (any(candidate_tss$fail_DHS != "Do not have a DHS check") &&
any(candidate_tss$fail_DHS != "All miRNA have DHS verified")) {
if (length(candidate_tss$fail_DHS) == 1) {
log_3 <- paste(candidate_tss$fail_DHS, "fail to have a DHS check", " ")
} else {
log_3 <- paste(candidate_tss$fail_DHS, collapse = ", ")
log_3 <- paste(log_3, "fail to have a DHS check", " ")
}
} else {
log_3 <- candidate_tss$fail_DHS
}
if (any(gene_fliter_tss$fail_genefilter != "All miRNAs have tss")) {
if (length(gene_fliter_tss$fail_genefilter) == 1) {
log_4 <- paste(gene_fliter_tss$fail_genefilter, "fail to have a tss", " ")
} else {
log_4 <- paste(gene_fliter_tss$fail_genefilter, collapse = ", ")
log_4 <- paste(log_4, "fail to have a tss", " ")
}
} else {
log_4 <- gene_fliter_tss$fail_genefilter
}
if (allmirgene_byforce == TRUE) {
log_4 <- "All miRNAs have tss"
}
if (seek_tf == TRUE) {
a <- mir_tf(gene_fliter_tss$tss_df$mir_name, gene_fliter_tss$tss_df$chrom,
gene_fliter_tss$tss_df$strand,
gene_fliter_tss$tss_df$predicted_tss,
tf_n, min.score)
result <- dplyr::left_join(gene_fliter_tss$tss_df, a, by = "mir_name")
} else {
result <- gene_fliter_tss$tss_df
}
list(tss_df = result,
find_nearest_peak_log = log_1,
eponine_score_log = log_2,
DHS_check_log = log_3,
gene_filter_log = log_4)
}
Any scripts or data that you put into this service are public.
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.