MEPmod | R Documentation |
MEPmod searches microexon-tags in 45 conserved microexon clusters in plants using gapped Position Weight Matrix (PWM). The only input file is plant genomic sequences or a plant genome. The sequences on both plus and minus strands will be scanned.
MEPmod(
genome,
min.score = "80%",
clusters = seq_len(45),
include.intronLoss = TRUE,
span = 20000,
min.intron = 20,
max.intron = 10000,
cores = 4
)
genome |
the path(s) to the fasta file(s) or a 'DNAStringSet' object. Any contig sequence < 10 kb will be excluded. |
min.score |
a character string containing a percentage specifying the
minimum score of each exon block (e.g. |
clusters |
an integer vector between 1 to 45 specifying microexon-tag clusters to search (default: all the 45 clusters) |
include.intronLoss |
|
span |
the maximum spanning region of the microexon-tag (default: 20 kb). |
min.intron |
minimum intron size |
max.intron |
maximum intron size |
cores |
number of cores to use. This parameter will pass to mclapply as *mc.cores* from parallel package. |
A GRanges of microexons with 8 metadata columns:
NT, 108 bp DNA sequence of microexon-tag (72 bp for cluster 1 and 2).
AA, 36 aa protein sequence translated from NT (24 aa for cluster 1 and 2).
region, the span region of microexon-tag.
block.starts, the genomic coordinates of block start positions (exon, intron, exon, ...).
block.sizes, block sizes (exon, intron, exon, ...).
score, match score of the microexon-tag. Large score indicate high confidence.
isMicroExon, if the microexon has any flanking intron loss, FALSE; otherwise, TRUE.
cluster, which cluster this microexon-tag belongs to.
suppressPackageStartupMessages(library(BSgenome))
if (!requireNamespace("BSgenome.Athaliana.TAIR.TAIR9", quietly = TRUE))
BiocManager::install("BSgenome.Athaliana.TAIR.TAIR9")
library(BSgenome.Athaliana.TAIR.TAIR9)
Chr1<-BSgenome.Athaliana.TAIR.TAIR9[["Chr1"]]
Chr1<-DNAStringSet(Chr1)
names(Chr1)<-'Chr1'
x<-subseq(Chr1, 1, 1e6)
x
mep<-MEPmod(genome=x)
mep
# GRanges object with 2 ranges and 8 metadata columns:
# seqnames ranges strand | NT AA
# <Rle> <IRanges> <Rle> | <DNAStringSet> <AAStringSet>
# [1] Chr1 456757-456761 + | TTTGATGCTA...ACTGTCTGAC FDARTAWSQC...AFGAVESLSD
# [2] Chr1 601547-601560 - | AAAGTTGAAT...ACATTCAGAT KVEFKDNEWK...RLDCLLKHSD
# region block.starts block.sizes score isMicroExon cluster
# <IRanges> <IntegerList> <IntegerList> <numeric> <logical> <integer>
# [1] 456595-456893 456595,456647,456757,... 52,110,5,... 0.9800 TRUE 8
# [2] 601327-601700 601653,601561,601547,... 48,92,14,... 0.9241 TRUE 35
# -------
# seqinfo: 1 sequence from an unspecified genome; no seqlengths
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.