Standardmäßig ist der Prompt ausgewählt, der zuerst die Quelle prüft. Sie können zu einem direkten Befehl wechseln oder eine lokale Kopie herunterladen.
Quelldateien prüfen
Lesen Sie SKILL.md und alle von SkillsMP angezeigten Begleitdateien, bevor Sie sich für eine Installation entscheiden.
Mit Codex oder Claude installieren Kopieren Sie diesen Prompt, fügen Sie ihn in Codex, Claude oder einen anderen Assistant ein und lassen Sie die Skill-Seite prüfen und installieren.
Ein direkter Befehl überspringt den Prüf-Prompt. Prüfen Sie die Quelle, bevor Sie ihn ausführen.
"Visualize ChIP-seq signal around features of interest" -> Generate normalized signal tracks (bigWig), heatmaps centered on TSS/peaks, average profile plots, and genome-browser views — with normalization that supports the biological claim (within-sample vs cross-sample vs spike-in scaled).
CLI (production): deepTools -> -> /
bamCoverage
computeMatrix
plotHeatmap
plotProfile
CLI (config-driven tracks): pyGenomeTracks (replaces Gviz for many use cases)
R (publication): Gviz, EnrichedHeatmap, ChIPseeker tag heatmaps
GUI: IGV with batch scripts for reproducible screenshots
The single most consequential choice is bigWig normalization — it determines whether visual comparison reflects biology. Get this right before generating any heatmap or browser view.
bigWig Normalization Decision Tree
Goal
Method
When to use
Within-sample profile of a single ChIP
--normalizeUsing CPM
Standard; reads per million; comparable within one library
Within-sample, length-aware
--normalizeUsing BPM
TPM-analog; useful for variable-width regions; less common for ChIP-seq
HDACi / BETi / EZH2i; see chip-seq/spike-in-normalization
ChIP vs input ratio
bamCompare --operation log2
Visualize enrichment over input
ChIP vs input control-subtracted
bamCompare --operation subtract
Absolute signal above background
ChIP vs input SES-corrected
bamCompare --scaleFactorsMethod SES --operation log2
More robust to library size; uses signal-extraction-scaling
ENCODE convention: RPGC with read-length-matched effective genome size. For visual comparison of treatment vs control on a fold-change biology, log2 bamCompare against shared input.
Spike-in scaled tracks (the right way):
# Compute scale factor from spike-in reads (ChIP-Rx Drosophila or CUT&RUN E. coli)
SCALE=$(echo"scale=6; 1.0 / $SPIKE_IN_READS_M" | bc) # 1 per million spike reads
bamCoverage -b chip.bam -o chip.bw --scaleFactor $SCALE --binSize 10
# DO NOT also pass --normalizeUsing; deepTools multiplies the two factors, reintroducing depth normalization
For pipeline-driven figure generation across multiple regions, pyGenomeTracks is easier to script than Gviz. For one-off publication figures with complex annotation, Gviz remains useful.
R: Gviz and EnrichedHeatmap
library(Gviz)
library(GenomicRanges)
library(TxDb.Hsapiens.UCSC.hg38.knownGene)
chr <-'chr1'; start <- 1e6; end <- 1.1e6
itrack <- IdeogramTrack(genome ='hg38', chromosome = chr)
gtrack <- GenomeAxisTrack()
dtrack <- DataTrack(range='sample.bw', genome ='hg38',
type ='histogram', name ='ChIP', col.histogram ='darkblue')
grtrack <- GeneRegionTrack(TxDb.Hsapiens.UCSC.hg38.knownGene,
genome ='hg38', chromosome = chr, name ='Genes')
plotTracks(list(itrack, gtrack, dtrack, grtrack), from = start, to = end, chromosome = chr)
library(EnrichedHeatmap)
library(rtracklayer)# Normalize bigWig signal to a matrix around target sites
signal <- import('sample.bw')
tss <- promoters(txdb, upstream =0, downstream =1)
mat <- normalizeToMatrix(signal, tss, extend =3000, mean_mode ='w0', w =50)# Heatmap with customization
EnrichedHeatmap(mat, name ='Signal', col =c('white','red'),
top_annotation = HeatmapAnnotation(lines = anno_enriched()))
ChIPseeker Tag Heatmap (R)
library(ChIPseeker)
library(TxDb.Hsapiens.UCSC.hg38.knownGene)
peaks <- readPeakFile('peaks.narrowPeak')
promoter <- getPromoters(TxDb = TxDb.Hsapiens.UCSC.hg38.knownGene,
upstream =3000, downstream =3000)
tagMatrix <- getTagMatrix(peaks, windows = promoter)# Tag heatmap and average profile# tagHeatmap in ChIPseeker >= 1.36 takes palette (RColorBrewer name), not xlim/color;# xlim is read from the tagMatrix window. plotAvgProf still uses xlim/conf.
tagHeatmap(tagMatrix, palette ='Reds')
plotAvgProf(tagMatrix, xlim =c(-3000,3000), conf =0.95,
xlab ='Distance from TSS (bp)', ylab ='Peak density')
bamCoverage -- --normalizeUsing and --scaleFactor conflict
Trigger: Passing both --normalizeUsing CPM and --scaleFactor X.
Mechanism: deepTools multiplies the --scaleFactor value by the factor computed from --normalizeUsing, so passing both compounds them and reintroduces library-depth normalization on top of the spike-in factor.
Symptom: Spike-in scaling appears to have no effect; tracks look like CPM.
Fix: Use ONE — --scaleFactor alone for spike-in; --normalizeUsing alone otherwise. Never both.
bamCompare -- log2 with zeros produces -Inf
Trigger:bamCompare --operation log2 without pseudocount; many bins have zero reads.
Mechanism: log2(0/x) = -Inf; downstream tools (plotHeatmap) may color these as NaN or fail.
Fix: Add --pseudocount 1 to both samples; or use --skipZeroOverZero to skip bins with zero in both samples.
computeMatrix -- Stranded bigWig vs unstranded reference points
Trigger: Using stranded bigWigs (separate plus/minus) with reference-point mode on a BED without strand info.
Mechanism: computeMatrix doesn't auto-detect strand; signal is plotted in genomic-strand orientation, breaking TSS-centered plots.
Fix: Use unstranded merged bigWig OR ensure BED has strand column 6.
plotHeatmap --kmeans -- Order depends on first sample only
Trigger: Using k-means with multiple samples and expecting consistent clustering.
Mechanism: k-means clusters by signal in the first -S bigWig only; other samples are plotted in the same row order.
Fix: Order samples in -S so the most-discriminating one is first; for combined clustering across samples, use --hclust or run k-means externally on combined matrix.
Spike-in scaled bigWig -- Wrong scale factor direction
Trigger: Computing scale_factor = spike_reads / 1e6 and passing to --scaleFactor.
Mechanism: deepTools multiplies signal by scaleFactor; the INVERSE is correct (sample with fewer spike reads gets larger scale factor to compensate).
Symptom: Treatment samples appear lower than control even when biology says higher.
Fix:scale_factor = MIN(spike_reads_all_samples) / spike_reads_this_sample. Always verify against known internal-control regions (blacklist should show no signal change post-scaling).
Gviz / EnrichedHeatmap -- Memory failure on whole-genome bigWigs
Trigger: Loading a 3 GB bigWig into R as a GRanges.
Mechanism: Gviz loads the entire bigWig into memory for genome-wide views.
Fix: Use chromosome parameter to restrict; use import.bw(con, which = GRanges(...)) to subset; consider pyGenomeTracks for whole-chromosome views.
pyGenomeTracks -- INI parsing strict
Trigger: Custom INI keys not recognized; or section names with spaces.
