Ecosyste.ms: Awesome

An open API service indexing awesome lists of open source software.

Awesome Lists | Featured Topics | Projects

https://github.com/cmdcolin/macs-accessories

visualizations for macs output
https://github.com/cmdcolin/macs-accessories

Last synced: 4 days ago
JSON representation

visualizations for macs output

Awesome Lists containing this project

README 

        
macs-accessories

================



The package Model-based analysis for ChIP-seq (MACS) is very useful software for peak finding on ChIP-seq data. The outputted files include peaks in BED and XLS formats and the 'tag pileup' in wiggle file format. We use macs-accessories for additional data analysis from the files including visualizations and calculations of normalized difference (NormDiff) scores from Zheng et al. (2010)



```{r setup, dev='png',cache=FALSE,echo=FALSE}

debug=FALSE

opts_knit$set(upload.fun = imgur_upload) # upload all images to imgur.com

source('readplot.R')

source('knitr.R')

```



We created a 'wiggle class', an S3 R object, for loading MACS output files and doing additional data analysis. The class automatically loads peak files (bed,xls,csv) and and wiggle files (macs output format)



```{r setup2, cache=TRUE}

wig1=WiggleClass('S96')

wig2=WiggleClass('HS959')

wig1$loadWiggles()

wig2$loadWiggles()

```



The normalized difference score gives us on average the expected value of the ChIP-seq subtracted from the input data using a simple random model. For $A$, $B$ representing chip-seq and input control data respectively



$A\sim Poisson(f+g)$

$B\sim Poisson(cg)$



Then the NormDiff score $Z$ is defined as



$$Z(x_i)=\frac{A(x_i)-B(x_i)/c}{\hat\sigma}$$



We use the data to estimate scaling factor $c$ and variance $\hat\sigma$



We can look at the average normalized difference scores of the peaks, and see how this compares with the same location in other experiments, simply by comparing wig1 with wig2 Z scores for the peak.



```{r d2,dev='png',cache=TRUE}

wig1$estimateScalingFactor()  

wig1$estimateVarianceAll()   

wig2$estimateScalingFactor()   

wig2$estimateVarianceAll()

wz1=wig1$Z(wig1$peaks)

wz2=wig2$Z(wig1$peaks)

wz4=wig2$Z(wig2$peaks)

wz3=wig1$Z(wig2$peaks)

r1=plotMaxAvgZscore('Max Avg  S96 peak NormDiff score vs HS959 synteny w=100', wig1, wig2, wz1, wz2,'#bb0000', '#001199')

r2=plotMaxAvgZscore('Max Avg HS959 peak NormDiff score vs S96 synteny w=100', wig2, wig1, wz4, wz3, '#bb0000','#009900')

```



We can calculate NormDiff scores for the whole genome using 

```{r zallcalc, cache=TRUE}

wza1=wig1$Zall()

wza2=wig2$Zall()

```



We can observe the distribution of NormDiff scores



```{r zall, dev='png',cache=TRUE}

datasort=as.numeric(wza1[[1]][,4])

d <- density(datasort,adjust=1.4) # returns the density data

plot(d, main='Kernel density of NormDiff scores') # plots the results 

polygon(d, col="#BB2222CC", border="#222244")

clone=datasort

qqnorm(clone)

qqline(clone,col=2)

```



By taking the pvalues of all Zscores we can see which ones are significant according to a p-value $P(\bar x \leq X)$

```{r color, dev='png', cache=TRUE}

plotZscoreColor('HS959 pvalue for S96 peaks', wig1,wig2,wz1,wz2)

```



We want to use hypothesis testing to observe NormDiff scores that are highly different from the background. We used a likelihood ratio test sgnificance level of 1%. When comparing ChIP-seq experiments from different strains of yeast.



```{r cutoff, dev='png',cache=TRUE}

ret=plotZscoreCutoff('S96 peaks vs HS959 synteny (Hypothesis testing)',wig1,wig2,wz1,wz2,0.05)

```