-
Notifications
You must be signed in to change notification settings - Fork 2
/
sm6_RNAseq_preprocessing.Rmd
254 lines (167 loc) · 17.7 KB
/
sm6_RNAseq_preprocessing.Rmd
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
---
title: 'STAT540 - Seminar 6: RNA-seq preprocessing - BAM files to count tables'
output:
github_document
---
## Attributions
This seminar was developed by Keegan Korthauer
# Overview
This seminar focuses on a portion of the upstream preprocessing of RNA-seq: going from aligned reads (BAM files) to count tables that will be used in differential expression analysis. Note that we are skipping the step of going from raw reads (fastq files) to aligned reads (BAM files), as this is outside the scope of this course. Refer to the lecture materials on High-dimensional genomics assays & data for high-level details and resources.
In order to use RNA-Seq data for the purposes of differential expression analysis, we need to obtain an expression value for each gene (or transcript). The digital nature of RNA-seq data allows us to use the number of reads (an integer count) which align to a specific feature (gene or transcript) as its expression value. In this seminar, we will explore how to use BAM (or SAM) files to generate a count table of reads aligning to the human transcriptome.
By the end of this seminar, you should be able to:
- use the [`Rsamtools`](https://bioconductor.org/packages/release/bioc/html/Rsamtools.html) package to **sort and index a BAM file**
- use the [`Rsubread`](https://bioconductor.org/packages/release/bioc/html/Rsubread.html) package to **count the number of reads that align to each gene/transcript**
- extract raw counts from the output of `featureCounts`, and calculate normalized expression units using the `cpm()` and `rpkm()` functions in `edgeR`
# Loading libraries
Before we begin, we'll need to load the necessary R packages for this seminar. It is likely you don't already have all of these installed on your machine unless you've done a similar analysis in the past. If this is the case, you'll need to run this code chunk to install them (currently set to `eval = FALSE`):
```{r, eval=FALSE}
library(BiocManager)
install(c("Rsamtools", "Rsubread", "RNAseqData.HNRNPC.bam.chr14", "edgeR"))
```
After they are successfully installed, we'll load them for use in our session.
```{r, message = FALSE}
library(Rsamtools)
library(Rsubread)
library(RNAseqData.HNRNPC.bam.chr14)
library(ggplot2)
theme_set(theme_bw())
library(dplyr)
library(edgeR)
library(testthat)
```
# BAM/SAM - Aligned Sequence data format
SAM (sequence alignment map) and BAM (the binary version of SAM) files are the preferred output of most alignment tools. SAM and BAM files carry the same information. The only difference is that SAM is human readable, meaning that you can visually inspect each of the reads. You can learn more about the SAM file format [here](https://en.wikipedia.org/wiki/SAM_(file_format)).
The [`Rsamtools`](https://bioconductor.org/packages/release/bioc/html/Rsamtools.html) package is one of many software tools that have been developed for working with SAM/BAM alignment files - it is an R implementation of the widely used command line tool [Samtools](http://www.htslib.org/). The [`Rsubread`](https://bioconductor.org/packages/release/bioc/html/Rsubread.html) package is a tool that summarizes SAM/BAM files into count tables - it is an R implementation of the command line program [Subread](http://subread.sourceforge.net/) which has tools for alignment and quantification. The tool we'll use within Subread is the [featureCounts](https://academic.oup.com/bioinformatics/article/30/7/923/232889) tool.
Although you could use the standalone command line versions of both of these tools, to make things as convenient as possible, we'll use the R implementations and run today's seminar entirely within R.
# HeLa cell line transcription
The alignment BAM file we will be working with was created by aligning raw reads from an RNA-seq experiment of HNRNPC gene knockdown cell line and control HeLa cells ([Zarnack et al. 2012](http://europepmc.org/article/MED/23374342)). We will specifically work with one of the control HeLa cell samples. [HeLa cells](https://en.wikipedia.org/wiki/HeLa) are immortalized cervical cancer cells derived taken from cancer patient Henrietta Lacks in 1951. Henrietta Lacks was a 31-year-old African-American mother of five and a patient at Johns Hopkins Hospital. She died from cervical cancer on October 4, 1951. Her cells were taken from her without her knowledge or consent, and the HeLa cell line is now the oldest and most used immortal cell line in scientific research. To read more about Henrietta's legacy and the ethical issues involved in procuring and using her cells, read [The Immortal Life of Henrietta Lacks](http://rebeccaskloot.com/the-immortal-life/).
In this tutorial, we are working with an already aligned BAM file, but the raw reads (fastq files) from this experiment are available in the [European Nucleotide Archive](https://www.ebi.ac.uk/ena/browser/view/PRJEB3048).
# RNA-seq upstream analysis pipeline
In this tutorial we will only be dealing with a portion of the RNA-seq analysis pipeline. Here is an overview of the entire pipeline (image source: [Yang & Kim 2015](https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4742321/)). Note that this figure is simplistic in its depiction of downstream analysis - differential expression is **not** the only downstream analysis in typical experiments. For example, we'll learn about things like gene network and enrichment analyses, clustering, and supervised learning later in the course.
As you can see, there are many steps to go from the raw reads to the aligned bam files (red to light blue). These have been done for us in this example, as we are making use of the aligned data in the Bioconductor package [RNAseqData.HNRNPC.bam.chr14](http://bioconductor.org/packages/release/data/experiment/html/RNAseqData.HNRNPC.bam.chr14.html). Briefly, this package has preprocessed the data from [Zarnack et al. 2012](http://europepmc.org/article/MED/23374342). The reads (paired-end) were aligned to the full human genome (version hg19) with TopHat2. In addition, the data have been subset to only include reads mapped to Chromosome 14. This will reduce the computational time for our tutorial compared to running on the entire genome.
**In this tutorial, we are only going to deal with the dark blue box: "Expression Quantification". Specifically, we are going to learn how to convert a SAM/BAM alignment file to a count table that can be used for differential expression analysis. Along the way, we will learn more about the SAM and BAM files.**
# Viewing BAM files
The first thing we are going to do is look at the alignment file, so we can learn more about the structure of BAM/SAM files. Since we're not downloading the BAM file directly, but instead using a BAM file included in the `RNAseqData.HNRNPC.bam.chr14` package files, we first need to find out the file location of where the BAM file is stored. From the package documentation, we learn that the file names are stored in the object `RNAseqData.HNRNPC.bam.chr14_BAMFILES`.
```{r}
RNAseqData.HNRNPC.bam.chr14_BAMFILES
```
We can see we have 8 BAM files. We'll use sample ERR127306, which we can determine from the [data repository](https://www.ebi.ac.uk/ena/browser/view/PRJEB3048) is a control HeLa sample.
```{r}
bamfile <- RNAseqData.HNRNPC.bam.chr14_BAMFILES[grepl("ERR127306",
RNAseqData.HNRNPC.bam.chr14_BAMFILES)]
```
This file is not human readable, so we are going to convert it into a SAM file, which we can read. The `asSam` function converts BAM files to SAM files. We'll save a SAM file to the current working directory.
```{r}
asSam(bamfile, destination = "hela")
```
We have created the file "hela.sam". We'll take a peek at the top of the header:
```{r}
cat(system("head hela.sam", intern = TRUE), sep = '\n')
```
This contains information on the format and contents of the file. And at the last few lines of the file, which illustrates some reads and their mapping location, along with lots of extra mysterious-looking information:
```{r}
cat(system("tail hela.sam", intern = TRUE), sep = '\n')
```
You can see that each line corresponds to a single read, and each field refers to a different descriptor of the read. For a detailed explanation of what each field means, see [this document](https://samtools.github.io/hts-specs/SAMv1.pdf). Understanding what each field means is not of immediate relevence to the following steps. However, it is always good to have a general understanding of the structure of the files you are working with.
# Sorting and indexing BAM files
In order to be able to do things like extract reads that align to a certain chromosome, or that map to a certain genomic region, we need to have a BAM file index. In this example dataset, one has already been generated for us. But for illustration we'll review how to create one. To do this, we first have to sort the file. We can do this using the following command to sort our BAM file by chromosome and positions of each read:
```{r}
sortBam(bamfile, destination="hela_sorted")
```
This generated a file 'hela_sorted.bam' in our current working directory. Once this file has been sorted, we can generate an index file.
```{r}
indexBam("hela_sorted.bam")
```
*Note: you must always sort before indexing a file*
# Viewing basic information about our BAM file
We don't want to have to look through our entire file line by line. Let's instead use some of the handy parsing functions in `RSamtools` to view some basic summary info about our file (note these functions can use the non-human readable BAM file, which is more efficient than the human-readable SAM file.
First, we create a special reference to the BAM file, which will be passed to the parsing functions (this contains more information than just the path to the BAM file - also points to the location of the BAM index file):
```{r}
# view basic info of bam
bamFile <- BamFile(bamfile)
bamFile
```
We'll pass this special reference to `countBam`, which will by default tell us how many reads and bases are contained in our BAM file.
```{r}
countBam(bamFile)
```
We can see we have `r countBam(bamFile)$records` reads, which amounts to a total of about `r signif(countBam(bamFile)$nucleotides/10E6, 3)` million nucleotides. From this output, can you determine how long are our reads?
Next, we'll use the `quickBamFlagSummary` function to look at the different 'flags' for our reads.
```{r}
quickBamFlagSummary(bamFile)
```
These 'flags' can be used to get detailed counts of certain types of reads. As a quick example, we might want to grab the count of reads that on the minus strand, we could do the following:
```{r}
params <- ScanBamParam(flag = scanBamFlag(isMinusStrand = TRUE))
countBam(bamFile, param = params)
```
There's a ton more possibilities - explore the documentation of the `RSamtools` package (in particular the `scanBamFlag` and `scanBamParam` functions), as well as the [meaning of SAM flags](https://www.samformat.info/sam-format-flag) if you'd like to learn more. These can be really helpful for quality control of our alignment - to make sure we have what we generally expect (e.g. if we performed stranded sequencing, we expect that roughly half of the reads map to each strand). We don't need to dive too deep into the flags if we are satisfied with our alignment and just want to proceed to get our count table, however.
# Generating a count table
Let's take stock of where we are: we have a BAM file, a SAM file and a BAI (BAM index) file. It is hard to do much analysis with what we have, because even though we know which chromosome our reads map to, we don't know which *genes* (or *transcripts*) they map to.
A variety of different tools can be used to generate a count table - i.e a table of how many reads align to each gene. `HTSeq`, `RSEM`, and `featureCounts` are examples of such programs that are command line tools. Each tool works optimally with the output of different aligners, so it is best to read up on how the BAM files you are working with were processed, and to use the feature counting tools.
Today, we are going to use `featureCounts` as implemented in the `RSubread` package, as the BAM files we are using were generated using the Tophat2 aligner, which generates BAM files that are compatible with `featureCounts`. Because we are trying to associate each read with a gene ID, we need an *alignment file* (such as a GTF file) that contains information on the genomic coordinates of each transcript.
This type of alignment file can be downloaded from a database such as [Gencode](https://www.gencodegenes.org/), but conveniently, `RSubread` has some RefSeq annotations built-in, including for mm10, mm9, hg38, and hg19. It is critical that you use the **same genome version as the genome version used to align raw reads to your BAM file**. Our BAM file was generated by aligning reads to hg19, so that's what we'll use. These built-in annotations allow for summarizing counts over genes (the default, and also the option we'll choose) or exons. Note that external annotation files can be used for other genomes and to summarize over transcript/isoform annotations.
Now we're ready to apply the `featureCounts` function that will do the counting of reads over genes. Note that this step can take a while if we have a large BAM file. We specify `isPairedEnd = TRUE` since as we mentioned above our sequencing protocol generated paired end reads. We also specify `strandSpecific = 1` to indicate that our sequencing protocol was stranded.
```{r}
genecounts <- featureCounts(bamfile,
annot.inbuilt = "hg19",
isPairedEnd = TRUE,
strandSpecific = 1)
```
You can change other options, such as the `minMQS`, the minimum mapping quality score to include in the counts. To learn more about different parameters, see the help fule for the `featureCounts` function.
Let's take a peek at the output.
```{r}
str(genecounts)
```
It looks like the counts are stored in the `counts` slot and the gene annotation is stored in the `annotation` slot. There are also a couple more slots for experiment metadata (`targets` has sample name, and `stat` contains some stats about our BAM file).
Let's take a peek at the count table, subsetted for genes on chromosome 14 (since our bam file was subsetted to reads mapping to chr 14).
```{r}
chr14 <- which(grepl("chr14", genecounts$annotation$Chr))
str(chr14)
head(genecounts$counts[chr14,])
sum(genecounts$counts[chr14,] > 0)
sum(genecounts$counts[-chr14,] > 0)
```
Looks like there are `r length(chr14)` genes on chromosome 14, `r sum(genecounts$counts[chr14,] > 0)` of which have non-zero expression. And as we expected, no genes on any other chromosomes have any counts.
Note that the gene IDs are RefSeq IDs. If we'd like to convert them to another ID, such as Hugo symbols or Ensembl IDs, we can use things like the [`biomaRt`](https://bioconductor.org/packages/release/bioc/html/biomaRt.html) package, or annotation packages specific to the organism we are dealing with (e.g. for human: [`org.Hs.eg.db`](https://www.bioconductor.org/packages/release/data/annotation/html/org.Hs.eg.db.html))
Let's look at a density of expression of the chr14 genes (log-transformed).
```{r, fig.align = 'center'}
data.frame(count = genecounts$counts[chr14,]) %>%
ggplot(aes(x = count + 1)) +
geom_density() +
scale_x_log10()
```
We can save the count table to a text file for importing into future R sessions for downstream analysis (such as with [`DESeq2`](https://bioconductor.org/packages/release/bioc/html/DESeq2.html) or [`edgeR`](https://bioconductor.org/packages/release/bioc/html/edgeR.html). We also include a column for gene length, since we might want to use this to obtain quantities like RPKM.
```{r}
write.table(data.frame(genecounts$annotation[,c("GeneID","Length")],
genecounts$counts),
file = "hela_counts.txt", quote = FALSE, sep = "\t", row.names = FALSE)
```
Lastly, we'll cleanup our directory by removing the SAM/BAM/index files we generated earlier, and the text file of counts (comment the last line if you'd like to keep the text file).
```{r, results = "hide", show = FALSE}
file.remove("hela.sam")
file.remove("hela_sorted.bam")
file.remove("hela_sorted.bam.bai")
file.remove("hela_counts.txt")
```
# Deliverable
Now we have a table that summarizes the number of reads that align to each transcript. The is the perfect input to use for methods mentioned above like [`DESeq2`](https://bioconductor.org/packages/release/bioc/html/DESeq2.html) and [`edgeR`](https://bioconductor.org/packages/release/bioc/html/edgeR.html), whose algorithm requires reads be in integer counts, not normalized by library size. For visualization purposes, however, it may be helpful to use reads that are normalized first.
Your exercise is to convert the existing count data into Counts per Million (CPM) and Reads Per Kilobase per Million mapped reads (RPKM) as defined by Bo Li et al. in [this paper](https://academic.oup.com/bioinformatics/article/26/4/493/243395/RNA-Seq-gene-expression-estimation-with-read).
Hint: check out the `cpm` and `rpkm` functions in the `edgeR` package. Format each of your answers as a numeric matrix of values.
```{r}
# your code here
CPM <- "FILL_THIS_IN"
RPKM <- "FILL_THIS_IN"
```
This next code chunk will test whether you have correctly calculated CPM and RPKM in the previous chunk. The answers are hidden as an encrypted hash, but the function will return Test passed if it matches. Note that it assumes your answers are each formatted as a numeric matrix.
```{r, error = TRUE}
test_that("Calculation of CPM:", {
local_edition(2)
expect_known_hash(CPM, "7b444ec637e82962a409bf283704fed5")
})
test_that("Calculation of RPKM:", {
local_edition(2)
expect_known_hash(RPKM, "16547a066c56abe9ee28f02f59042a89")
})
```