Duplication level of RNA-seq data

[gary]

Hello everyone!

I got my 100bp paired-end RNA-seq data today, FastQC told me that the duplication rate is above 60%. I searched around and found that it is common to get a high dup level with RNA-seq. Is that normal?

Should I remove the duplication? I heard some discussions in this forum that if the duplicates were removed then I cannot compare the highly expressed genes since the max depth of coverage at one point is 200 with 100bp sequencing data.

thanks!

[Chipper]

200 is the maximum coverage per base if you have unstranded single end reads, with paired ends you can have many fragments starting at the same point if the other end differs.

FastQC only looks at the reads, what you should do is calculate the library complexity after alignment (e.g. with Picard). Looking at the alignments at lowly expressed genes also helps to determine if the library is over-sequenced.

[gary]

Thank you Chipper! I will try your suggestions.

In your experiences, do you think the >=60% duplication level given by FastQC is too high? Or I will need to look at the alignment results to see if that was too high.

[NextGenSeq]

What was the yield? We usually aim for less than 100 ng/ul but some RNA samples amplify better than others.

[arkal]

In my experience (from what i've seen and what i've read), you can expect to see 60-90% duplication when u run a fastqc on ur data! Don't really know how it correlates to the real picture though!

[TonyBrooks]

FastQC bases it's duplication estimates on the first 50bp of sequence only. It also makes no allowance for paired end data. High FastQC duplication rates for RNA-Seq is normal.
To get a better idea you need to look at the mapping co-ordinates.
As Chipper says, use the Picard library complexity estimator. For example, I just ran and RNA-Seq sample with 64.59% (Read1) and 58.13% (Read 2) FastQC duplication, but only 0.32% duplication using the Picard library complexity estimator.

[Baoqing]

Interpretation of the PICARD results

Hi, Guys
I was also trying to estimate library complexity with PICARD with my paired end data, I used the tophat aligned reads as input, according to the picard, "One or more files to combine and estimate library complexity from." what does it exactly mean ? Does multiple inputs mean the bam files from each duplicate of biological samples?

If so, how to add? Should I just add extra files in the INPUT argument

java -Xmx2g -jar ~/Desktop/apps/picard-tools-1.92/EstimateLibraryComplexity.jar INPUT= accepted_hits025.bam <more bam files here?> OUTPUT= picard_file MIN_IDENTICAL_bases=6 MAX_DIFF_RATE=0.02

I also get a table from the run of the command, but any clue how to understand this file? I also attached part of the table (It is too long to attach)

Thank you in advance!
## METRICS CLASS net.sf.picard.sam.DuplicationMetrics
LIBRARY UNPAIRED_READS_EXAMINED READ_PAIRS_EXAMINED UNMAPPED_READS UNPAIRED_READ_DUPLICATES READ_PAIR_DUPLICATES READ_PAIR_OPTICAL_DUPLICATES PERCENT_DUPLICATION ESTIMATED_LIBRARY_SIZE
Unknown 0 6501528 0 0 2054652 1644248 0.316026 27100851

## HISTOGRAM java.lang.Integer
duplication_group_count Unknown
1 3808007
2 376471
3 78976
4 57845
5 19068
6 26550
7 8180
8 17601
9 4826
10 10475
11 3229
12 5810

[Chipper]

Picard estimates that your library has 27100851 molecules.

[Baoqing]

Thanks. Still not quite clear to me. Does that mean 27,100,851 mRNA, exon or something else? seems unlikely are mRNAs, the read pair examined in total is 6,501,528. Also, it seems that the duplicate_group_count are some kind of id, how could i get back and check those duplicated reads? Really appreciate your help in clarifying this.