FASTQ alignment metrics (RNA-Seq)?

[dan]

Hello,

How do people judge the quality of a FASTQ (short read) alignment? In particular I'm interested in evaluating RNA-Seq alignments, typically (but not exclusively) from ILLUMINA instruments.

What comes to mind is:
* Fraction of reads mapped
* Fraction of reads mapped uniquely
* Fraction of 'good' pairs (right orientation, right distance)

and for RNA-Seq specifically
* Fraction of reads mapping within a gene

Anything based on read mapping quality?

What other metrics can we think of?

[annaprotasio]

hi Dan,

Have a look at "samtools flagstat"

The output will looks something like this and I think it contains all the info you requested.

7276199 + 0 in total (QC-passed reads + QC-failed reads)
0 + 0 duplicates
7276199 + 0 mapped (100.00%:-nan%)
7276199 + 0 paired in sequencing
3787000 + 0 read1
3489199 + 0 read2
6195536 + 0 properly paired (85.15%:-nan%)
6795026 + 0 with itself and mate mapped
481173 + 0 singletons (6.61%:-nan%)
480036 + 0 with mate mapped to a different chr
480036 + 0 with mate mapped to a different chr (mapQ>=5)

good luck

[GenoMax]

Also take a look at RSeQC: http://rseqc.sourceforge.net/

Most aligners will produce stats on alignments e.g. BBMap, TopHat and probably STAR as well.

[maxsalm]

FastQC may also be of general use: http://www.bioinformatics.babraham.a...ojects/fastqc/

[dan]

FastQC may also be of general use: http://www.bioinformatics.babraham.a...ojects/fastqc/

I agree it's useful, but it's not what I want here.

[jwfoley]

How about proportion of duplicate fragments? This will depend on whether you've done single- or paired-end reads, though, since with single RNA-seq reads you do expect a certain amount of duplication by chance (with paired reads it's a much smaller chance).

[bjackson]

I do primarily single ended reads, but for alignment quality I look primarily at
1) pct of reads mapped
2) pct of reads uniquely mapped

It sounds like you are also asking about post-alignment qc in general and I add
3) read duplication (ie how many reads align to identical location) - most reads should have only one or several.
4) reads biotype distribution (most should map to protein-coding regions)
5) cumulative pct measures - I sort genes by count or fpkm and graph # of genes vs cumulative percentage. That will tell you if you are sinking a lot of reads into very common transcripts and tell you that you might need more depth to see certain less common transcripts.