MG-RAST ambigious base count

[sfranzenburg]

Hi,
I wanted to analyze my 150bp Paired-End Metagenome Reads (Illumina) with MG-RAST.

After uploading, I get the file statistics.

My forward sequences (R1) have <5% ambiguous characters, but my reverse sequences (R2) have ~75% ambiguous characters (according to MG-RAST). Needless to say, that I can not apply MG-RASTs default quality filtering to these reads, they all would get removed.

However, when I check my reads with FastQC, the R2 file seems fine:

So why does MG-RAST tell me, that these reads are of low quality, and why is that no problem with the R1 file?

A few lines from the R1 file:
@HWI-ST1348:258:H8UMCADXX:1:1101:1062:2034 1:Y:0:ATCANG
NATGNTCCTTGATAAGAGNAAAGATAAGTGGCAGATTGTCCGGAAAGCGTATAAATAATATAGTAATGCTAAAAGGTNNCNNCGATGCCGGTATTTATAGTACTATGCAAATTCTCAGCCAGGTTACCATTAATGAAANCATGATAGAAAT
+
#0;@#2@@@@@@?@@@?@#3@>@??@??>@??>??????????????????????????>??@?????==?????############################################################################
@HWI-ST1348:258:H8UMCADXX:1:1101:1237:2154 1:N:0:ATCACG
CAATACGCAGCAACAATCACATACAGAGCGTACATACAGTATGCTCCGACATGTTAGAACTTGATTATTTTGTGCCATAGTTGCGTACCTAGCGAAGCGAGTTTCGGTGCTTATGCTTTCACGTTTCTATAAAAACTGCGTGACAACGCCA
+
CCCFFFFFHAHHHJJJJJIJJIJJIJJGGIGHHIIGIIIIIIJJDHIJGIIIJJJIJJJHHHHEHHFFFFFFAACEEECCDDCCB@?@?CCC@<B><>@BBBBC>B>?@9CCCC@>ADACAC<BD?CCDADDED:5>4499928A485>B9
@HWI-ST1348:258:H8UMCADXX:1:1101:1370:2035 1:Y:0:ATCACG
NTGGNACAGTTTAATGACGCGTTTATATGTCGGCGCGGGCAGTTCGTGCTGTGTATGAGTGTGAGGCATTTCCCACGAATCTGGCCGACGCGGGCAGTTCGTTGTATGAGTGTGAGGCATTTCGCAGGAATGTTATTTCATCATTATTTGC
+
#0;@#2=@@@@@@?@?@????@@????????????8??#################################################################################################################
@HWI-ST1348:258:H8UMCADXX:1:1101:1382:2089 1:N:0:ATCACG
ATTCTAGAGAAACAAATCCTTAATTTTTCTCGCCAGATTTCGCGCTCCTAAAGATCATAGAAGGAGCAAAAAAAATTATCTATATTCTTCAATGACTTATTGGGGGGGGGGGCGACCCCAGCTATAGATCGGAAGACCACACGTCTGAACC


A few lines from the R2 file:
@HWI-ST1348:258:H8UMCADXX:1:1101:1237:2154 2:N:0:ATCACG
CTTTAAGAAGTACCGTTTTAGCGAGACGACAGCGCTACAACGCTGTCGTCGTGTGTGGTTCGGAGCTATGCGGGGAAGTAAGTCTGGCGTTGTCACGCAGTTTTTATAGAAACGTGAAAGCATAAGCACCGAAACTCGCTTCGCTAGGTAC
+
CCCFFFFFHHFHHJJJJJJJJJJJJJJJJJJJGIIJJJJIIGGIEHEHFDF<>;?=ABCBDDD?BDDDCDCBDDD99@:CCDCDDDCC@DBD@DECBDBDBCDDDDDDDDDDDD<ADDDDCCCDDDDACDDDDBBCCBBDBDD@BBBB>>:
@HWI-ST1348:258:H8UMCADXX:1:1101:1370:2035 2:Y:0:ATCACG
GGCGACAGCGTTGTGGCGCTGTCGCCGCGCTAAAACGGTACCTTCCAAAACTACCAAATGCCATCAGCCAATCGCAACTCAATTAAAGGCTGATGCAATTGCCAAATAATGATGAAATAACATTCCTGCGAAATGCCTCACACTCATACAA
+
@CCFFFFFHHHHHIJJJJJIIGIIJJGGGIJJHHHHFFBEDDEDEDDDDDDDCDDDDBD:CDCCDDACDDDDD@BD@DDCDDDEDDDDCDCBBCCCDCCDDCCCDDDDDEDDDDDDD>CC@ACCDDDCDDB@>AACC:>@<:@9>?AA:@3
@HWI-ST1348:258:H8UMCADXX:1:1101:1382:2089 2:N:0:ATCACG
ATAGCTGGTGTCGCCCCCCCCCCCAATAAGTCATTGGAGAATATAGATAATTTTTTTTGCCCCTTCTAGGATCTTTAGGAGCGCGAAATCTGGGGAGAAAATTTAAGGATTTGTTTCTCTAGATTAGATCGGAAGAGCGTCGTGTAGGGAA
+
@@<D;DDDHHH?F=C@HGHGIIF>>B6;>C#########################################################################################################################
@HWI-ST1348:258:H8UMCADXX:1:1101:1478:2092 2:Y:0:ATCACG
GCATTAGACCTTTGCAAACCAGTTTTCCCCCCCATTGCGCGACATTACTCATACGCCAAGTGGAACATTAGCACTTCTCTCCGATGAATAGCTCAGTTCTTCGCCAAATTATGTGAATTTGAAACGTGTGAGCGGGCAAAGAGGTTCATTC


best regards,
Soeren

[GenoMax]

Your sequence files should have no ambiguous characters (just A/C/G/T). It sounds like your file may be corrupted in some way or there is the usual problem with PC/Mac/Unix file format differences to consider (if these files were moved among any of those systems).

[sfranzenburg]

Just got a reply from the MG-RAST support (didn't expect them to react that fast, thats why I got here for help too). Seems like I have misinterpreted their numbers.
When I saw average ambig chars: 0.75 and saw that 75% of my reads fail fastq-trimming, I assumed that 75% of my bases must be interpreted as ambiguous. Thats not the case, but however, 75% of my reads (which according to FASTQC have good quality, fail default QC (min score 15, max 5 bases). But I think MG-RAST support will help me from now on.

[Brian Bushnell]

With 150bp reads (I assume this is Hi-seq 2000?) I don't recommend failing reads just because some of the last few bases are below Q15, because you'll throw away tons of good data. Depending on what you want to do with the reads (e.g., map them), you don't necessarily even need to quality-trim them.

[sfranzenburg]

Thanks.
I meanwhile found my lost reads. Most of them were removed in MG-RASTs dereplication step (50% removed). Is dereplication state of the art for Illumina data today? 50% loss sound very high for me.

[Brian Bushnell]

It depends on what you're doing and how the data was generated. If you don't do any amplification, then duplicate removal may not be necessary, which is ideal. The shorter your insert size, the smaller the genome, the more PCR cycles are performed, the less random your shearing protocol is, and the more reads you have, the more duplicates will be detected.

50% is very high for what I consider a "normal" library, but that could be very different from your "normal". Looks like you're studying drosophila... can you say anything about the library prep and the purpose of the study?

[sfranzenburg]

Hi Brian,
thanks alot.
I did MDA for 8h to amplify my original DNA (library prep protocol wanted to have 1 µg, which was far more than I had). Library Prep was basically the same as with the Illumina Kit. Insert size is 200bp, shearing was performed by sonification (Covaris) and 11 PCR cycles were used in the final PCR.

[Brian Bushnell]

Since you started with less DNA than required, and performed a lot of amplification, 50% duplicates is not unexpected. My only experience with MDA is in single-cell sequencing, but our amplified single cells often have 95%+ duplicate rates and highly erratic coverage ranging from 1x to 100000x for the same library.

[sfranzenburg]

Ok, thanks alot Brian.
Than I will the discard the reads.
I read some papers for MDA before I did it and I don't recall that they brought up this problem.

[Brian Bushnell]

It might not be due to MDA, just large amounts of amplification in general will create duplicates.