unmapped SOLiD reads actually mapped via Bowtie

[hoagiang]

Hello everyone,

We have troubles with mapping color-space SOLiD reads using bowtie. So I'm looking for your help to solve this weird issue.
We have yeast mRNA reads from SOLiD 4. We mapped these reads with bowtie using index from the mRNA sequences. About 30% of the reads mapped to these mRNAs. We checked with ABI rep, they said that 30% is normal for these libraries. So we went on with our analysis.
Now that we have some spare time. We want to know what the unmapped reads are, so we first mapped them to the yeast genome.
Surprisingly, ~50% of the unmapped reads mapped to the genome (presumably the non-mRNA regions). When we looked at the location of these genome-mapped reads, 10% are located in mRNA region in the yeast genome.
We went back and forth to check and it seems these genome-mapped are real and actually mapping to mRNAs.

Our question is how bowtie seems to miss these genome-mapped reads when mapping to the mRNAs. We also wonder how many of the remaining unmapped reads are actually unmapped, since it's not a viable option to convert from color-space to base-space and then mapping them.

We appreciate any advice on why this happens.

Thanks,

Hoa

[hoagiang]

some examples

Here are two examples we picked out:

test_bowtie.csfasta:
>1279_48_327_F3
T30202210310311223303333213232132131
>1279_109_287_F3
T21011322013012332221022113232132131

test_bowtie.qual:
>1279_48_327_F3
33 33 30 31 32 31 31 31 33 27 22 25 27 23 27 32 26 30 22 25 18 5 31 18 4 8 6 31 23 10 19 15 17 14 22
>1279_109_287_F3
30 29 33 30 25 21 31 33 33 31 12 26 32 29 5 5 25 31 31 23 4 31 29 16 17 9 26 24 21 26 6 30 24 15 25

the bowtie commands we used:

test_bowtie_genome.sam:@PG ID:Bowtie VN:0.12.7 CL:"bowtie -t -f -C -v 2 --trim3 12 -a -m 1 --best --sam -p 4 -Q test_bowtie.qual /home/hoagiang/bin/bowtie/0.12.7/indexes/S288C_cs/S288C_cs test_bowtie.csfasta test_bowtie_genome.sam --un test_bowtie_genome_unmapped"

test_bowtie_refseq.sam:@PG ID:Bowtie VN:0.12.7 CL:"bowtie -t -f -C -v 2 --trim3 12 -a -m 1 --best --sam -p 4 -Q test_bowtie.qual /home/hoagiang/bin/bowtie/0.12.7/indexes/S288C_SGD_R64_transcriptome_cs/S288C_mRNA_cs test_bowtie.csfasta test_bowtie_refseq.sam --un test_bowtie_refseq_unmapped"

The results we got from mapping to the genome:
test_bowtie_genome.sam:1279_48_327_F3 0 chrXI 163830 255 21M * 0 AGGAGTTACCGTGAGCGGCGA `^``__a]RPUSS\[YUPL.; XA:i:1 MD:Z:21 NM:i:0 CM:i:1
test_bowtie_genome.sam:1279_109_287_F3 16 chrXII 283214 255 21M * 0 CTTGAGAATCAACAAGATGTT ]D<W_Y5!C^[/4acaUOX`_ XA:i:2 MD:Z:21 NM:i:0 CM:i:2

The results we got from mapping to the mRNAs:
test_bowtie_refseq.sam:1279_48_327_F3 4 * 0 0 * * 0 AGAGGCATCATCCGGTTATTTT B?@A@@@B<7:<8<A;?7:3&@ XM:i:1
test_bowtie_refseq.sam:1279_109_287_F3 4 * 0 0 * * 0 CACCTGGACTACGTTGGGCAGG >B?:6@BB@-;A>&&:@@8%@> XM:i:1

whereas the mRNAs are:
>YKL152C GPM1 SGDID:S000001635, Chr XI from 164385-163642, Genome Release 64-1-1, reverse complement, Verified ORF, "Tetrameric phosphoglycerate mutase, mediates the conversion of 3-phosphoglycerate to 2-phosphoglycerate during glycolysis and the reverse reaction during gluconeogenesis"
>YLR075W RPL10 SGDID:S000004065, Chr XII from 282927-283592, Genome Release 64-1-1, Verified ORF, "Protein component of the large (60S) ribosomal subunit, responsible for joining the 40S and 60S subunits; regulates translation initiation; has similarity to rat L10 ribosomal protein and to members of the QM gene family"

The two indexes are built from the SGD website: R64-1-1 version of the S288C genome and orf_genomic.fasta

We did notice the differences in sequences when mapping to either the genome or mRNAs, but couldn't explain it.

[gringer]

We did notice the differences in sequences when mapping to either the genome or mRNAs, but couldn't explain it.

I can answer you on this, at least. The unmapped sequences in the SAM output for SOLiD reads are double-encoded colour space (i.e. A->0,C->1,G->2,T-3). When a sequence is mapped, the SAM file instead has the base-space sequence. You're not necessarily going to get an exact match of base space to its colour space equivalent due to errors in the colour space sequence.