Puzzling to SOLID data analysis

[Lv Ray]

Can anyone help me about the SOLID 4 sequencing data ? I got them from SRA database. And they seems have been treated. they are mate paired data.
mate1:
@SRR586064.1 ugc_357_358_MatePair_2x50bp_solid0032_20100528_MP_ugc_357_854_13_97/1
T30..01121.12.1032100213131122200031222022101302313
+
!AB!!?:>@<!@B!;AB?AA@@<2<@?@@<?AB>A?9?:;@@>;-=>>7=@
@SRR586064.2 ugc_357_358_MatePair_2x50bp_solid0032_20100528_MP_ugc_357_854_13_137/1
T00..00000.00.0000000002220301333000020000303000000
+
!<B!!97=<A!<@!?>7?@;9+%+2%%-&-%%+6(,0*)3((%<%4.+8.1

mate2:
@SRR586064.1 ugc_357_358_MatePair_2x50bp_solid0032_20100528_MP_ugc_357_854_13_97/2
G10330000122222033220201000000220002000000000000000
+
!@BBABBBB@>?@@(.))35.%-.3((%1+%((82-'.*-*/3*14*'696
@SRR586064.2 ugc_357_358_MatePair_2x50bp_solid0032_20100528_MP_ugc_357_854_13_137/2
G01002233300021321333003300011021333000110330003000
+
!AA>>@A@?3=?:1+6@:?+;A+0+921:-:B78'6?(/5/?5:26=&*

MY questions are:
1)what is the first read in mate1?(some like the first read have the symbol “.” ,what does it mean?) But in mate1 ,there are many reads like mate2.
2) can I convert the SOLID fastq(color space) to necleotide space? Anything lose?
3) If I convert the color space field, how about the QV lines?
4) I am similar to Bowtie2, can I use it to do mapping?
5)last, should I do the QC before conversion or after?

[Brian Bushnell]

1a) The first read is "T30..01121.12.1032100213131122200031222022101302313". The reads are a single letter followed by numbers.
1b) '.' means 'N' (a no-call).
2) Not without mapping. Only mapped reads can be correctly translated to base-space.
3) The QV lines can be sort of interpolated by averaging adjacent values.
4) You need to map with a colorspace-aware aligner. I'm pretty sure BWA allows this.
5) QC must be done before mapping.

[westerman]

Actually Brian is wrong about #2. It is quite trivial to convert from color-space to base-space without mapping. I use to be able to do so by hand -- just for mental exercise -- when we owned a SOLiD. For example your read:

G01002233300021321333003300011021333000110330003000

converts into:

GGTTTCTATAAAAGTAGTATAAATAAAACAAGTATAAAACAATAAAATTTT

[BTW: I see that Wikipedia says "... There is one unambiguous conversion of a base reference sequence into color-space, but there are four possible conversions of a color string into base strings" which is wrong. I should pull out my old conversion table and correct this.]

That said it is quite dangerous to rely on the conversion from color-space to base-space. If there is any uncertainty in the color-space then the probability of going off-track in base-space is high with the consequent corruption of your base-space data. Mapping the (older SOLiD generated) read to a known reference is the way to keep the corruption from occurring.

Color-space can be a very powerful self-correcting format <em>as long as you remain in color space</em>. Do not convert from color-space into base-space until you are completely done with your analysis.

BWA dropped support of color-space as of version 0.6 so you will have to use an older version.

Bowtie2 never supported color-space although the former program Bowtie does.

BTW: Not that this help you but the new SOLiD has a self-correcting method which re-aligns the read properly ever fifth base.

[Brian Bushnell]

Actually Brian is wrong about #2. It is quite trivial to convert from color-space to base-space without mapping.

I wouldn't say "wrong"... though maybe misleading. It is impossible to correctly translate an imperfect colorspace read to base-space without mapping. And since SOLiD has an extremely high error rate - a minority, maybe around 20%, of reads were error-free in the SOLiD 4 data I've processed - in practice, it's not a useful thing to do. And indeed it is completely impossible to translate a read with a no-called base, before mapping.

It's also not possible to unambiguously translate a colorspace read to base-space even AFTER mapping; the translation requires assumptions about the probability of errors versus SNPs versus indels and the results are highly subjective.