[inesdesantiago]

Good to know! Thanks for the reply!
Ines

[kmcarr]

Sorry for this basic question. But what is the tile? Is it the pictures of the lane? So, each little photographed square in a given lane is called a tile (tile 1, tile 2 ,etc etc)?
ines

Yes, that is correct. Tiles do not exist in any physical sense, they are just the sections of each lane as they are imaged by the camera. On the current generation instrument, GAII, there are 100 tiles per lane, made up of two columns of 50 tiles each. The tiles are numbered starting with #1 at the top left of a lane, down to #50 at the bottom left, over to #51 at the bottom right then up to #100 at the top right.

[inesdesantiago]

Sorry for this basic question. But what is the tile? Is it the pictures of the lane? So, each little photographed square in a given lane is called a tile (tile 1, tile 2 ,etc etc)?
ines

[SillyPoint]

Further to the OP's original question:

The _seq.txt file is just as cgb says: lane, tile, X, Y, sequence. X & Y are in pixels relative to the upper left corner of each tile image, with +X to the right, and +Y down (don't ask).

The _sig2.txt also file starts with lane, tile, X, Y. The rest is intensities for each base, each cycle. Intensities have been corrected for crosstalk and phasing. Pay attention here: For each cycle, there are four values (a,c,g,t). They are separated by *blanks*. Cycles (4 values) are in turn separates by *tabs*.

The _prb.txt file contains base probabilities arranged the same way. No lane/tile/x/y here, though. The probabilities are given Solexa-style: Q = 10 * log (P/(1-P)), where P is the probability that the base is a/c/g/t. Not to be confused with phread-style scores, encoded as Q = -10 * log (E), where E is the probability of an *incorrect* call.

Having said all that, I'm moved to enquire: Why are you looking at what are really intermediate data files? The end product of the pipeline for most purposes is the _sequence.txt files produced by the Gerald step. There you will find what amounts to fastq-format files, containing sequence and base scores, plus lane/tile/X/Y. Only beware that the scores are Solexa-style and encoded as ascii by adding 64 (so Q40='h'). maq expects a true fastq file, with phred-style scores plus 33 (Q40='I').

[SillyPoint]

To amplify a bit on cgb's posting: If you align your reads to a known, error-free reference (e.g., PhiX), you can then count the true errors and establish a true error rate. Compare this to the estimated error rate embodied in the Q scores. They should match: Out of all the Q30 bases in all the reads, there should be 1 error in 1000, and so on for each Q value.

An easy place to find this information is in the s_<lane>_qreport.txt file produced by Gerald when you do an alignment on the lane (ANALYSIS default or Eland). What you'll see there is that what's called Q40 really has 0.5% errors = Q23.

[cgb]

the scores are supposed to reflect the chances of a basecall being in error, 20 = 1 in 100 etc. If they do this accurately they are "calibrated". Raw Bustard scores are not well calibrated - it tends to over score and underscore bases and shove a lot into a Q40 bin (wrongly). he scores can be adjusted after the fact using several well known methods - the newer (0.4) / 1.0 release of the GAPipeline allows for some degree of recalibration using control lane data.

[ScottC]

Note - your quality values are raw Qscores emitted by Bustard and will not be well calibrated.

Hi cgb,

Can you expand on this a bit more please?

Cheers,

Scott.

[swbarnes2]

cgb,
There is this _sequence.txt output per lane as well, that is the reads in seq file minus the QC reads that fail chastity filter. This can then be converted to fastq using one of the MAQ utilities.

On our pipeline, the _sequence.txt file only has 32 bases of sequence. If you are using SOAP or Maq, or you are doing more than 36 bases, you don't want to lose all those bases. Maybe you can fool around with the pipeline to get it to output more, but I don't know how. It also uses a non-standard quailty scoring format, but that's not a deal-breaker.

I made a <50 line perl thingie to take the .prb and .seq files to make a fastq. If I can do it, it can't be that hard

[cgb]

have a look on the sanger site - if not mail jkb@sanger.ac.uk or ts6@sanger.ac.uk

[bioinfosm]

the sig2 files are processed "traces" you can draw a bar chart with them for each sequence. The seq files are the final data - its trivial to convert the seq and prb files into a fastq file - there are tools floating around to do this.

generally the key is the first 4 columns : lane, tiles, x, y for the given cluster that gave the sequence.

cgb,
can you say more on these programs that convert prb + seq into fastq format?
There is this _sequence.txt output per lane as well, that is the reads in seq file minus the QC reads that fail chastity filter. This can then be converted to fastq using one of the MAQ utilities.

Any advantage of using seq + prb, instead of the filtered _sequence? I have heard from MAQ, SSAHA and other authors that using the filtered file is preferred to get better alignment results using their tools

sm

[cgb]

... on the sig2 files - your row (= cluster) has the same key for the first 4 cols. then you have 4 values for A,C,G,T <Tab> A,C,T,G etc.... up to cycle number

note - your quality values are raw Qscores emitted by Bustard and will not be wel calibrated.

[cgb]

Not quite....

the flowcell has 8 lanes. lane number is the lane. each lane has up to 330 'tiles' they are numbered in a snakey pattern, the X,Y is the cluster co-ordinate on the given tile

[rebrendi]

Lane = 1-8 (which channel of the flowcell)

X,Y = physical location of the cluster on the flowcell...

thanks....

[ECO]

Lane = 1-8 (which channel of the flowcell)

X,Y = physical location of the cluster on the flowcell...

[rebrendi]

generally the key is the first 4 columns : lane, tiles, x, y for the given cluster that gave the sequence.

well, so what do they mean these columns?