blast hit problem

[NicoBxl]

Hi,

I've a problem with a blast alignment (blast2 -p blastn or blastall -p blastn or megablast )

When I align a fasta file with about 3M sequences ( ~ 20 nt each sequence ) with a DB, I don't have any good alignment. But if I take a little sample of the sequences in the fasta file ( ~ 100 sequences ) , I have a few very good alignment ( ID = ~100% and e-value < 1e-5 )

I don't understand how this can occur ?

Anybody knows how to solve this problem ?

Thanks,

Nicolas

[malachig]

That is strange. In the past I have used BLAST to align blocks of 500,000 to 1 million reads to various databases (not the whole genome, but transcriptome or junction databases). In that scenario I used the '-m 8' option to produce tabular output, and filtered it as a stream to remove alignments below a specified score. If you do not do this, the output files can be massive (because BLAST reports all multi-match hits). Is it possible that you are producing a massive file that consumes all your disk space. Does the job actually run to completion? Remember that the output is buffered so it can take a while before anything gets written to disk if you have a large number of reads but only a small number of alignments being found, this effect could be exaggerated.

[Torst]

When I align a fasta file with about 3M sequences ( ~ 20 nt each sequence ) with a DB, I don't have any good alignment. But if I take a little sample of the sequences in the fasta file ( ~ 100 sequences ) , I have a few very good alignment ( ID = ~100% and e-value < 1e-5 )

Older BLAST engines aligned each input sequence one by one, and wrote the result into the output file. Newer BLAST iterate over large chunks of the input to avoid scanning the whole database for each query sequence. Sometimes it can take hours before any output will be written.

Are you letting BLAST finish running first?