Mapping benchmarks of the GMS

[kmay]

Hi,

As I promised in another thread here some first controlled benchmarks from the Genomatix Mapping Station ( GMS):

• Random pick of 1Mio reads (32bps) across the human genome (NCBI build 36)

• Generate 1bps insertions and 1bps deletions in 1% of the reads

• Generate 1 bps substitutions in 10% of the reads

Mapping time: 375 seconds

Correctly identified reads: 921.799 (92 %)
Found perfect alignment for read with indel: 2.653 (0.2%)
Ignored because ambiguous: 43.057 (4%)

=====================================
Real life data:

Illumina: ChipSeq STAT1 human
Total: 5.069.935 reads
Length of reads: 36

Time: 1.524 sec (about 5 min per 1.000.000 reads)

Result:
4.150.890 unique best hits (81 %)
3.151.593 with alignment score > 0.9
207.636 of the unique hits with alignment score > 0.9 have at least on insertion or deletion in the alignment
560.920 reads with multiple best hits

====================================
Real life data:

Helicos DGE human tissue

Total: 4.397.276 reads
Length of reads: 5bps to 140bps

Time: 527 sec

Result:
1.662.837 sequences were ignored (too small) (37 %)
2.011.375 unique best hits (45 %)
777.434 with alignment score > 0.9
665.631 of the unique hits with alignment score > 0.9 have at least on insertion or deletion in the alignment
723.064 reads with multiple best hits

[zee]

This looks good.

What Hardware conditions e.g. RAM, CPU clock speed, etc are these benchmarks running on?

[Chipper]

How does it compare to the Eland alignments (e.g. # aligned sequences etc) for STAT1? Was the DGE alignments done with or without splice junctions?

[kmay]

Hi,

no fancy hardware, just an opteron processor, 2GHz
but 64GB of main memory!

->Chipper: I am travelling in the US this week. I´ll try to give you some answers later, or ask a colleague to jump in.

Klaus

[clivey]

Eland was designed to avoid having 64Gb of memory. It should do about 4000 reads per second per CPU on a modern CPU so ~250 seconds.