TimeLogic BLAST Performance

The following examples are real-world application benchmarks recently performed by TimeLogic using several DeCypher configurations.

Tthe TimeLogic CodeQuest workstation will deliver the performance of a 1-card DeCypher Engine installed in a 2-CPU system, so many of these benchmarks are indicative of the workstation's performance.

Translated BLAST search

Tera-BLASTX was used to compare 976,256 genome sequence tags (682 bp average length) with NCBI NT (18 GB). The search is processed on a DeCypher system consisting of 2 dual-CPU servers with a total of 4 DeCypher G4 engines. The search required 7 days, 20 minutes to report the top 10 hits and alignments with e-values at or above 10-5. We expect this benchmark would take 2-3 weeks on a CodeQuest Workstation, depending upon configuration.

Comparing 70-mers to the Rat Genome

Using BLASTN to compare 26K oligos to the Rat genome. A 6 CPU cluster required 7 days to complete this search, while a 2-CPU server with a single DeCypher card ran the search in 20 minutes. The DeCypher system delivered performance equivalent of thousands of CPUs. The CodeQuest workstation would deliver equivalent performance.

All Known Bacterial Proteins vs. 192 Translated Bacterial Genomes

Algorithm: Tera-T-BLASTN

Using the smallest configuration BLAST System installed within a Sun V880 server, all known bacterial proteins (4,289 proteins sequences totaling 1,358,990 symbols from NCBI’s E. coliK12 U00096.faa dataset) were searched against 192 bacterial genomes (775,191,168 symbols in 6-frames) in 1 hour, 36 minutes.

The same 8-CPU server without DeCypher acceleration completes the task in 13 days, 9 hours and 40 minutes (321.66 hrs).

Results: DeCypher improves performance by 200-fold, operating at a speed equivalent to 1,600 Sun Ultra Sparc III CPUs.

Human Chromosome 19 vs. Mouse genome

DeCypher was used to compare Human chromosome 19 to the complete mouse genome with BLASTN and returned the top 500 scores and top 250 alignments (with e-values less than 10) in only 2.3 hours.

Details: This benchmark consisted of a Tera-BLASTN search of 1 nucleic sequence containing 15,825,424 bases against the complete mouse genome in the file chromFaMasked.zip containing 37 sequences totaling approximately 2,711,077,328 bases. This was processed on a BLAST System with 2 DeCypher accelerator cards installed in a Sun Microsystems V880 (8 CPUs).

Results: The search completed in 2 Hours, 20 Minutes.

Large BLASTX Search

Algorithm: Tera-BLASTX using nucleic query vs AA target

System: BLAST System (dual DeCypher accelerator cards)
hosted in a Dell server (2 CPUs, 4 GB RAM)

Query: 500K nucleic sequences randomly selected from recent release of NCBI's gbest dataset

Target: 1 million AA sequences randomly selected from a recent release of NCBI's genpept dataset

Details: We searched the forward 3-frame translations and the reverse 3-frame translations of the query sequence against the direct strand of the target sequence and returned gapped alignments.

Results: DeCypher required 27 hrs to complete this search.

See suggested host servers for various DeCypher configurations.

Interested in further discussion of our benchmark data? Would you like to see custom benchmarks for your toughest bioinformatics bottlenecks? Contact a TimeLogic account executive today to learn more about DeCypher!