Smith-Waterman Performance

When comparing protein sequences, Smith-Waterman is often the most optimal algorithm, especially when % identity is low. For even more distant evolutionary relationships, double affine (spans introns) and frameshift-tolerant Smith-Waterman methods can be employed.

TimeLogic's accelerated Smith-Waterman can achieve 10.125 billion cells/second per DeCypher Engine. Throughput scales linearly with the addition of cards to a server. In comparison, Software Smith-Waterman achieves only 20 million cells/second on a single CPU, and will not scale linearly with additional processors.

Amino Acid Comparisons

IPI mouse sequences (42,799) were compared to a database of 49,078 IPI Human on a DeCypher system comprised of a dual 2.8GHz CPU server and two DeCypher Engine™ accelerator cards. With hardware-based Smith-Waterman, the top 100 descriptions and top 50 alignments were calculated within 7 hours and 13 minutes.

A subset (1%) of the same Smith-Waterman search was run on a dual 2.8 GHz CPU server using ssearch, which does not benefit from the TimeLogic hardware. From this subset, we expect that the top 100 descriptions and top 50 alignments would require 947 hours to complete.

The DeCypher system represents a 131-fold speed up over software execution of this Smith- Waterman search. In this direct comparison, the two DeCypher Engine system demonstrated the speed of 262 CPUs, minimally. *We expect that the software search could not be scaled linearly across a server farm and would require significantly more CPUs to match the performance of DeCypher. DeCypher, on the other hand, scales linearly up to 8 cards per server. Each additional server will double the throughput.

Smith-Waterman Sensitivity at BLASTspeeds for oligo searches?

Comparing probes to a genomic database is important for primer and probe design, SNP mapping and many other informatics techniques. We configured a system with 8 DeCypher Engine™ accelerator cards in a dual-CPU HP DL585 server running Red Hat Enterprise 4, and compared 1,000 30-mer sequences to Unigene (3.29 million mouse sequences) using Smith-Waterman in 2 hours, 41 minutes. 10,000 alignments were reported.

TimeLogic achieve tremendously improved performance using Tera-Probe™ application for the same search, which is a fusion of Tera-BLAST and Smith-Waterman. Using the same system configuration, we are able to complete this search in 3 minutes and achieve the same sensitivity.

For immediate information, please contact a TimeLogic account executive today to discuss a custom benchmark project.