System designs to perform bioinformatics sequence alignment

References (1)

1. Speed-up CLUSTALW Speed-up sequences (s) (s) (s) 100 0.859 1 28 828 08 200 188 6 03 672 07 400 500 1 24 1500 00 600 688 2 15 40625 88 800 969 0 16 6391 87 18 iterations. Note that Table 2 contains only hardware-related results. In general, the cited previous work does not include any information about performance results of the systems for practical usage. However, the end user is more interested in the wall-clock time required to get the alignment results than the hardware execution times. Therefore, wall-clock time required for alignments is also included in this work. Table 3 presents overall performances for linear and affine gap penalty alignment systems. Here, the first column is the database size to be scanned while the other columns present the performance of the proposed system and the software-only solution. As seen from Table 3, the proposed system performance increases as the database gets larger. For example, the proposed system scans a 100 MB database in less than 30 s, while the software-only application uses over 90 min to do the same operation. Performance evaluations of the multiple sequence alignment systems Proposed multiple sequence alignment designs are modeled in VHDL and mapped on a Xilinx Virtex II XC2V6000 FPGA platform. There are 185 MSA cells placed on the FPGA. Each cell includes 91 flip flops and 354 LUTs mapped on 188 FPGA slices. According to the synthesis results, the clock frequency can be set to 57 MHz at maximum. Therefore, performance of the multiple sequence alignment system is 10.54 GCUPS. Table 4 presents how much the first step of the CLUSTALW algorithm is accelerated. To achieve this information, 100 to 800 sequences with 185 residues were aligned with FPGA-supported CLUSTALW and pure CLUSTALW. As is seen, the first step is accelerated over 80 times as the number of sequences surpasses 600. However, this acceleration is only for the first step of CLUSTALW, and the overall gain is still obscure. A realistic prediction can be made by using Amdahl’s Law [29]. According to Figure 2, about 80% of overall time is spent on the first step. Because of that, the hardware can achieve a maximum of 5 times of acceleration over the software-only solution. For this reason, the first steps are compared in Table 4, as it was made in the previous work. The proposed design is compared with a similar design that aimed to accelerate CLUSTALW, presented in [17]. Table 4 presents results directly obtained from [17], since they used exact