A hybrid short read mapping accelerator.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Short read mapping faces challenges in sensitivity and speed due to large datasets.
• Existing methods often sacrifice accuracy for speed or are too slow for large-scale use.
• Hardware-based designs, particularly using Field Programmable Gate Arrays (FPGAs), are underutilized for this problem.

Purpose of the Study:

• To develop a hybrid system combining FPGA hardware and CPU software for efficient short read mapping.
• To accelerate the computationally intensive alignment and seed generation stages of the mapping process.
• To compare the performance of the hybrid system against established CPU-based software (GASSST, BWA).

Main Methods:

• Developed a hybrid system integrating FPGA and CPU components.
• Implemented computation-intensive tasks (alignment, seed generation) on an FPGA.
• Designed a parallel block-wise alignment structure (Align Core) as an approximation of dynamic programming.
• Compared performance against multi-threaded GASSST and BWA software.

Main Results:

• The hybrid system demonstrates faster processing speeds than GASSST and BWA for single-end alignment, with comparable or higher sensitivity.
• For pair-end alignment, the hybrid system offers higher processing speed than BWA, with slightly lower sensitivity.
• The system achieves high sensitivity and reduced execution time with modest FPGA resource utilization.

Conclusions:

• The hybrid FPGA-CPU system effectively accelerates short read mapping.
• The design shifts the performance bottleneck from alignment to seed generation, guiding future aligner development.
• This approach offers a viable solution for handling the increasing volume of short read sequencing data.