Optimized ASIP architecture for compressed BWT-indexed search in bioinformatics applications

Abstract

Compressed indexes are adopted by a vast set of bioinformatics applications that deal with extremely large datasets, mainly due to the inherently high memory requirements of uncompressed alternatives. However, the additional computa- tional overhead that is imposed by the usage of such indexes makes them harder to implement in embedded computational platforms, such as biochips, with strict processing and power restrictions. Furthermore, compressed indexes are often char- acterized by a significant usage of bit-level operations, some of which are not commonly available on General Purpose Processors (GPPs). To circumvent this limitation, an Application-Specific Instruction-set Processor (ASIP) architecture is proposed to accelerate the processing of biological sequences (e.g., alignment, mapping, etc.) using compressed full-text indexes based on the Burrows-Wheeler Transform (BWT). The proposed processor was built over a RISC micro-architecture and extends the Xilinx MicroBlaze ISA with additional bit-level operations, especially tailored for compressed indexes. When used to perform search operations over the considered compressed index, the proposed architecture provides a reduction of the number of required instructions by about one half. Furthermore, when prototyped on a Xilinx Virtex-7 FPGA, the ASIP proved to offer an overall speedup between 3.1x and 4.5x for the execution of a single threaded operation. To ensure a further processing scalability, the proposed ASIP was designed in order to be easily used as the basic processing unit of multi-core systems, especially tuned for the parallel processing of massive datasets of biological reads. Keywords—Application-Specific Instruction-Set Processor, Compressed text indexes, DNA alignment, Heuristic algorithms