RASSA: Resistive Prealignment Accelerator for Approximate DNA Long Read Mapping

Abstract

DNA read mapping is a computationally expensive bioinformatics task, required for genome assembly and consensus polishing. It requires to find the best-fitting location for each DNA read on a long reference sequence. A novel resistive approximate similarity search accelerator, RASSA, exploits charge distribution and parallel in-memory processing to reflect a mismatch count between DNA sequences. RASSA implementation of DNA long read pre-alignment outperforms the state-of-art solution, minimap2, by 16-77× with comparable accuracy and provides two orders of magnitude higher throughput than GateKeeper, a short-read pre-alignment hardware architecture implemented in FPGA.Constructing human DNA sequence in real time is paramount to development of precision medicine 1 and on-site pathogen detection of disease outbreaks 2 . Single-molecule, real-time sequencing from Pacific Biosciences 3 (PacBio) and Oxford Nanopore Technologies 4 (ONT) are new technologies that can produce long reads within minutes, potentially enabling real time genomic analysis. However, long read DNA sequencing poses new challenges. First, long reads contain many thousands of base pairs (bps). Second, long reads tend to exhibit about 15-20% insertion, deletion (indel) and substitution errors 3,4 .To construct a complete host sequence, in case a reference sequence exists (from a previously sequenced organism), long reads are mapped to high-similarity locations of the reference sequence. Determining the edit distance between every mapped read and the reference sequence requires a computationally intensive local alignment procedure (e.g., Smith-Waterman 4 ). Its computational time complexity is typically ( ) for two sequences with lengths and . Reference sequences vary from several millions to billions of bps. It is therefore computationally prohibitive to perform optimal alignment of every long read with the entire reference sequence.Read mappers (e.g., minimap 6 , minimap2 7 ) find regions of high similarity (mappings) between reads or Ran GinosarTechnion, Israel Institute of Technology IEEE MICRO between a read and a reference sequence, followed by an alignment step to determine the exact edit distance and verify that the mapping is correct. In case that a pre-alignment algorithm identifies a specific region in the reference suitable for mapping, the alignment can be performed only on that region, reducing alignment's duration and resource requirements 8 . Therefore, read mapping can be viewed as a twostep process: (1) pre-alignment filtering and (2) accurate alignment verification. The pre-alignment step reduces the problem size for aligners by narrowing the regions to ones with potentially high-scoring alignment.Existing pre-alignment hardware solutions 9,10 target short reads (up to several hundred bps) which contain a small number of indel and substitution errors (less than 5%) and have a different error profile than that of PacBio or ONT long reads 3,4 . High edit distance threshold is required for mapping long but error-prone reads. However, c...

RRAM-Based CAM Combined With Time-Domain Circuits for Hyperdimensional Computing

“…Area and energy savings were improved at the expense of increased latency by less than 20% due to digital-to-analog conversions. Another type of AM is RASSA with a 2T1R bitcells structure and depends on discharging the ML which consumes a lot of power 4 . The outputs of RASSA are locations on the reference input sequence, where alignment may result in a high score.…”

“…Content addressable memory (CAM) is an attractive hardware solution for applications that significantly rely on high-speed search, match, and retrieve operations [1][2][3][4] . Unlike conventional SRAM, that takes several cycles for a search operation, a CAM directly performs the search within its pre-stored content in a parallel fashion with potential single cycle access, naturally realizing in-memory computing (IMC).…”

Scaling Genomics Data Processing with Memory-Driven Computing to Accelerate Computational Biology

“…Alser et al [5] propose another FPGA based tool for pre-alignment; they achieve a 10× acceleration for this task using a Virtex-7 FPGA using Xilinx VC709 board running at 250 MHz. Kaplan et al [18] propose a novel resistive approximate similarity search accelerator (RASSA) that exploits charge distribution and parallel in-memory processing to reflect a mismatch count between DNA sequences. Their pre-alignment software achieves 16-77× improvements in long reads.…”

Hardware acceleration of genomics data analysis: challenges and opportunities

“…Thus, a search usually extends the full reference genome for each read resulting in billions of searches, making it computationally intensive ( McVicar et al , 2016 ). Previous attempts to increase read alignment throughput have included multistage alignment algorithms ( McVicar et al , 2016 ), pre-alignment filters ( Kaplan et al , 2019 ) and many-core processing ( Liu et al , 2017 ).…”