[Paper Review] GRIM-filter: fast seed filtering in read mapping using emerging memory technologies

Jeremie S. Kim, Damla Senol+8 more|arXiv (Cornell University)|Aug 14, 2017

Genomics and Phylogenetic Studies27 references17 citations

TL;DR

GRIM-Filter is a novel seed filtering algorithm that leverages 3D-stacked memory's processing-in-memory (PIM) capabilities to accelerate DNA read mapping. By using coarse-grained genome segment representations and massively parallel in-memory bit vector operations, it reduces false negatives by 5.59x–6.41x and achieves 1.81x–3.65x end-to-end speedups over state-of-the-art mappers like mrFAST with FastHASH at a 5% error threshold.

ABSTRACT

Motivation: Seed filtering is critical in DNA read mapping, a process where billions of DNA fragments (reads) sampled from a donor are mapped onto a reference genome to identify genomic variants of the donor. Read mappers 1) quickly generate possible mapping locations (i.e., seeds) for each read, 2) extract reference sequences at each of the mapping locations, and then 3) check similarity between each read and its associated reference sequences with a computationally expensive dynamic programming algorithm (alignment) to determine the origin of the read. Location filters come into play before alignment, discarding seed locations that alignment would have deemed a poor match. The ideal location filter would discard all poor matching locations prior to alignment such that there is no wasted computation on poor alignments. Results: We propose a novel filtering algorithm, GRIM-Filter, optimized to exploit emerging 3D-stacked memory systems that integrate computation within a stacked logic layer, enabling processing-in-memory (PIM). GRIM-Filter quickly filters locations by 1) introducing a new representation of coarse-grained segments of the reference genome and 2) using massively-parallel in-memory operations to identify read presence within each coarse-grained segment. Our evaluations show that for 5% error acceptance rates, GRIM-Filter eliminates 5.59x-6.41x more false negatives and exhibits end-to-end speedups of 1.81x-3.65x compared to mappers employing the best previous filtering algorithm.

Motivation & Objective

• To address the growing memory bandwidth bottleneck in read mapping, which has shifted from alignment to filtering due to the adoption of seed filters.
• To design a filtering algorithm that efficiently exploits the high internal bandwidth and in-memory computation of emerging 3D-stacked DRAM technologies.
• To minimize wasted alignment computation by rapidly discarding non-viable mapping locations before expensive dynamic programming steps.
• To develop a universal, high-performance filter compatible with any seed-and-extend read mapper, particularly hash-table-based mappers.
• To demonstrate that algorithmic innovation tailored to emerging memory architectures can significantly accelerate bioinformatics workloads.

Proposed method

• GRIM-Filter represents the reference genome as coarse-grained segments, each associated with a precomputed bit vector indicating the presence of small read segments.
• It uses massively parallel in-memory operations on 3D-stacked DRAM to count how many of a read’s segments are present in each genome segment.
• If the segment count falls below a threshold, the corresponding mapping location is discarded as a poor match, avoiding subsequent alignment.
• The algorithm is optimized for processing-in-memory (PIM) workloads, minimizing data movement and leveraging the high internal bandwidth of 3D-stacked memory.
• The bit vector representation enables efficient, parallel comparison operations directly within the memory layer, reducing computational latency.
• The method is orthogonal to existing filters and aligners, allowing it to be stacked with other optimizations for further performance gains.

Experimental results

Research questions

• RQ1 Can a seed filtering algorithm be designed to fully exploit the high bandwidth and in-memory computation capabilities of 3D-stacked DRAM?
• RQ2 To what extent can such an algorithm reduce false negatives compared to existing filters while maintaining high filtering speed?
• RQ3 How much performance gain can be achieved in end-to-end read mapping by replacing traditional filtering with a PIM-optimized algorithm?
• RQ4 Is the proposed filtering approach universally applicable across different read mappers and datasets?
• RQ5 Can the memory bandwidth bottleneck in seed filtering be effectively overcome through algorithm-hardware co-design?

Key findings

• GRIM-Filter reduces the false negative rate by 5.59x–6.41x compared to FastHASH at a 5% error threshold, significantly improving sensitivity.
• It achieves 1.81x–3.65x end-to-end speedup over mrFAST with FastHASH, the previous fastest mapper, across multiple benchmarks.
• The algorithm reduces computation time on false negative locations by 83.7% on average, minimizing wasted alignment effort.
• Performance gains increase with higher error thresholds due to greater filtering efficiency, with the largest gains observed at 5%.
• GRIM-Filter is compatible with any read mapper and can be combined with other hardware-optimized components for further acceleration.
• The study demonstrates that algorithmic design tailored to emerging memory technologies like 3D-stacked DRAM can overcome memory bandwidth bottlenecks in bioinformatics.