Local DNA Alignment Using Bit-Matrix Trace Regions for Variant Calling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for variant calling in next-generation sequencing, such as the GATK HaplotypeCaller, rely solely on the Smith-Waterman algorithm for local alignment, which is inefficient in terms of memory consumption and latency, and require multiple large adders for hardware acceleration, leading to suboptimal performance.
Innovation Solution
A method utilizing a bit matrix for local alignment, combined with a pair hidden Markov model and Bayes' Theorem, to determine haplotypes, and a system with distributed processing devices for efficient variant calling, minimizing memory overheads and latency through in-cache processing and optimized memory access.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the Smith-Waterman algorithm is used for local alignment in variant calling, then alignment accuracy is maintained, but memory consumption increases and processing latency increases
Solution Approach 1:
The patent segments the bit matrix into multiple blocks that can be processed independently and stored in cache memory. By dividing the large alignment matrix into smaller manageable blocks, the system maintains alignment accuracy while reducing peak memory consumption and improving cache utilization during the variant calling process.
Solution Approach 2:
The patent transforms the traditional two-dimensional bit matrix representation into a blocked structure that exploits spatial locality in both dimensions. This dimensional organization enables efficient cache access patterns and reduces memory bandwidth requirements while preserving the accuracy of local alignment results.
2Measurement precision
If the Smith-Waterman algorithm is used for local alignment in variant calling, then alignment accuracy is maintained, but processing speed decreases
Solution Approach 1:
The patent performs preliminary blocking and pre-computation of alignment scores within each block before final traceback. By preparing data structures and computing partial results in advance within cacheable blocks, the system maintains accurate alignment results while significantly improving processing speed through reduced memory access latency.
Solution Approach 2:
The patent ensures continuous processing by organizing computation to flow smoothly through blocked regions without interruption. The method maintains computational continuity by systematically processing adjacent blocks in sequence, keeping processing units utilized and avoiding idle wait states, thus improving overall processing speed while preserving alignment accuracy.
3Productivity
If multiple large adders are used for hardware acceleration of Smith-Waterman algorithm, then processing throughput is improved, but device complexity and hardware resource requirements increase
Solution Approach 1:
The patent segments the computational workload into independent block-processing units that can be executed with simpler, smaller adders. By dividing the alignment computation into blocked regions, the system achieves high throughput through parallel block processing rather than requiring multiple large adders, thus reducing hardware complexity while maintaining productivity.
4Ease of manufacture
If traditional memory access patterns are used in local alignment, then implementation simplicity is maintained, but memory access efficiency decreases
Solution Approach 1:
The patent reorganizes memory access patterns by introducing a block dimension to the traditional row-column access scheme. This blocked access pattern improves memory efficiency by exploiting spatial locality and prefetching opportunities, reducing memory access latency while maintaining implementation feasibility through systematic block traversal.
Data Source
AI summary
A method for performing local alignment based on a query sequence of DNA and a reference sequence of DNA includes: obtaining a bit matrix H; determining at least one diagonal based on the bit matrix H; for each of the at least one diagonal, calculating an initial score for the diagonal, determining at least one trace region, determining a sub-alignment for each of the at least one trace region, consolidating the diagonal and the sub-alignment respectively of the at least one trace region to obtain an alignment, and obtaining an alignment score based on the initial score and the partial score respectively of the sub-alignment respectively of the at least one trace region; and among each of the at least one alignment thus determined respectively for each of the at least one diagonal, reserving one of the at least one alignment that has the highest alignment score therefrom.