FET: Small: AlignMEM: Fast and Efficient DNA Sequence Alignment in Non-Volatile Magnetic RAM

最先进的DNA测序技术可以在单次运行中产生tb的DNA序列数据，其吞吐量预计将在未来几年每年增加3-5倍。为了将这些大的dna数据应用于后续复杂的疾病诊断/预后，如癌症风险评估、量身定制患者治疗和产前检测，它们必须首先与32亿长度的人类参考基因组相匹配。然而，为了这个目的，现有的软件工具可能需要数小时或数天才能对齐如此大量的DNA序列数据，即使是在今天非常强大的计算系统中，由于最先进的计算架构中的“内存墙”挑战，描述了存储单元和计算单元之间的速度不匹配。为此，该项目利用了基于非易失性纳米磁铁的磁性随机存取存储器（MRAM）技术和内存计算架构的创新。如果成功，它可以为下一代DNA序列分析系统实现高达两个数量级的计算性能，速度和能源效率，从而实现大规模快速基因组数据分析，以支持各种疾病研究和生物医学应用的研究。该项目将开发新的本科/研究生水平的内存计算架构和生物信息学课程模块。该项目将遵循两个主要研究方向。第一篇论文探讨了如何利用MRAM器件固有的非易失性特性，高效地开发DNA比对计算所需的超并行、可重构内存逻辑及其大型DNA数据内存处理（PIM）加速器架构。第二个研究方向将研究如何开发基于Burrows-Wheeler变换的快速内存DNA比对算法，以匹配所提出的基于MRAM的PIM平台及其在疾病表型预测中的大规模基因组分析应用。生成的比对将用于估计基因表达，并识别患者样本的单核苷酸突变事件，从而为疾病风险评估提供分子特征。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

The state-of-the-art DNA sequencing technologies could generate Terabytes of DNA sequence data in a single run, and their throughput is expected to increase 3-5 times each year in the coming years. In order to apply these big DNA-data into follow-up complex disease diagnostics/prognostics, such as cancer risk assessment, tailor patient treatment, and prenatal testing, they must be first aligned to a 3.2-billion-length human reference genome. However, the existing software tools for this purpose may need hours or days to align such large amount of DNA sequence data even with very powerful computing systems of today due to the 'memory wall' challenge in state-of-the-art computing architecture that describes the speed mismatch between memory units and computing units. To this end this, project leverages innovations from non-volatile nano-magnet based Magnetic Random Access Memory (MRAM) technology and in-memory computing architecture. If successful, it can achieve up to two orders magnitude higher computing performance, speed and energy efficiency for next-generation DNA sequence analysis system, which enables large-scale fast genomic data analytics to support research on various disease studies and biomedical applications. This project will develop new undergraduate/graduate level course modules on in-memory computing architecture and bioinformatics. This project will follow two main research tracks. The first one explores how to leverage the intrinsic non-volatile MRAM device property to efficiently develop ultra-parallel, reconfigurable in-memory logic required by DNA alignment computation and its big DNA-data Processing-in-Memory (PIM) accelerator architecture. The second research track will investigate how to develop fast DNA alignment-in-memory algorithm based on Burrows-Wheeler Transformation to match with the proposed MRAM based PIM platform and its large-scale genomic analysis application in disease phenotype prediction. Alignments generated will be used to estimate gene expression, and identify single nucleotide mutation events for patient samples, leading to molecular signatures for disease risk assessment.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.