Novel domain-specific languages and compiler optimization methods for computational biology

动机：下一代测序（NGS）实验产生的大量数据需要开发高效的计算方法，因为计算方面是目前NGS管道的最大瓶颈。然而，许多有前途的方法无法应对当前和新兴的天然气监测技术的规模，而且难以使用和复制。这一问题的根源在于广泛使用的通用开发环境不能有效地表达和优化生物数据工作流。用户被迫要么使用高级但速度慢的语言，如Python，要么使用低级语言，如C语言，它可以产生高效的工具，但需要花费大量的时间和可维护性成本。方法：专门为生物和测序数据定制的领域特定语言（DSL）和相关的编译器优化技术套件将为实验新的计算算法提供灵活性、简单性和模块化，同时生成高性能代码，并使程序从资源受限的体系结构移植到最大的超级计算机。目的：我们提出了一种新的DSL和相关的编译器，名为Seq，可以快速轻松地开发高性能测序管道。为了实现这一目标，我们将：(i)设计一种编程语言和编译器，使开发高级语言（如Python）变得容易，同时提供低级语言（如C）的原始性能（目标1）；（ii）探索基因组工作流程中的数据访问模式，并设计出可以在编译器级别利用这些模式的方法，以便在各种计算环境（如多核cpu、gpu和手持设备）中自动进行低级优化（目标2）；（iii）提供将Seq轻松集成到流行的生物信息学和科学环境中的方法，并为NGS数据开发一个精心策划的算法原语库（目标3）。短期目标是开发一种能够在通用架构上有效处理各种NGS数据的DSL。长期目标是建立一个全面和广泛使用的基础设施，允许快速和容易的方法开发生物数据。由于加拿大对计算生物学HQP的需求很大，本提案的关键目标之一是在五年的时间里培养HQP。影响：我们设想我们的DSL将显著促进加拿大基因组学和健康研究，使研究人员能够以更自然的方式表达他们的想法，并允许他们使用最好的算法方法来完成这项工作。此外，我们希望我们的DSL能够通过节省大量的时间和成本来帮助加拿大的大型科学卫生项目。我们还预计Seq将成为广泛使用的生物信息学工具的关键组成部分。最后，我们期望通过这个项目培养的HQP能够为加拿大的知识经济做出贡献。

Motivation: The vast scale of data generated by next-generation sequencing (NGS) experiments necessitates the development of efficient computational methods, as the computational aspect is currently the biggest bottleneck of NGS pipelines. However, many promising methods cannot handle the scale of current and emerging NGS technologies and are too hard to use and replicate. Root cause of this problem lies in widely used general-purpose development environments that cannot efficiently express and optimize biological data workflows. Users are forced to use either high-level but slow languages such as Python, or low-level languages such as C that produce efficient tools but at a significant time and maintainability costs. Approach: A domain-specific language (DSL) and associated suite of compiler optimization techniques specifically tailored for biological and sequencing data would provide flexibility, simplicity and modularity for experimenting with new computational algorithms, while generating high-performance code and making the programs portable from resource-constrained architectures to the biggest supercomputers. Objectives: We propose a novel DSL and associated compiler named Seq that enables rapid and easy development of high-performance sequencing pipelines. To achieve this, we will: (i)design a programming language and a compiler that allows ease of development of high-level languages such as Python, while providing raw performance of low-level languages such as C (Objective 1); (ii)explore data access patterns in genomic workflows and devise methods that can exploit these patterns at the compiler level for automatic low-level optimizations across various computational environments, such as multicore CPUs, GPUs and handheld devices (Objective 2); and (iii)provide means to easily integrate Seq into popular bioinformatics and scientific environments and develop a curated library of algorithmic primitives for NGS data (Objective 3). The short-term goal is to develop a DSL that can efficiently handle various kinds of NGS data on common architectures. The long-term goal is to build a comprehensive and widely used infrastructure that allows rapid and easy method development for biological data. As computational biology HQP are in high demand in Canada, one of the key goals of this proposal is to train HQP over the course of five years. Impact: We envision our DSL to significantly boost Canadian genomics and health research by enabling researchers to express their ideas in a more natural way and by allowing them to use the best algorithmic methods for the job. Furthermore, we expect our DSL to aid large-scale scientific Canadian health projects by providing huge time and cost savings. We also anticipate Seq to become a key building block in the wide specter of widely used bioinformatics tools. Finally, we expect that HQP trained by this program will contribute to the Canadian knowledge-based economy.