Novel algorithm development, user support and maintenance for STAR

点击翻译按钮获取中文摘要

Abstract. Sequencing of transcribed RNA molecules (RNA-seq) is an invaluable tool for studying cell transcriptomes at high resolution and depth. STAR is a popular RNA-seq analysis suite that combines high accuracy and ultra- fast speed of mapping with a reach collection of built-in features and tools. STAR is used by hundreds of researchers, including several major consortia and institutions. We propose to significantly enhance and expand STAR capabilities in the following important areas. 1. Develop novel algorithms and tools integrated directly into STAR. RNA-seq analyses require combining multiple tools into “processing pipelines” which is demanding task owing to bottlenecks and compatibility issues. We aim to overcome these impediments by integrating novel tools directly into STAR software: (i) mapping of RNA-seq reads to personal genomes utilizing genotype information to produce more accurate allele aware alignments, thus increasing precision of personal genomics analyses; (ii) mapping of long RNA reads from emerging sequencing technologies such as PacBio and Oxford Nanopore. 2. Increase accuracy and speed and of the core mapping algorithm. New applications, such as personal genomics, require significant improvements in mapping accuracy. We will enhance STAR mapping algorithm with (i) spliced seed extension through mismatches/indels; and (ii) limited local alignment so of the read ends. Tremendous increase of sequencing throughput has put a significant emphasis on the efficiency of the computational algorithms. To keep up with the increasing sequencing throughput, we will boost STAR algorithm with (i) vectorization of query-text comparisons using SIMD/SSE instructions; (ii) dynamical programming for seed stitching. The improvements in accuracy and speed will be validated in both simulated and real RNA-seq data. Mapping accuracy depends strongly on choosing the best mapping parameters for a particular dataset. We will devise automated parameter optimization procedures to eliminate guesswork in parameter selection. 3. Enhance user-friendliness, user support/education, and software maintenance. User-friendliness is crucial for bioinformatics software usefulness to the broadest audience. We aim to significantly enhance users' experience by developing STAR web user interfaces for both pre-run data input, and post-run exploring of results. To enable STAR analysis in the cloud, we will create STAR virtual machines on popular Amazon and Google cloud computing services, and develop Hadoop-based tools for distribute processing of the big datasets. We will also expand user support and education, continue to implement user- requested features and debug user-reported issues.