[Paper Review] Hot RAD: A Tool for Analysis of Next-Gen RAD Tag Data

Lauren A. Assour, Nicholas LaRosa+1 more|arXiv (Cornell University)|Nov 20, 2015

Genomics and Phylogenetic Studies8 references3 citations

TL;DR

Hot RAD is a GUI-enabled, distributed pipeline for analyzing next-generation RAD-tag sequencing data, using Smith-Waterman-based clustering and consensus sequence generation to produce fewer, deeper sequence stacks from raw Illumina reads. It improves data usage and reduces memory load compared to existing tools, supporting both single- and paired-end reads with customizable filtering and distributed execution via Makeflow and Weaver.

ABSTRACT

Restriction site Associated DNA (RAD) tagging (also known as RAD-seq, etc.) is an emerging method for analyzing an organism's genome without completely sequencing it. This can be applied to a non-model organism without a reference genome, though this creates the problem of how to begin data analysis on unmapped and unannotated reads. Our program, Hot RAD, presents a straightforward and easy-to-use method to take raw Illumina data that has been RAD tagged and produce consensus contigs or sequence stacks using a distributed framework, creating a basis on which to begin analyzing an organism's DNA. The GUI (graphical user interface) element of our tool makes it easy for those not familiar with the command line to take raw sequence files and produce usable data in a timely manner.

Motivation & Objective

• To address the challenge of analyzing unmapped, unannotated RAD-tag sequencing data from non-model organisms without a reference genome.
• To reduce memory usage and improve data utilization by clustering reads into fewer, deeper sequence stacks rather than traditional de novo assembly.
• To provide a user-friendly, GUI-based interface for non-expert researchers to process RAD-seq data without command-line expertise.
• To enable scalable, distributed execution of RAD-tag analysis using Makeflow and Weaver for high-throughput data sets.
• To offer flexibility by allowing substitution of core components, such as alternative assemblers or alignment tools, for customized workflows.

Proposed method

• Uses a Python-based trimmer to remove barcodes, cut sites, and low-quality sequences (e.g., those with excessive Ns), while supporting variable-length barcodes and paired-end reads.
• Employs a banded Smith-Waterman algorithm to compute sequence identity and group reads into clusters based on user-defined percent identity thresholds.
• Applies a modified center star alignment method to identify the most representative consensus sequence (equivalent to a contig) within each cluster.
• Uses a tiered, distributed merging process with Smith-Waterman alignment to resolve overlapping or duplicate clusters across distributed processing nodes.
• Supports execution via GUI or command-line shell scripts, with optional integration into cloud or cluster environments using Makeflow and Weaver.
• Allows replacement of the default assembler with alternative tools, enabling modular customization of the analysis pipeline.

Experimental results

Research questions

• RQ1 How can RAD-tag sequencing data be processed efficiently when no reference genome is available?
• RQ2 Can a distributed, GUI-based pipeline improve data usage and reduce memory consumption compared to traditional assemblers in RAD-seq analysis?
• RQ3 To what extent can variable-length barcodes and heterogeneous read data be handled effectively in RAD-tag analysis?
• RQ4 How does the use of consensus sequence clustering compare to de novo assembly in terms of memory efficiency and data utilization?
• RQ5 Can a modular pipeline design support integration with alternative assemblers or alignment tools without sacrificing performance or usability?

Key findings

• Hot RAD processes more sequences than other RAD-tag tools, achieving higher data usage by clustering reads into fewer, deeper loci.
• With the 'BP + Fast Align' filter, Hot RAD reduced runtime from 773.1 minutes to 16.9 minutes while processing 18,524 reads and producing 3,108 clusters.
• The tool consistently uses more sequences than competing tools like Stacks and RADTools, especially in heterogeneous data sets such as oak, field-caught flies, and mosquitoes.
• Hot RAD’s distributed architecture enables scalable execution on clusters or clouds using Makeflow and Weaver, significantly reducing processing time for large data sets.
• The pipeline supports both single-end and paired-end reads, though it does not use the second read for placement, and allows optional removal of low-coverage clusters to speed processing.
• Hot RAD’s modular design allows replacement of the default assembler with other tools, such as SeqMan NGen, while maintaining compatibility with the core clustering and consensus generation pipeline.