Practical selection of representative sets of RNA-seq samples using a hierarchical approach

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• RNA-sequencing (RNA-seq) data is abundant in public databases.
• Evaluating bioinformatics tools requires representative sample sets, but existing methods struggle with large datasets.
• Current sequence-based methods for selecting representative RNA-seq samples face memory and runtime challenges.

Purpose of the Study:

• To develop an efficient computational method for selecting representative RNA-seq samples from large databases.
• To enable comprehensive and unbiased evaluation of RNA-seq analysis tools.
• To overcome the computational limitations of direct similarity matrix computation.

Main Methods:

• Developed 'hierarchical representative set selection,' a novel divide-and-conquer algorithm.
• Applied a hierarchical approach to break down the selection process into manageable sub-selections.
• Utilized k-mer counting for sequence-based similarity assessment within a hierarchical framework.

Main Results:

• Hierarchical representative set selection achieves summarization quality comparable to direct methods.
• Achieved significant reductions in runtime (up to 8.4x) and memory usage (up to 5.35x).
• Demonstrated superior performance over random sampling on large RNA-seq datasets from the Sequence Read Archive (SRA).

Conclusions:

• Hierarchical representative set selection offers a practical and scalable solution for representative subset selection from large RNA-seq databases.
• The method facilitates more efficient and thorough evaluation of bioinformatics tools.
• Enables robust analysis of large-scale genomic data.