MLSeq: Machine learning interface for RNA-sequencing data.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• RNA sequencing (RNA-Seq) is preferred over microarrays for gene expression analysis due to lower data noise.
• Limited algorithms exist for RNA-Seq data classification compared to microarray data.
• The MLSeq package addresses this gap by integrating existing and novel classification methods for RNA-Seq.

Purpose of the Study:

• To develop MLSeq, an R package for comprehensive RNA-sequencing data classification.
• To provide a user-friendly platform for both preprocessing and classification of gene expression data.
• To introduce novel classification algorithms tailored for RNA-Seq data.

Main Methods:

• MLSeq preprocesses RNA-Seq data (normalization, filtering, transformation).
• It supports two classification strategies: direct RNA-Seq algorithms or transforming data for microarray algorithms.
• Novel algorithms like voom-based nearest shrunken centroids (voomNSC) were developed and included.

Main Results:

• Three real-world RNA-Seq datasets (cervical cancer, lung cancer, aging) were used for evaluation.
• Discrete distribution algorithms (PLDA, NBLDA) achieved >0.92 accuracy for cervical cancer and aging data.
• The voomNSC algorithm demonstrated high sparsity, selecting minimal features, particularly for cancer datasets.

Conclusions:

• MLSeq offers a comprehensive and user-friendly interface for RNA-Seq data classification.
• It serves as an integrated pipeline for preprocessing and classification tasks.
• The package facilitates advanced analysis of gene expression data using both established and novel methods.