A comprehensive simulation study on classification of RNA-Seq data.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• RNA sequencing (RNA-Seq) is a key technology for gene expression profiling.
• Gene expression-based classification is crucial for molecular-level disease diagnosis and monitoring.
• Existing statistical methods often fail with RNA-Seq data due to its count-based nature and distributional assumptions.

Purpose of the Study:

• To compare the performance of various classification algorithms for RNA-Seq data.
• To investigate the impact of key parameters on classification accuracy.
• To identify optimal classification strategies for RNA-Seq gene expression data.

Main Methods:

• Comparison of count-based classifiers (PLDA, NBLDA) and microarray-based classifiers (SVM, RF) with data transformations.
• Evaluation of parameters including overdispersion, sample size, number of genes, and differential-expression rate.
• Comprehensive simulation study and validation with miRNA and mRNA experimental datasets.

Main Results:

• Classification accuracy improves with larger sample sizes and higher differential-expression rates.
• Decreasing the dispersion parameter and number of groups enhances classification performance.
• Data transformation is critical for applying microarray-based classifiers to RNA-Seq data.

Conclusions:

• Count-based classifiers, such as power-transformed Poisson Linear Discriminant Analysis (PLDA), are suitable for RNA-Seq data.
• Microarray-based classifiers, like Random Forests (RF) and Support Vector Machines (SVM), perform well after appropriate transformations (e.g., vst, rlog).
• The R/BIOCONDUCTOR package MLSeq offers tools for RNA-Seq data classification.