Minerva and minepy: a C engine for the MINE suite and its R, Python and MATLAB wrappers

Area of Science:

• Bioinformatics
• Computational Biology
• Data Analysis

Background:

• Maximal Information-based Non-parametric Exploration (MINE) algorithms are valuable for assessing variable dependence.
• Previous implementations, like the Java version, had significant memory requirements, limiting their use with large datasets.
• Integrating robust dependence measures into bioinformatics pipelines is crucial for biological data analysis.

Purpose of the Study:

• To introduce a novel ANSI C implementation of the MINE algorithms.
• To develop libraries with a low memory footprint for efficient computation on large datasets.
• To facilitate the integration of maximal information-based measures of dependence into bioinformatics pipelines.

Main Methods:

• Developed ANSI C libraries for MINE algorithms, focusing on memory efficiency.
• Created minerva (R interface) and minepy (Python, MATLAB, Octave, C++) libraries.
• Implemented native parallelization for the R interface.
• Tested performance on MINE benchmarks and large-scale transcriptomics datasets (microarray and RNA-seq).

Main Results:

• The C implementation significantly reduces memory requirements compared to the Java version.
• The software demonstrates good upscaling properties for handling large datasets.
• Performance was validated on datasets with up to 1340 samples.
• Native parallelization in the R interface enhances computational speed.

Conclusions:

• The new C implementation of MINE algorithms provides an efficient and scalable solution for calculating maximal information-based measures of dependence.
• These libraries offer a low-memory footprint, making them suitable for large-scale bioinformatics analyses.
• The availability of minerva and minepy facilitates their adoption in diverse computational environments.