A General-Purpose Machine Learning R Library for Sparse Kernels Methods With an Application for Genome-Based...

Area of Science:

• Computational biology
• Bioinformatics
• Machine Learning

Background:

• Machine learning adoption is expanding beyond computer science due to user-friendly tools.
• Existing tools often require advanced programming knowledge, limiting accessibility.

Purpose of the Study:

• To present the sparse kernel methods (SKM) R package for accessible supervised machine learning.
• To provide efficient implementation of six popular algorithms with optional sparse kernels.
• To introduce novel sparse kernel computation for resource reduction.

Main Methods:

• Developed an R package (SKM) implementing generalized boosted machines, generalized linear models, support vector machines, random forest, Bayesian regression, and deep neural networks.
• Integrated a function for computing seven kernel types (Linear, Polynomial, Sigmoid, Gaussian, Exponential, Arc-Cosine 1, Arc-Cosine L) and their sparse versions.
• Evaluated SKM performance in a genome-based prediction framework using maize and wheat datasets.

Main Results:

• The SKM package offers user-friendly access to key machine learning algorithms.
• The sparse kernel functionality significantly reduces computational resources for kernel methods.
• SKM demonstrates effective performance in genome-based prediction, comparable to standard methods.

Conclusions:

• The SKM package democratizes machine learning application across various scientific fields.
• Sparse kernel computation is crucial for efficient and performant kernel machine learning.
• SKM is a versatile tool applicable to diverse prediction problems beyond genomics.