Positional frequency chaos game representation for machine learning-based classification of crop lncRNAs

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Alignment-based methods are computationally intensive for large-scale genomic analyses.
• Alignment-free methods like k-mer analysis are faster but often lose crucial positional information.
• Long non-coding RNAs (lncRNAs) in plants are vital regulators of gene expression but require efficient identification methods.

Purpose of the Study:

• To introduce a novel alignment-free encoding method, positional frequency chaos game representation (PFCGR), for plant lncRNA classification.
• To enhance traditional frequency chaos game representation (FCGR) by incorporating positional statistics of k-mers.
• To enable accurate and computationally efficient classification of plant lncRNAs directly from genomic sequences using machine learning.

Main Methods:

• Developed PFCGR by integrating four statistical moments (mean, standard deviation, skewness, kurtosis) of k-mer positions into a multi-channel image representation.
• Utilized machine learning models including Logistic Regression, Random Forests, and Convolutional Neural Networks for lncRNA classification.
• Evaluated PFCGR performance on seven major crop species.

Main Results:

• PFCGR-based classifiers achieved classification accuracies comparable to or exceeding the computationally intensive DNABERT model.
• The proposed method requires significantly fewer computational resources compared to existing advanced models.
• Demonstrated the effectiveness of PFCGR in accurately classifying plant lncRNAs across diverse crop species.

Conclusions:

• PFCGR offers an efficient and accurate approach for plant lncRNA identification.
• The method preserves vital positional information lost in traditional alignment-free techniques.
• PFCGR facilitates large-scale computational genomics studies by reducing computational demands.