MLDeCNV: A machine learning approach for predicting copy number variation types in plant genomes

Area of Science:

• Plant genomics
• Bioinformatics
• Computational biology

Background:

• Copy number variations (CNVs) are key drivers of genetic diversity and plant traits.
• Plant genomes present challenges for CNV characterization due to complexity and repetitive sequences.

Purpose of the Study:

• To introduce MLDeCNV, an open-source machine learning tool for classifying CNV types in plants.
• To improve the accuracy of CNV type categorization in downstream genomic analyses.

Main Methods:

• Developed MLDeCNV using the XGBoost model, incorporating 32 CNV-related features.
• Trained the model on a high-confidence dataset of validated and predicted CNVs.
• Evaluated performance across various CNV sizes and dataset sizes.

Main Results:

• MLDeCNV achieved high accuracy (89.27%), precision, recall, and F1-score (89.3%), with an AUC of 0.9783.
• XGBoost outperformed traditional machine learning models in handling complex CNV data.
• The tool reliably classifies deletions, duplications, and non-CNVs from pre-identified regions.

Conclusions:

• MLDeCNV is a robust and reliable post-detection classification tool for plant CNVs.
• Its integration into CNV detection pipelines can streamline genomic analyses and enhance understanding of genetic variation.
• The tool is accessible online for broader research application.