Identification of small open reading frames in plant lncRNA using class-imbalance learning

Area of Science:

• Genomics
• Computational Biology
• Plant Science

Background:

• Small open reading frames (sORFs) in long noncoding RNAs (lncRNAs) encode peptides crucial for organismal growth and development.
• Computational methods offer a cost-effective alternative to experimental approaches for identifying sORFs.
• Limited computational tools and resources exist for identifying sORFs in plant lncRNAs, particularly addressing class imbalance issues in machine learning models.

Purpose of the Study:

• To present a novel computational method, sORFplnc, for identifying sORFs with coding potential in plant lncRNAs.
• To address the challenge of class imbalance in machine learning for sORF identification.
• To provide a robust computational tool for plant lncRNA research.

Main Methods:

• Developed a hybrid sampling technique, WCDSMOTE-ENN, combining SMOTE with weighted cosine distance (WCDSMOTE) for minority class synthesis and weighted edited nearest neighbor (WENN) for majority class cleaning.
• Implemented a heterogeneous classifier ensemble for the classification task.
• Integrated feature selection with the WCDSMOTE algorithm.

Main Results:

• The proposed sORFplnc method demonstrates superior performance in identifying plant lncRNA sORFs with coding potential compared to existing computational methods.
• The WCDSMOTE-ENN approach effectively handles imbalanced datasets, improving classifier accuracy.
• The heterogeneous classifier ensemble contributed to robust classification performance.

Conclusions:

• sORFplnc offers a significant advancement in the computational identification of coding-potential sORFs in plant lncRNAs.
• The class-imbalance learning strategies enhance the reliability of sORF prediction.
• This work provides a valuable resource for plant biology, agriculture, and biomedicine.