Machine learning algorithm for precise prediction of 2'-O-methylation (Nm) sites from experimental RiboMethSeq...

Area of Science:

• Molecular Biology
• Bioinformatics
• Genomics

Background:

• Deep sequencing enables RNA modification analysis, but interpretation of large datasets, especially for low-abundance RNAs, is challenging due to false positives/negatives.
• Current methods rely on subjective classification of modification sites using arbitrary thresholds, limiting accuracy for complex datasets like mRNA and lncRNA.

Purpose of the Study:

• To develop a robust and accurate method for analyzing epitranscriptomic RNA modifications.
• To overcome limitations of existing approaches in identifying RNA modification sites, particularly in complex RNA molecules.

Main Methods:

• Applied a machine learning algorithm, specifically Random Forest (RF), to create a predictive model.
• Trained the RF model on extensive human ribosomal RNA (rRNA) datasets with established modification profiles.
• Validated the model's predictive performance using experimentally determined profiles from eukaryotic rRNAs (S. cerevisiae and A. thaliana).

Main Results:

• Developed a Random Forest model for predicting 2'-O-methylated RNA sites from RiboMethSeq data.
• Demonstrated the model's effectiveness by assessing its performance on diverse eukaryotic rRNA datasets.
• Established a more objective and accurate approach for classifying potentially modified RNA positions.

Conclusions:

• The developed Random Forest model offers a significant improvement for analyzing epitranscriptomic data, particularly for complex RNA molecules.
• This machine learning approach enhances the specificity and selectivity of RNA modification site identification, reducing false positives and negatives.
• The study discusses the potential application of this predictive model for detecting other RNA modifications and analyzing more complex datasets.