Identification of non-coding RNAs with a new composite feature in the Hybrid Random Forest Ensemble algorithm

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Non-coding RNAs (ncRNAs) play crucial roles in gene regulation, but their identification in complex genomes remains challenging.
• Accurate discrimination between various ncRNA types, including short and long non-coding RNAs (lncRNAs), is essential for understanding their functions.

Purpose of the Study:

• To develop and validate a computational tool for efficient and accurate identification of ncRNA signals in genomic regions.
• To improve the characterization of long non-coding RNA (lncRNA) elements through a novel feature-based approach.

Main Methods:

• A hybrid random forest (RF) and logistic regression model was developed for ncRNA classification.
• A discriminative feature set was curated, including a novel feature named SCORE, derived from structural, sequence, modularity, robustness, and coding potential attributes.
• The classifier was trained on a balanced dataset and evaluated for accuracy, sensitivity, and specificity.

Main Results:

• The RF-based classifier achieved high performance metrics: 92.11% accuracy, 90.7% sensitivity, and 93.5% specificity.
• The novel SCORE feature significantly enhanced the RF classifier's ability to identify Rfam lncRNA families.
• The genome-wide framework demonstrated high sensitivity (>90% for prokaryotes, 77.7% for eukaryotes) in identifying known ncRNAs across diverse organisms.

Conclusions:

• The developed hybrid RF classifier provides an efficient and accurate method for ncRNA identification.
• The SCORE feature offers improved characterization of lncRNA elements, advancing ncRNA research.
• The framework's applicability across various organisms highlights its potential for broad genomic analysis and discovery.