Improving tRNAscan-SE Annotation Results via Ensemble Classifiers.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Transfer RNA (tRNA) annotation is crucial for understanding gene expression and regulation.
• Existing tRNA detection tools like tRNAscan-SE exhibit an unacceptable false positive rate for large genomic sequences.
• Improving the accuracy of tRNA prediction is essential for comprehensive genomic analysis.

Purpose of the Study:

• To develop a machine learning-based predictor, tRNA-Predict, to enhance the accuracy of tRNA detection.
• To reduce the false positive rate associated with current tRNA annotation methods.
• To provide a tool that can refine genome-wide tRNA predictions.

Main Methods:

• Utilized machine learning to design tRNA-Predict, an ensemble classifier using LibMutil.
• Generated training datasets comprising real and pseudo-tRNA sequences identified by tRNAscan-SE.
• Constructed three distinct tRNA feature sets for model training and validation.
• Employed a positive dataset of 623 tRNA sequences from tRNAdb 2009 and a negative dataset of false positives from tRNAscan-SE.

Main Results:

• Achieved a prediction accuracy rate of 95.1% for tRNA-Predict using 10-fold cross-validation in silico.
• Demonstrated tRNA-Predict's capability to effectively distinguish functional tRNAs from pseudo-tRNAs.
• Showcased tRNA-Predict's utility in improving tRNAscan-SE's genome-wide annotation results.

Conclusions:

• tRNA-Predict offers a significant improvement over existing methods for accurate tRNA identification.
• The tool effectively filters false positives, enhancing the reliability of tRNA annotation pipelines.
• tRNA-Predict serves as a valuable complementary tool for genomic sequence analysis and tRNA research.