Delineating the impact of machine learning elements in pre-microRNA detection

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Gene regulation involves transcription factors and post-transcriptional mechanisms like microRNAs (miRNAs).
• Experimental identification of all miRNA targets is challenging, necessitating computational methods.
• Machine learning (ML) is widely used for miRNA target prediction, but optimal approaches are unclear.

Purpose of the Study:

• To investigate the impact of different machine learning components on pre-microRNA detection performance.
• To establish the relative importance of various ML aspects in pre-miRNA identification.
• To analyze the effect of novel negative datasets on ML model training and testing.

Main Methods:

• Exploration of various machine learning algorithms, feature extraction techniques, and parameter optimization strategies.
• Development and application of two new negative datasets for training and testing ML models.
• Systematic analysis of the interplay between different ML components for pre-miRNA detection.

Main Results:

• No single ML component was found to be definitively superior; all parts are intricately connected.
• The performance of ML models for pre-miRNA detection is highly dependent on the combination of chosen components.
• The newly established negative datasets influenced model training and testing outcomes.

Conclusions:

• A comprehensive exploration of various scenarios is necessary for effective ML-based pre-miRNA detection.
• The interconnectedness of ML components suggests a holistic approach is required, rather than optimizing individual parts.
• Further research should focus on understanding these complex interactions for improved miRNA target prediction.