Comparative analysis and prediction of nucleosome positioning using integrative feature representation and machine...

Area of Science:

• Genomics
• Bioinformatics
• Molecular Biology

Background:

• Nucleosome positioning is crucial for genome expression, DNA replication, repair, and transcription.
• Understanding nucleosome positioning is vital for numerous biological processes.
• Diverse DNA sequence representation methods exist, prompting exploration of integrated approaches.

Purpose of the Study:

• To investigate the impact of integrating multiple DNA sequence features on nucleosome positioning analysis.
• To enhance the theoretical understanding of nucleosome positioning.
• To evaluate novel feature integration strategies for improved prediction accuracy.

Main Methods:

• Utilized frequency chaos game representation (FCGR) for DNA sequence feature construction.
• Integrated FCGR with other sequence features.
• Applied principal component analysis (PCA) for feature dimensionality reduction.
• Employed machine learning models including SVM, ELM, XGBoost, MLP, and CNN for prediction.

Main Results:

• Integrated feature vectors demonstrated significantly superior prediction quality compared to single features.
• PCA-based dimensionality reduction notably improved prediction accuracy for the H. sapiens dataset.
• The study confirmed the effectiveness of various machine learning algorithms in nucleosome positioning prediction.

Conclusions:

• Frequency chaos game representation (FCGR) is a feasible approach for nucleosome positioning analysis.
• Integrative feature representation offers superior performance over single-feature methods.
• The findings are validated across multiple datasets, including H. sapiens, C. elegans, D. melanogaster, and S. cerevisiae.