An assessment of prediction algorithms for nucleosome positioning

Area of Science:

• Genomics
• Computational Biology
• Molecular Biology

Background:

• Nucleosome configuration influences eukaryotic genome regulation.
• Numerous computational tools exist for nucleosome positioning prediction.
• A third-party benchmark for evaluating these tools is lacking.

Purpose of the Study:

• To evaluate the performance of computational tools for nucleosome positioning prediction.
• To identify general and species-specific DNA motifs within nucleosomal and linker DNA.
• To assess cross-species prediction accuracy and its implications.

Main Methods:

• Genome-scale in vivo nucleosome maps from two vertebrates and three invertebrates were used.
• Performance evaluation of updated Segal's model and Gupta's SVM with RBF kernel.
• Analysis of over- and under-represented DNA oligomers in nucleosomal and linker DNA.

Main Results:

• Updated Segal's model and Gupta's SVM showed higher prediction accuracy.
• Performance varied significantly across species, indicating species-specific nucleosomal DNA characteristics.
• Commonly enriched oligomers across all five eukaryotes were CA/TG and AC/GT.
• Cross-species prediction highlighted specific DNA features in medaka and budding yeast.

Conclusions:

• Species-specific training data is desirable for high-performance nucleosome positioning prediction.
• Understanding DNA motifs in nucleosomal and linker regions is key to improving prediction accuracy.
• Gupta's SVM with RBF kernel demonstrates potential for accurate nucleosome positioning prediction.