Training-free measures based on algorithmic probability identify high nucleosome occupancy in DNA sequences.

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Nucleosomes are fundamental units of DNA packaging in eukaryotes.
• Accurate prediction of nucleosomal binding sites is crucial for understanding genome regulation.
• Existing models like the Kaplan model rely on specific features such as GC content and k-mer training.

Purpose of the Study:

• To introduce and evaluate novel training-free methods for identifying nucleosomal binding sites in DNA sequences.
• To assess the efficacy of information-theoretic and algorithmic complexity measures for predicting nucleosome occupancy.
• To compare the performance of these new methods against established models.

Main Methods:

• Application of information-theoretic and algorithmic complexity measures to DNA sequences.
• Testing the methods on diverse, well-studied genomic datasets of varying sizes.
• Analysis of signals within and beyond nucleosome length to identify predictive features.
• Comparison with the established Kaplan model.

Main Results:

• The developed measures successfully identified potential nucleosomal binding sites.
• The methods revealed discrepancies between in vivo and in vitro predictions.
• Complexity indices were found to be informative of nucleosome occupancy levels (high and low).
• Entropy and complexity-based scores provide complementary information to the Kaplan model, especially for high occupancy prediction.

Conclusions:

• Training-free, complexity-based methods offer a promising approach for predicting nucleosomal binding sites.
• These novel scores can effectively pinpoint regions of high and low nucleosome occupancy.
• The proposed methods can serve as valuable complements to existing predictive models, enhancing their accuracy and scope.