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

Abstract

We introduce and study a set of training-free methods of an information-theoretic and algorithmic complexity nature that we apply to DNA sequences to identify their potential to identify nucleosomal binding sites. We test the measures on well-studied genomic sequences of different sizes drawn from different sources. The measures reveal the known in vivo versus in vitro predictive discrepancies and uncover their potential to pinpoint high and low nucleosome occupancy. We explore different possible signals within and beyond the nucleosome length and find that the complexity indices are informative of nucleosome occupancy. We found that, while it is clear that the gold standard Kaplan model is driven by GC content (by design) and by k-mer training; for high occupancy, entropy and complexity-based scores are also informative and can complement the Kaplan model.

A Review of Methods for Estimating Algorithmic Complexity: Options, Challenges, and New Directions

“…Indeed, when applied to short strings, LZW and other statistical compression algorithms would fail, so they are found to be of very limited use in the short string regime. For example, in an approach that tested and compared compression in a problem of molecular biology, it was shown that CTM and BDM can be informative of nucleosome occupancy [ 41 ] when applied to short strings (a nucleosome of DNA is only around 146 base pairs long), a key feature in genetic regulation. The problem of nucleosome occupancy is one of the greatest challenges in molecular biology, believed to be second only to protein folding.…”

“…CTM offered the first ever classification and ranking of short strings (see Figure 3). This regime of short strings is of no small interest; they are among the most relevant for real-world applications such as perturbation analysis [ 40 ], molecular biology and genetics [ 41 ].…”

“…Some of the approaches to speeding things up have included the introduction of bias and making the search domain specific. It was not until we proposed numerical estimations based on AP [ 29 , 30 , 38 ] that applications of AP to diverse areas of science began to be conceived of and explored, including our own most recent applications [ 39 , 40 , 41 ]. We see our work as having offered an opportunity to open up a discussion and to explore alternatives to the otherwise dominant use of statistical lossless compression (see Table 1 ).…”

piCRISPR: Physically Informed Deep Learning Models for CRISPR/Cas9 Off-Target Cleavage Prediction

“…We have found that nucleosome organisation related features are likely to enhance model performance beyond the said five epigenetic markers, and therefore include here three base-pair resolved nucleosomal channels: W/S Score [15], Strong-Weak Nucleotide BDM (SW-BDM) [16], NuPoP (Occupancy) [17]. These scores showed the largest Spearman correlation 1 with cleavage activity when considering 13 distinct nucleosome organisation-related scores.…”

“…The importance of nucleosome presence informed features for successful predictions of some of our best-performing models supports our emphasis on physically informed features. For a more detailed version of these features, we refer to [15,16,17].…”