Gomez-Vela, F. | Search bioRxiv

bioRxiv2020

A comparative study of supervised machine learning algorithms for the prediction of long-range chromatin interactions

The role of three-dimensional genome organization as a critical regulator of gene expression has become increasingly clear over the last decade. Most of our understanding of this association comes from the study of long range chromatin interaction maps provided by Chromatin Conformation Capture-based techniques, which have greatly improved in recent years. Since these procedures are experimentally laborious and expensive, in silico prediction has emerged as an alternative strategy to generate virtual maps in cell types and conditions for which experimental data of chromatin interactions is not available. Several methods have been based on predictive models trained on one-dimensional (1D) sequencing features, yielding promising results. However, different approaches vary both in the way they model chromatin interactions and in the machine learning-based strategy they rely on, making it challenging to carry out performance comparison of existing methods. In this study, we use publicly available 1D sequencing signals to model chromatin interactions in two human cell lines and evaluate the prediction performance of 5 popular machine learning algorithms: decision trees, random forests, gradient boosting, support vector machines and multi-layer perceptron. Our approach accurately predicts long-range interactions and reveals that gradient boosting significantly outperforms the other four algorithms, yielding accuracies of ~ 95%. We show that chromatin features in close genomic proximity to the anchors cover most of the predictive information. Moreover, we demonstrate that gradient boosting models trained with different subsets of chromatin features, unlike the other methods tested, are able to produce accurate predictions. In this regard, and besides architectural proteins, transcription factors are shown to be highly informative. Our study provides a framework for the systematic prediction of long-range chromatin interactions, identifies gradient boosting as the best suited algorithm for this task and highlights cell-type specific binding of transcription factors at the anchors as important determinants of chromatin wiring.

bioinformatics↗

bioRxiv2020

Genome-wide prediction of topoisomerase IIβ binding by architectural factors and chromatin accessibility

DNA topoisomerase II-{beta} (TOP2B) is fundamental to remove topological problems linked to DNA metabolism and 3D chromatin architecture, but its cut-and-reseal catalytic mechanism can accidentally cause DNA double-strand breaks (DSBs) that can seriously compromise genome integrity. Understanding the factors that determine the genome-wide distribution of TOP2B is therefore not only essential for a complete knowledge of genome dynamics and organization, but also for the implications of TOP2-induced DSBs in the origin of oncogenic translocations and other types of chromosomal rearrangements. Here, we conduct a machine-learning approach for the prediction of TOP2B binding sites using publicly available sequencing data. We achieve highly accurate predictions, with accessible chromatin and architectural factors being the most informative features. Strikingly, TOP2B is sufficiently explained by only three features: DNase I hypersensitivity, CTCF and cohesin binding, for which genome-wide data are widely available. Based on this, we develop a predictive model for TOP2B genome-wide binding that can be used across cell lines and species, and generate virtual probability tracks that accurately mirror experimental ChIP-seq data. Our results deepen our knowledge on how the accessibility and 3D organization of chromatin determine TOP2B function, and constitute a proof of principle regarding the in silico prediction of sequence-independent chromatin-binding factors. Author summaryType II DNA topoisomerases (TOP2) are a double-edged sword. They solve topological problems in the form of supercoiling, knots and tangles that inevitably accompany genome metabolism, but they do so at the cost of transiently cleaving DNA, with the risk that this entails for genome integrity, and the serious consequences for human health, such as neurodegeneration, developmental disorders or predisposition to cancer. A comprehensive analysis of TOP2 distribution throughout the genome is therefore essential for a deep understanding of its function and regulation, and how this can affect genome dynamics and stability. Here, we use machine learning to thoroughly explore genome-wide binding of TOP2B, a vertebrate TOP2 paralog that has been linked to genome organization and cancer-associated translocations. Our analysis shows that TOP2B-DNA binding can be accurately predicted exclusively using information on DNA accessibility and binding of genome-architecture factors. We show that such information is enough to generate virtual maps of TOP2B binding along the genome, which we validate with de novo experimental data. Our results highlight the importance of TOP2B for accessibility and 3D organization of chromatin, and show that computationally predicted TOP2 maps can be accurately obtained using minimal publicly available datasets, opening the door for their use in different organisms, cell types and conditions with experimental and/or clinical relevance.

bioinformatics↗