Probabilistic method corrects previously uncharacterized Hi-C artifact

Abstract

Three-dimensional chromosomal structure plays an important role in gene regulation. Chromosome conformation capture techniques, especially the high-throughput, sequencing-based technique Hi-C, provide new insights on spatial architectures of chromosomes. However, Hi-C data contains artifacts and systemic biases that substantially influence subsequent analysis. Computational models have been developed to address these biases explicitly, however, it is difficult to enumerate and eliminate all the biases in models. Other models are designed to correct biases implicitly, but they will also be invalid in some situations such as copy number variations. We characterize a new kind of artifact in Hi-C data. We find that this artifact is caused by incorrect alignment of Hi-C reads against approximate repeat regions and can lead to erroneous chromatin contact signals. The artifact cannot be corrected by current Hi-C correction methods. We design a probabilistic method and develop a new Hi-C processing pipeline by integrating our probabilistic method with the HiC-Pro pipeline. We find that the new pipeline can remove this new artifact effectively, while preserving important features of the original Hi-C matrices.