The nuclear repeat length (NRL) was calculated using NRLfinder as previous publication.33 Briefly, read lengths were extracted and converted into a frequency histogram, which was then smoothed using a digital 6th-order Butterworth filter with a zero-phase shift and a cutoff frequency of 0.04 cycles/read. This cutoff was empirically optimized to reduce noise from mononucleosomal DNA winding artifacts. Local minima and maxima were identified from the first derivative of the filtered histogram, with the second peak maximum corresponding to the dinucleosomal periodicity. The NRL shift between conditions (e.g., control vs. NuMA-depleted HCT116 cells) was calculated the mean difference between the first two peak maxima of each sample. All analyses were performed in Python 3.9 with NumPy, SciPy, and Matplotlib libraries.
For chromatin-state modeling, we used the ChromHMM (v.1.19).32 The input data of ATAC-seq and RNA-seq reported in this manuscript was generated as described above. Additional input data including ChIP-seq for CTCF, H3K4me3, H3K27me3, H3K4me1, H3K36me3 and H3K9me3 were download from ENCODE (https://www.encodeproject.org). briefly, raw bam files were download and replicates were combined. BinarizeBam and LearnModel tools in ChromHMM was used to generate chromatin state model with default settings. Emissions parameters were visualized in R.







