Extended Data Fig. 3: Predictors of chromatin radiality.

(a) Correlation between the log2 GPSeq score and the mean number of transcription start sites (TSS, one TSS per gene) at 1 Mb resolution (overlapping genomic windows, 100 kb step size). Each dot represents one out of 26,330 genomic windows analyzed. (b) Correlation between the log2 GPSeq score and the average RNA-seq reads count at 1 Mb resolution (overlapping genomic windows, 100 kb step size). Each dot represents one out of 26,330 genomic windows analyzed. (c) Correlation between the log2 GPSeq score (1 Mb resolution, overlapping genomic windows with 100 kb step size) and chromosome size in base-pairs (bp). Each dot represents a single 1 Mb genomic window. (d) Correlation between the log2 GPSeq score (chromosome resolution) and the median GC-content per Mb per chromosome. Each dot represents one chromosome. (e) Correlation between the log2 GPSeq score (1 Mb resolution, overlapping genomic windows with 100 kb step size) and the median GC-content per Mb per chromosome. Each dot represents a single 1 Mb window. n = 25,026 genomic windows (points) were analyzed. (f) Same as in (e) but at 100 kb resolution (non-overlapping windows). n = 25,342 genomic windows (points) were analyzed. (g) Predicted over observed chromosome-wide GPSeq score. The prediction is based on a multivariable model including both chromosome size and GC-content as described in the Methods. PE, prediction error. Dotted red line: bisector. Each dot represents one chromosome. (h) Same as in (g) but using 1 Mb overlapping genomic windows with 100 kb step and using GC-content, chromosome size, gene expression and gene density to model the GPSeq score. n = 26,293 genomic windows (points) were analyzed. In all the plots in the figure, PCC and SCC are the Pearson’s and Spearman’s correlation coefficient, respectively. Dashed red lines: linear regressions. All the source data for this figure are from HAP1 cells.