Application of machine learning methods to histone methylation ChIP-Seq data reveals H4R3me2 globally represses gene expression

Table 1 Multilinear model terms and statistics

Multilinear model term	β (trim mean)	Z (trim mean)	p (median)	Impact (trim mean)
H3K79me1 ^a	6.741	18.234	0	1.331
H3K36me3 ^a	4.087	17.802	0	0.922
H3K79me3 ^a	3.078	23.916	0	0.598
H4K20me1 ^a	0.977	21.446	0	0.450
H3K4me2 ^a * - H3R2me1 ^r	18.270	7.850	1.66E-15	0.437
H3K27me2 ^r * - H3R2me1 ^r	70.468	15.280	0	0.381
H3K9me2 ^r * - H3K27me1 ^r * - H4K20me1 ^a	37.041	5.643	9.47E-09	0.156
H3K4me3 ^a	0.133	5.729	5.20E-09	0.151
H2BK5me1 ^a * - H3K36me3 ^a	1.286	3.800	7.95E-05	0.115
H2BK5me1 ^a * - H4K20me1 ^a - H3R2me1 ^r	1.531	6.034	1.14E-09	0.030
Intercept	4.026	64.131	0	0
H3K9me3 ^r * - H3K36me3 ^a	-2.747	-12.439	0	-0.010
H3K4me2 ^a * - H3K36me3 ^a - H3K79me3 ^a	-1.274	-3.743	1.02E-04	-0.018
H3K36me3 ^a - H3K79me2 - H3R2me2	-5.855	-3.497	1.73E-04	-0.022
H3K27me3 ^r * - H3K79me2 - H3K79me3 ^a	-26.341	-6.380	5.78E-11	-0.026
H3K4me1 ^a * - H3K9me2 ^r * - H4K20me1 ^a	-4.563	-3.578	2.65E-04	-0.041
H3K9me1 ^a * - H3K27me1 ^r * - H4K20me1 ^a	-2.350	-5.078	1.92E-07	-0.077
H2BK5me1 ^a * - H4K20me1 ^a	-0.600	-9.627	0	-0.095
H3K36me1 - H3K79me1 ^a - H3K79me3 ^a	-27.840	-8.478	0	-0.115
H3K4me2 ^a * - H3K9me1 ^a *	-1.578	-3.340	3.57E-04	-0.123
H4R3me2 ^r	-11.121	-13.233	0	-0.301
H3K27me2 ^r * - H3K36me3 ^a	-31.772	-8.911	0	-0.311
H3K27me2 ^r * - H3K79me1 ^a	-56.535	-9.535	0	-0.449
H3R2me1 ^r	-11.937	-16.504	0	-0.596

Terms appearing in the final multilinear model, and associated statistics. Trim mean of the β coefficients, Z-scores and impact factors (β multiplied by amplitude interquartile range). Trim mean is defined as the mean of the population excluding the lowest and highest 5% of the data. The superscript labels each mark as activating (a) or repressive (r). Unstarred marks correspond to monovalent terms in the model, starred marks do not have a monovalent contribution in the model, but correlated/anti-correlated with gene expression based on univariate analysis, and uncolored marks do not have clear correlation with gene expression. Rows are sorted by the impact term value.