TABLE 2.

Results of the permutation statistics–the community configurations are similar but still significantly specific for each group at each timepoint

	TP1
	Community 1	Community 2	Community 3
REL versus LDC	0% (p < .0002)	0.02% (p = .0006)	0.02% (p < .0006)
ABS versus LDC	0% (p < .0002)	0.04% (p =.0012)	0% (p < .0002)
REL versus ABS	0% (p < .0002)	0% (p < .0002)	2.96% (ns)
ABS versus REL	0% (p < .0002)	0.42% (p = .0126)	0.06% (p = .0018)
	TP 2
	Community 1	Community 2	Community 3
REL versus LDC	0.52% (p = .0156)	0% (p < .0002)	5.2% (ns)
REL versus ABS	3.36% (ns)	0.02% (p = .0006)	0.82% (p = .0246)
	Core‐Community 1	Community 2a	Community 2b	Community 3
ABS versus LDC	0% (p < .0002)	0% (p < .0002)	0% (p < .0002)	0% (p < .0002)
ABS versus REL	0% (p < .0002)	0% (p < .0002)	0.12% (p = .0048)	0.92% (p = .0368)

The first column of the table specifies the two groups whose data were permutated, e.g, REL versus LDC at TP1 means that we tested how probable it is that the relapser‐specific community configuration at timepoint 1 could have been observed by chance in the controls at timepoint 1. The first number is the frequency in percent that the community configuration was observed across the 5,000 iterations, the corresponding p value (Bonferroni corrected) is given in brackets. For example, the exact node combination of community 2 in relapsers versus controls at timepoint 1 was only detected in 0.02% of the 5,000 iterations. A percentage that small corresponds to a p value = 0.0006 thereby community 2 is significantly specific for relapsers at timepoint 1.

Abbreviations: ABS, abstainers; LDC, light drinking controls; REL, relapsers; TP, timepoint.