Network measures estimated on fruit and seed datasets.

1p value based on 1,000 permutations.Individual ILs were ranked according to increasing difference in variance of the fruit dataset compared to the average variance in the seed dataset (Table S8). Network properties were calculated from a network reconstructed by using the data from the ordered list of ILs. For instance, for n = 25, the 25 ILs from Table S8 were used in creation of the correlation network associated to the data from only these n = 25 ILs. Subsets comprising the first 15 to 25, 50 and 76 of the fruit ILs ranked in non-decreasing order with respect to their variance were used to construct correlation-based networks (r≥0.3, p≤0.01). Four network properties were calculated for each subset-based fruit network: density, degree, clustering coefficient, and diameter (value). Values represent the estimates of the respective network measures for each subset of fruit ILs. By performing the classical permutation test with 1,000 repetitions, the statistical significance of the differences in measures between the subset-based fruit networks and the seed network (first data row in the table) were measured. In each permutation, the order of each metabolite within the subset was randomized, and the newly ordered dataset was subjected to correlation analysis and network measures estimation. The difference between newly generated network property values in the seed and the fruit, upon randomization, were tested to check whether their value is at most that of the difference for the original networks. Subsequently, the total number of occurrences meeting this criterion formed the basis for the empirical p-value estimation. With the exception of the network diameter, density, degree, and clustering coefficient of the fruit IL subset networks are significantly different from the corresponding measures in the seed network.