Fig. 4

Subnetwork comparison for the network constructed by Pearson, Spearman, FS-Weight and PCA-PMI methods for human gut microbiome. a Network generated based on four methods based on top 100 gut genera. Genera whose average abundance ranked top 100 are selected, based on which the four methods were compared. b Correlation of genus in the same subnetwork. Pearson and Spearman method detect non-related correlations which could be filtered out by loose definition method. Based on indirect neighbor (FS-Weight) and non-linear correlations (PCA-PMI) calculations, previously hidden correlations (between Syntrophomonas and Methanosphaera, for example) could be discovered. In (a) and (b), each node represents a genus and the node size represented the average abundance in all samples. Edges indicates the correlations and red edges represent the positive correlations and green edges represent the negative correlations. c The possible interaction mechanism in the subnetwork for (b). Non-digestible carbohydrates could be degraded by Bacteroides and Ruminococcus to produce the short-chain fatty acid. Further reactions are generated between Syntrophomonas and methanogens to transfer the short-chain fatty acid into the methane. d Correlation distribution for four methods in non-digestible carbohydrate pathway. We calculated the correlations between Syntrophomonas and other genera (Bacteroides, Ruminococcus, Methanosphaera, Methanobrevibacter) detected by different methods (P: Pearson, S: Spearman, F: FS-Weight, P-P: PCA-PMI), and a tick represented the correlation was detected, and no detection was indicated by a cross