This study investigates the ability of co-occurrence networks to reveal microbial interactions and identify keystone species in complex microbial ecosystems. Using generalized Lotka-Volterra dynamics, we simulated multi-species microbial communities with known interaction patterns and constructed co-occurrence networks to evaluate how well they capture underlying interactions. We found that co-occurrence networks can recapitulate interaction networks under certain conditions, but lose interpretability when habitat filtering becomes significant. Networks suffer from local hot spots of spurious correlation near hub species with many interactions. We identified topological features associated with keystone species in co-occurrence networks. This study provides a framework for environmental microbiologists to construct and interpret co-occurrence networks from microbial survey datasets. Co-occurrence networks are constructed from microbial survey sequencing data to identify interactions between community members. While this approach has potential to reveal ecological processes, it has been insufficiently validated due to technical limitations in studying complex microbial ecosystems. We simulated microbial communities with known interaction patterns using generalized Lotka-Volterra dynamics and constructed co-occurrence networks to evaluate their ability to reveal underlying interactions. We found that co-occurrence networks can recapitulate interaction networks under certain conditions, but lose interpretability when habitat filtering becomes significant. Networks suffer from local hot spots of spurious correlation near hub species with many interactions. We identified topological features associated with keystone species in co-occurrence networks. This study provides a substantiated framework to guide environmental microbiologists in the construction and interpretation of co-occurrence networks from microbial survey datasets. Co-occurrence networks were produced by applying an association metric or correlation coefficient to the simulated abundance data in a pair-wise manner. Statistically significant aggregation or avoidance was determined by generating a null distribution for each species pair by shuffling the site-abundance of one of the species and re-calculating the association metric. This resampling was performed 1000 times and the resulting distribution was used to generate p-values for observed association metric. P-values were corrected for multiple comparisons using the method of Benjamini and Hochberg (1995) and p-values less than p = 0.05 were considered to be statistically significant edges in the network. The sparCC program (Friedman and Alm, 2012), which was used for treatment of relative abundance data, uses a similar approach based on matrix permutation and null distribution generation. We evaluated the topological properties of both the interaction and the co-occurrence network using the package igraph (Csardi and Nepusz, 2006) in the R environment. We were particularly interested in properties potentially relevant for community roles and functioning as previously hypothesized in (Faust and Raes, 2012) and references therein, these are: (i) mean degree <k>: the degree of a node counts the number of edges it has. The mean degree is calculated over allThis study investigates the ability of co-occurrence networks to reveal microbial interactions and identify keystone species in complex microbial ecosystems. Using generalized Lotka-Volterra dynamics, we simulated multi-species microbial communities with known interaction patterns and constructed co-occurrence networks to evaluate how well they capture underlying interactions. We found that co-occurrence networks can recapitulate interaction networks under certain conditions, but lose interpretability when habitat filtering becomes significant. Networks suffer from local hot spots of spurious correlation near hub species with many interactions. We identified topological features associated with keystone species in co-occurrence networks. This study provides a framework for environmental microbiologists to construct and interpret co-occurrence networks from microbial survey datasets. Co-occurrence networks are constructed from microbial survey sequencing data to identify interactions between community members. While this approach has potential to reveal ecological processes, it has been insufficiently validated due to technical limitations in studying complex microbial ecosystems. We simulated microbial communities with known interaction patterns using generalized Lotka-Volterra dynamics and constructed co-occurrence networks to evaluate their ability to reveal underlying interactions. We found that co-occurrence networks can recapitulate interaction networks under certain conditions, but lose interpretability when habitat filtering becomes significant. Networks suffer from local hot spots of spurious correlation near hub species with many interactions. We identified topological features associated with keystone species in co-occurrence networks. This study provides a substantiated framework to guide environmental microbiologists in the construction and interpretation of co-occurrence networks from microbial survey datasets. Co-occurrence networks were produced by applying an association metric or correlation coefficient to the simulated abundance data in a pair-wise manner. Statistically significant aggregation or avoidance was determined by generating a null distribution for each species pair by shuffling the site-abundance of one of the species and re-calculating the association metric. This resampling was performed 1000 times and the resulting distribution was used to generate p-values for observed association metric. P-values were corrected for multiple comparisons using the method of Benjamini and Hochberg (1995) and p-values less than p = 0.05 were considered to be statistically significant edges in the network. The sparCC program (Friedman and Alm, 2012), which was used for treatment of relative abundance data, uses a similar approach based on matrix permutation and null distribution generation. We evaluated the topological properties of both the interaction and the co-occurrence network using the package igraph (Csardi and Nepusz, 2006) in the R environment. We were particularly interested in properties potentially relevant for community roles and functioning as previously hypothesized in (Faust and Raes, 2012) and references therein, these are: (i) mean degree <k>: the degree of a node counts the number of edges it has. The mean degree is calculated over all