Over 100 monthly bacterioplankton DNA samples, from each of the surface and 200 m depths at the Bermuda Atlantic Time-series Study (BATS) site, were analyzed for community assembly processes. Correlation networks, filtered for potential autocorrelation artifacts, were constructed for each depth. Network characteristics for the two depths were remarkably similar and the nodal taxonomic units (NTUs) with the greatest number of connections for both networks were dominated by the SAR86 clade in the Gammaproteobacteria class. Categories of NTUs within each network constructed using a weighted, phylogenetically-based similarity of connections measure suggested ecological drift processes. A clustering approach based on Tarjan's algorithm used directed graph theory to link taxa by similarity in connections to other taxa, revealing apparent examples of niche-based processes among taxa that have similar connections. An algorithm that used hierarchical Dirichlet Processes to model neutral communities based on learned parameters indicated that community assembly processes were neutral at the local 200 m level but not at the 200 m metacommunity level nor at either level of surface samples. However, surface samples restricted to SAR11 NTUs supported the hypothesis of neutral assembly processes, suggesting that neutral processes may apply to lower taxonomic groupings within the whole community. The greater annual disturbance from the deep mixing event at the surface of BATS may alter dispersal rates and drive community assembly away from neutral processes towards niche-based processes.