Natural microbial community compositions compared by a back-propagating neural network and cluster analysis of 5S rRNA

Abstract

The community compositions of free-living and particle-associated bacteria in the Chesapeake Bay estuary were analyzed by comparing banding patterns of stable low-molecular-weight RNA (SLMW RNA) which include 5S rRNA and tRNA molecules. By analyzing images of autoradiographs of SLMW RNAs on polyacrylamide gels, band intensities of 5S rRNA were converted to binary format for transmission to a back-propagating neural network (NN). The NN was trained to relate binary input to sample stations, collection times, positions in the water column, and sample types (e.g., particle-associated versus free-living communities). Dendrograms produced by using Euclidean distance and average and Ward's linkage methods on data of three independently trained NNs yielded the following results. (i) Community compositions of Chesapeake Bay water samples varied both seasonally and spatially. (ii) Although there was no difference in the compositions of free- living and particle-associated bacteria in the summer, these community types differed significantly in the winter. (iii) In the summer, most bay samples had a common 121-nucleotide 5S rRNA molecule. Although this band occurred in the top water of midbay samples, it did not occur in particle-associated communities of bottom-water samples. (iv) Regardless of the season, midbay samples had the greatest variety of 5S rRNA sizes. The utility of NNs for interpreting complex banding patterns in electrophoresis gels was demonstrated.