Application of Artificial Neural Networks for Interpreting Molecular Profiles of Microbial Communities.

Low resolution methods for profiling microbial communities use phospholipids and 16S rRNA genes patterns to characterize the composition of microorganisms in environmental samples. All of the DNA based methods used today such as DGGE, ARISA, RFLP, have pitfalls in being over interpreted with respect to describing microbial diversity, but provide useful inexpensive tools for comparing relative similarities of microbial community structures. PLFA has the advantage of providing estimates of fungal and bacterial biomass, and can also be associated with certain functional groups. The major limitation for interpreting the data generated by these methods has been the inability to separate out the effects of different factors (soil type, pH, plant species, growth stage, etc..) on markers for different functional groups. In many cases, combinations of variables will generate chaotic patterns that cannot be separated using routine statistical procedures. New methods employing artificial neural networks (ANN) for data analysis offer a novel and powerful approach for separating out the effects of environmental variables on microbial community structures. An example application of the ANN approach will be presented using soil FAME data from the Biota Project in the Atlantic forest of Brazil.