Coupling of functional gene diversity and geochemical data from environmental samples

Area of Science:

• Environmental microbiology
• Bioinformatics
• Geochemistry

Background:

• Genomic techniques for environmental microbial analysis often yield small sample sizes.
• Analyzing gene distributions in limited datasets presents challenges for inferring geochemical relationships.

Purpose of the Study:

• To investigate the utility of artificial neural networks (ANNs) for analyzing microbial gene distributions in small sample sets.
• To identify geochemical factors influencing the distribution of nitrite reductase genes (nirS, nirK) and dissimilatory sulfite reductase genes (dsrAB1, dsrAB2).

Main Methods:

• Application of ANNs to analyze distributions of nirS, nirK, dsrAB1, and dsrAB2 genes.
• Utilized data reduction, cross-validation, weight decay, and importance analysis for model development and interpretation.
• Evaluated model generalization error to assess the reliability of inferred relationships.

Main Results:

• ANN models successfully inferred relationships between geochemistry and gene distributions, particularly for unusual dsrAB-related genes (dsrAB2).
• Uranium and sulfate concentrations were strongly associated with dsrAB2 gene distribution.
• pH, nickel, non-purgeable organic carbon, and total organic carbon correlated with other gene groups (nirS, nirK, dsrAB1).
• Models for nirS, nirK, and dsrAB1 showed poorer generalization compared to dsrAB2 models.

Conclusions:

• ANNs are valuable tools for analyzing microbial gene distributions in small environmental datasets.
• Geochemical factors significantly influence microbial gene abundances, with specific relationships varying by gene type.
• Validation approaches are crucial when developing models from small sample sizes to avoid high generalization errors.