Data integration in microbial genomics ... - Jacobs University

84 6. Domains of unknown function
performed a least squares, linear fit of chlorophyll data with significance
(P(>R)) determined by permutation (n=1000). To explore
non-linear relationships between chlorophyll concentrations and the
ordination, we visualized generalized additive model (GAM) fits as
smoothed, non-parametric isoclines (Fig. 6.2) with significance determined
by ANOVA [Wood, 2008]. After Virtanen et al., we interpreted
coefficients of determination (R2) as goodness-of-fit measures
for linear vectors (Rv2) and non-parametric surfaces (Rs2). Analyses
were performed in R (http://www.r-project.org). We observed that
these DUF abundances moderately, but significantly, structure GOS
sites along chlorophyll concentration (Rv2 ≈ 0.52, P (> R) ≈ 9.99 × 10 −
4; Rs2 ≈ 0.91, p ≈ 2.00 × 10 − 16). An improved, albeit less significant, fit
(Rv2 ≈ 0.64, P (> R) ≈ 4.00 × 10 − 3; Rs2 ≈ 0.98, p ≈ 5.8 × 10 − 2) and a more
even resolution of sites may be observed when ordinating geographically
localized sample groups such as that along the North American
East Coast (GS002, GS004-8, GS012-14; n=9, plot not shown). The
GAM surface reveals considerable non-linear effects below chlorophyll
concentrations of 2.0 µg kg-1 seawater, where most sites – particularly
from oligotrophic waters – are ordinated. Such effects may rise
from the diverse functions, multi-functionality, and the selective interactions
between elements in biological systems [Kitano, 2002]. The
global coverage of GOS, across numerous ecoregions, may also introduce
unexpected variation. Nonetheless, if these chlorophyll measurements
are understood as a proxy for phytoplankton abundance, these
results tentatively support hypotheses linking the functional community
structure described by these DUFs to the abundance of photoreactive
plankton. This manner of environmental contextualization may
provide useful perspectives on the function of microbial genomic features
in their surrounding ecosystems.
6.5 Conclusion
Ecogenomic datasets promise to deliver valuable insight into the roles
of uncharacterized genes and proteins. The prospects are greater if
future ‘omics’ sampling is performed along clear environmental gra-