ABSTRACT
Gene expression signatures of toxicity and clinical response benefit
both safety assessment and clinical practice. However, gene signa-
tures’ predictive performance varies greatly between studies and
tends to diminish in repetitive studies and large data sets. Also,
difficulties in understanding the association of the signatures gene
content to the predicted endpoints have limited their application.
The recently completed Food and Drug Administration (FDA)-led
international microarray quality control II (MAQCII) project gen-
erated 262 signatures for 10 clinical and 3 toxicological endpoints
from six gene expression data sets. We conducted a comprehensive
functional analysis of these signatures and their nonredundant
unions using ontology enrichment, biological network building, and
interactome connectivity analyses. The analyses revealed different
functional correlations between the genes within and between
the signatures, although not at the level of commonly encoded
functional groups such as pathways. Different signatures for a given
endpoint were more similar at the level of biological entities and
transcriptional control than at the gene level. Signatures tended
to be enriched in function and pathway in an endpoint-and
model-specific manner and showed a topological bias for incoming
interactions. Importantly, the level of biological similarity between
different signatures correlated positively with the accuracy of the
signature predictions. These findings have implications for the
design, understanding, and application of predictive genomics.