What Makes A Good Egg? – Ovarian Gene Expression Profiling Via Artificial Intelligence Predicts Egg Quality In Striped Bass
Craig Sullivan, North Carolina State University
3 Feb 2012
Ovary biopsies taken from striped bass before the breeding season were subjected to microarray analysis. The relationship of ovarian gene expression (transcriptome) profiles to the quality of eggs spawned was assessed using artificial neural networks (ANNs) and supervised machine learning. Production of well-formed, mid-blastula stage embryos was the primary measure of egg quality. Collective minor (usually < 0.2-fold) changes in the expression of a limited suite of genes (N=232) representing < 2% of the transcriptome queried explained > 90% of the variance in egg quality, whereas expression of individual gene transcripts explained very little (< 1%). When ovarian transcript profiles from new fish were input to the ANN models, they predicted egg quality with nearly 80% accuracy. Analyses of the inherent correlation of transcripts performed by modulated modularity clustering, followed by the construction of gene relevance networks, indicated that collective dysfunction of the ubiquitin-26S proteasome, COP9 signalosome, and subsequent control of the cell cycle may be the primary cause of embryo developmental incompetence in striped bass. To our knowledge, the combined differences in the expression of discrete clusters of genes involved in these and related processes exhibited the most powerful molecular relationship to egg quality and, hence, reproductive fitness that has ever been described. Our findings emphasize the need to evaluate collective subtle changes in gene expression when exploring the genetic underpinnings of biologically-complex traits, a task to which the application of artificial intelligence appears to be especially well suited. Implications of these findings for aquaculture and fisheries management will be discussed.