Application of Diagnostic Indicators for Biological Test Validation in Assisted Reproductive Technology
In the field of medical reproductive research, the selection of embryos with the best potential for implantation is the main challenge for biologists. Several studies suggest that genes involved in the Oocyte-Cumulus Cell crosstalk could represent candidate genebiomarkers for selecting embryos with the highest implantation potential.Therefore, the principle objective of this study is to verify the transcriptomic experimental data from 21 biomarker genes by RT-qPCR (Real-Time quantitative Polymerase Chain Reaction) of 102 embryonic/cumulus cell samples from patients undergoing in vitro fertilization. Since variability (noises) from various sources (biological, technical, etc.) was observed,there is a reasonable doubt about the capability of these transcriptomic data to provide a reliable and robust pregnancy predictive model. So, our goal is to verify if the genomic signature could be used as biomarker. If so, one can stipulate that the transcriptome is predictable and could generate a reliable mathematical model. Stochastic modelling is based on the Multiple Logistic Regression (MLR) which is bimodal and therefore binary, seems adequate to give a conclusion regarding this genomic signature capacity to predict the absence or presence of Pregnancy (Pr) event. In this work, the observed event will be represented by a dependent random vector Y that takes the value 1 if pregnancy occurs and 0 if not. The prediction value of this vector also depends on the noise (ε) induced by the variabilities mentioned above. Bio-stochastic tools such as the ROC (Receiver Operating Characteristic measure) curve and its AUC (Area Under the ROC Curve), the probabilistic likelihood indicators, the Odds Ratio (OR), and finally the Youden Index (YI), appear as a simple and an effective biological decision tools to verify the validity of this genomic signature as biomarker to predict pregnancy (Pr). The analysis of the bio-statistical indicator results indicate that the obtained predictive model is non-discriminant, suggesting a bias in the transcriptomic data.