TY - JOUR T1 - Reproducibility Of Parameter Learning With Missing Observations in Naive Wnt Bayesian Network Trained on Normal/Adenomas Samples and Doxycycline Treated LS174T Cell Lines JF - bioRxiv DO - 10.1101/014076 SP - 014076 AU - Shriprakash Sinha Y1 - 2015/01/01 UR - http://biorxiv.org/content/early/2015/01/22/014076.abstract N2 - Insight, Innovation and Integration Doxycycline, a derivative of tetracycline, induces gene expression via reversible transcriptional activation. Levels of /3-catenin and other intra/extracellular genetic factors have been influenced in colorectal cancer cell lines, which make doxycycline a potential candidate for cancer chemotherapy. With the aim to build better computational models that show good prediction on test datasets, doxycycline treated cell lines might provide best training samples. This work tests the reproducibility of parameter learning and predictions based on the estimated parameters, using the Naive Bayesian Networks for Wnt pathway in case of missing observations for different nodes. The in silico experiments show the efficacy of causal models as one of the emerging diagnostic tools in development of targeted cancer therapy.Recent efforts in predicting Wnt signaling activation via inference methods have helped in developing diagnostic models for therapeutic drug targeting. In this manuscript the reproducibility of parameter learning with missing observations in a Bayesian Network and its effect on prediction results for Wnt signaling activation is tested, while training the networks on doxycycline treated LS174T cell lines as well as normal and adenomas samples. This is done in order to check the effectiveness of using Bayesian Network as a tool for modeling Wnt pathway when certain observations are missing. Experimental analysis suggest that prediction results are reproducible with negligible deviations. Anomalies in estimated parameters are accounted for due to the Bayesian Network model. Also, an interesting case regarding usage of hypothesis testing came up while proving the statistical significance of different design setups of the BN model which was trained on the same data. It was found that hypothesis testing may not be the correct way to check the significance between design setups for the aforementioned case, especially when the structure of the model is same. Finally, in comparison to the biologically inspired models, the naive bayesian model may give accurate results but this accuracy comes at the cost of loss of crucial biological knowledge which might help reveal hidden relations among intra/extracellular factors affecting the Wnt pathway. ER -