Prior Biological Knowledge And Epigenetic Information Enhances Prediction Accuracy Of Bayesian Wnt Pathway

TL;DR

This study develops Bayesian network models incorporating biological and epigenetic knowledge to improve cancer prediction accuracy from Wnt pathway data, demonstrating significant enhancements in test sample classification.

Contribution

It introduces simple static Bayesian network models that integrate prior biological and epigenetic information, advancing cancer prediction and understanding of Wnt signaling.

Findings

Increased prediction accuracy with epigenetic data inclusion

Significant ROC and AUC improvements in tumor vs. normal classification

Statistical validation confirms model effectiveness

Abstract

Computational modeling of Wnt signaling pathway has gained prominence for its use as computer aided diagnostic tool to develop therapeutic cancer target drugs and predict of test samples as cancerous and non cancerous. This manuscript focuses on development of simple static bayesian network models of varying complexity that encompasses prior partially available biological knowledge about intra and extra cellular factors affecting the Wnt pathway and incorporates epigenetic information like methylation and histone modification of a few genes known to have inhibitory affect on Wnt pathway. It might be expected that such models not only increase cancer prediction accuracies and also form basis for understanding Wnt signaling activity in different states of tumorigenesis. Initial results in human colorectal cancer cases indicate that incorporation of epigenetic information increases prediction accuracy of test samples as being tumorous or normal. Receiver Operator Curves (ROC) and their respective area under the curve (AUC) measurements, obtained from predictions of state of test sample and corresponding predictions of the state of activation of transcription complex of the Wnt pathway for the test sample, indicate that there is significant difference between the Wnt pathway being on (off) and its association with the sample being tumorous (normal). Two sample Kolmogorov-Smirnov test confirm the statistical deviation between the distributions of these predictions. At a preliminary stage, use of these models may help in understanding the yet unknown effect of certain factors like DKK2, DKK3-1 and SFRP-2/3/5 on {\beta}-catenin transcription complex.