Estimation of gynecological cancer networks via target proteins and risk factors

Bahçıvancı, Başak
Purutçuoğlu Gazi, Vilda
Purutçuoğlu, Eda
Abstract—The construction of biological networks has certain challenges due to its high dimension, sparse structure and very limited number of observations. Thus, specific modeling approaches have been suggested to deal with these problems such as Gaussian graphical model, loop-based multivariate adaptive regression splines (MARS) with/without interaction effects and Gaussian copula graphical model. From previous analyses via these methods, it has been shown that they can successfully estimate the systems with comparative accuracies. Hereby, in this study, as the novelty we use all these complex mathematical models in inference of gynecological cancer networks whose target genes are gathered from biological literature. The observations for these target genes are collected from the ArrayExpress database with other associated risk factors such as stage of the cancers which are denoted by categorical variables. Then, under different dimensions of systems, sample sizes and measurement types, we compare the performance of all models with the criteria of accuracy and F-measure. From the results, we observe that the suggested models can successfully estimate the real cancer systems under different conditions and are promising approaches to describe the complexity in biological networks.
Citation Formats
B. Bahçıvancı, V. Purutçuoğlu Gazi, and E. Purutçuoğlu, “Estimation of gynecological cancer networks via target proteins and risk factors,” 2018, Accessed: 00, 2021. [Online]. Available: