Combining affinity propagation clustering and mutual information network to investigate key genes in fibroid

Abstract

The aim of the present study was to investigate key genes in fibroids based on the multiple affinity propogation-Krzanowski and Lai (mAP-KL) method, which included the maxT multiple hypothesis, Krzanowski and Lai (KL) cluster quality index, affinity propagation (AP) clustering algorithm and mutual information network (MIN) constructed by the context likelihood of relatedness (CLR) algorithm. In order to achieve this goal, mAP-KL was initially implemented to investigate exemplars in fibroid, and the maxT function was employed to rank the genes of training and test sets, and the top 200 genes were obtained for further study. In addition, the KL cluster index was applied to determine the quantity of clusters and the AP clustering algorithm was conducted to identify the clusters and their exemplars. Subsequently, the support vector machine (SVM) model was selected to evaluate the classification performance of mAP-KL. Finally, topological properties (degree, closeness, betweenness and transitivity) of exemplars in MIN constructed according to the CLR algorithm were assessed to investigate key genes in fibroid. The SVM model validated that the classification between normal controls and fibroid patients by mAP-KL had a good performance. A total of 9 clusters and exemplars were identified based on mAP-KL, which were comprised of CALCOCO2, COL4A2, COPS8, SNCG, PA2G4, C17orf70, MARK3, BTNL3 and TBC1D13. By accessing the topological analysis for exemplars in MIN, SNCG and COL4A2 were identified as the two most significant genes of four types of methods, and they were denoted as key genes in the progress of fibroid. In conclusion, two key genes (SNCG and COL4A2) and 9 exemplars were successfully investigated, and these may be potential biomarkers for the detection and treatment of fibroid.

Unified feature association networks through integration of transcriptomic and proteomic data

“…It is also important to point out that the dataset we have here, containing nearly 100 datapoints is actually somewhat small compared to other studies using network inference. While there have been several studies that use datasets of a size similar to ours or smaller [54–56] most studies contain many hundreds of samples. It will be of interest to determine how the choice of network inference method, particularly the use of GENIE3, may be affected by very large sample sizes.…”