Supervised prediction of aging-related genes from a context-specific protein interaction subnetwork

Abstract

Human aging is linked to many prevalent diseases. The aging process is highly influenced by genetic factors. Hence, it is important to identify human aging-related genes. We focus on supervised prediction of such genes. Gene expression-based methods for this purpose study genes in isolation from each other. While protein-protein interaction (PPI) network-based methods for this purpose account for interactions between genes' protein products, current PPI network data are contextunspecific, spanning different biological conditions. Instead, here, we focus on an aging-specific subnetwork of the entire PPI network, obtained by integrating aging-specific gene expression data and PPI network data. The potential of such data integration has been recognized but mostly in the context of cancer. So, we are the first to propose a supervised learning framework for predicting aging-related genes from an aging-specific PPI subnetwork. We find that using an aging-specific subnetwork indeed yields more accurate aging-related gene predictions than using the entire network. Also, predictive methods from our framework that have not previously been used for supervised prediction of aging-related genes outperform existing prominent methods for the same purpose. These results justify the need for our framework.However, cellular processes, including aging, are carried out by genes' protein products interacting with each other in complex ways [20]. So, it is essential to consider interactions between proteins.This is exactly what PPI network-based methods do. They predict a gene as aging-related according to how similar its position (i.e., node representation/embedding/feature) in the PPI network is to the network positions of known aging-related genes [7,8,17,18]. State-of-the-art approaches of this type are UniNet [8] and mBPIs [7]. UniNet's feature concatenates 14 network centrality measures (e.g., degree or betweenness), where centrality of a node captures its importance in the network. The feature of mBPIs is constructed as follows. First, m nodes with the highest degrees in the network are identified. Then, for each node v in the network, v's feature has m dimensions corresponding to the top m highest-degree nodes, where each dimension j of node v's feature indicates whether v interacts with the top m highest-degree node j. A limitation of these and other PPI network-based methods is that the current PPI network data are context-unspecific, meaning that the PPIs span different conditions (cell types, tissues, diseases, environments, patients, etc.) [21]; consequently, the current PPI network data are also static [4]. We argue that it is essential to consider aging-specific context of the entire PPI network, i.e., its aging-specific subnetwork, by integrating aging-specific gene expression data with PPI network data.Indeed, the need for at least some level of data integration in the context of aging is being recognized. There exist approaches for supervised prediction of aging-related genes that extract genes' features from each of...

Improved supervised prediction of aging-related genes via weighted dynamic network analysis

“…Some of our considered features have high dimensions, such as DGDV, GoT, and cSGDV. So, as in our previous work [17,18], for each feature, we consider: (i) the full feature (i.e., no dimensionality reduction), (ii) linear dimensionality reduction via principal component analysis (PCA) that considers as few principal components as needed to account for at least 90% of variation in the data corresponding to the given feature, and (iii)-(viii) nonlinear dimensionality reduction via t-distributed stochastic neighbor embedding (tSNE) under six different perplexity parameters (5,13,21,30,40,50). This totals to 1 + 1 + 6 = 8 considered dimensionality reduction choices.…”

“…• CentraMV, centrality mean and variation [17,18], of a node v measures, for each of the four considered centrality-based features (GDC, ECC, KC, and DegC), two quantities: the mean and variation over v's 37 centrality values. These two quantities for each of the four centrality-based features combined form an 8-dimensional CentraMV node feature.…”

“…First, m highest-degree nodes in the network are found. Then, for a node v, v's feature has m dimensions corresponding to the m nodes, where each dimension d of node v's feature has a value of 1 if v interacts with the top m highest-degree node d, or otherwise it has a value of 0 if v does not interact with d. Just as [17,18], and as suggested in the original mBPIs publication, we use m = 30 in this paper. This results in a 30-dimensional mBPIs node feature.…”

“…We consider the same nine classifiers under the same parameters as in our previous work [17,18], all implemented using a Phython library scikit-learn (version 0.20.3) [37]: Adaptive Boosting (AB), Random Forest (RF), Decision Tree (Dtree), Logistic Regression (LR), Gaussian Naïve Bayes (NB), K-Nearest Neighbors (KNN), Support Vector Machine with a linear kernel (SVM-linear) and a non-linear kernel (radial basis function (SVM-rbf)), and Multilayer Perceptron (MLP); see Supplementary Section S1.…”

“…Supervised analyses from our other, even more recent work [17,18] are directly relevant for this paper. In that work, we performed supervised prediction of aging-related genes from three networks: the dynamic aging-specific subnetwork by Faisal et al [4], a static (still aging-specific) counterpart of the dynamic subnetwork, and the entire (also static) context-unspecific network from which the dynamic aging-specific subnetwork was inferred.…”