ulpd: an unsupervised learning model to identify party and date hubs

It has been claimed that protein-protein interaction (ppi) networks are scale free that contain a few hubs with ability to bind multiple proteins. hubs are classified as party and date hubs. party hubs generally bind different proteins in specific module simultaneously, while date hubs interact with multiple proteins in different modules at different times and locations. generally, they have been divided into two classes based on the average pear- son correlation coefficient (avpcc) of expression over all partners or their functions. in this study, we propose a two-step algorithm to classify party and date hubs based on their topological features of ppi network. in the first step, we calculate some topological features for each hub vertex in ppi network. in the second step, we apply an unsupervised learning model to calculate laplacian score for each feature. the laplace value for each hub vertex are considered based on laplacian scores. finally, the hub vertices are classified into two classes date hubs and party hubs with respect to laplace values. we evaluate our method on reference hubs based on the avpcc on ppi network. we show that the combination of topological features based on ulpd can improve the performance of each topological feature. finally, we investigate the performance of our method for human dataset and analyze two types of hubs as drug targets for covid-19.