Neural Network Based Prediction of Less Side Effect Causing Cancer Drug Targets in the Network of MAPK Pathways

Abstract

Computational side-effect prediction tools assist in rational drug design to decrease the late-stage failure of the drugs. Irrational selection of cancer drug targets in the deregulated MAPK pathways causes side effects. Network centralities and biological features - Degree, Radiality, Eccentricity, Closeness, Bridging, Stress, Pagerank centralities, essentiality, pathway-specific proteins, disease-causing proteins, protein domains are exploited quantitatively. We train an artificial neural network (ANN) with 15 selected features for the binary classification of side effects causing and less side-effect causing drug targets among the non-targeted proteins. Top ranked proteins among the Degree, Eccentricity, betweenness centralities, possessing GO-based molecular function, involved in more than one Biocarta pathways, domain content are prone to cause a number of side effects than other centralities and functional features. We predicted the following 15 less side effect causing cancer drug targets - Shc, Rap 1a, Mos, Tpl-2, PAC1, 4EBP1, GAB1, LAD, MEF2, ZAK, GADD45, TAB2, TAB1, ELK1 and SRF.