DCGG: drug combination prediction using GNN and GAE

Recent findings show that drug combination therapy can increase efficacy, decrease drug resistance, and reduce drug side effects. Due to the enormous number of possibilities in the selection of drugs, it is clinically impossible to screen all available combinations. Fortunately, artificial intelligence has opened up new perspectives for solving this problem by applying computationally intensive operations to predict drug combinations with high potential efficacy. These computational methods can be extremely resourceful for doctors and medical researchers to select drug combinations for the treatment of simple and complex diseases more cleverly and efficiently. In this paper, we propose an innovative solution for drug combination prediction called the DCGG method, in which the combination of node2vec, word2vec, indication, side effect, drug finger print, and drug targets is exploited for more enhanced prediction. DCGG is a combination of multiple Graph Auto Encoder (GAE) models that use Graph Neural Network (GNN) to prioritize potential novel, efficacious combination therapies. The comparison of DCGG with eight of the previous state-of-the-art models indicates the superiority of DCGG, which outperforms them by an average of 5% w.r.t AUC score (AUC = 0.974). Also, it is important to note that our method is used for a wide variety of drugs in contrast to many of the previous studies in this area. In addition to numeral evaluation, we constructed a graph of newly predicted drug combinations that are biologically interpreted with interesting patterns. We successfully found drug combinations that were not available in DCDB, but are mentioned and discussed to be efficacious in recent medical papers. Overall, the results indicate that DCGG provides a promising tool for predicting drug pairs that are most likely to have combinatorial efficacy.