SELF-SUPERVISED GRAPH NEURAL NETWORKS FOR PREDICTING SIDE EFFECTS FROM DRUG-DRUG INTERACTIONS

Because of their effects on mortality, morbidity, and healthcare expenses, adverse drug reactions (ADRs) brought on by drug-drug interactions are a global public health concern that requires study. The ever-increasing complexity of treatments and the ageing of the population in many areas have presented significant hurdles for the healthcare industry. There is currently no established technique for identifying these negative drug interactions until patients report them after the medication is put on the market. Furthermore, a number of studies demonstrate how difficult it is to identify these uncommon occurrences in clinical trials conducted prior to the drug's distribution. Therefore, there is an urgent need for a trustworthy and effective method to anticipate these adverse effects prior to the drug's entry into the market. We created an efficient framework to describe drug-drug interactions by using the spatial and physical characteristics of pharmaceuticals by expressing them as molecular graphs, utilising the power of Graph Neural Networks and the knowledge representation capabilities of self-supervised learning. We created a method that mimics the kinetics of a chemical reaction using this method. We attain state-of-the-art results by achieving a precision of 75% and an accuracy of 90% on the test dataset after training and testing our strategy on the Therapeutic Data Commons (TDC) Two SIDES Polypharmacy Dataset. In order to test our technique on the drug-drug interaction domain, we also conduct a case study on the DrugBank dataset and compare our findings on the interaction type prediction job. We get outstanding results with precision, F1, and accuracy of 99%. Our investigation and experimental methods provide a foundation for future studies on drug-drug interaction-based side-effect prediction and the use of graph neural networks in molecular biology.