Application of Graph Attention Network for Breast Cancer Data and Explanation of Prediction Basis

Breast cancer is one of the most prevalent cancers globally. Applying machine learning meth-ods to predict breast cancer prognosis is essential for developing effective clinical treatment plans. However, predicting models alone does not ensure the reliability of prognostic outcomes. Therefore, reliable predictions require clear explanations of the underlying basis for its predicted results. This study proposes a method employing a machine learning model for predicting cancer prognosis and a method for explaining the basis for the predicted results. In this study, a machine learning method was introduced to predict the prognosis of breast cancer patients using a graph attention network. We used GNNExplainer to obtain a graph containing important genes and their interrelationships as the basis for the prediction. This method enables the visuali-zation and explanation of the rationale behind the predictive outcomes. This study identified 55 genes critical to predicting survival status within a 10-year timeframe. This study proposed a method employing a graph attention network to breast cancer data, and to explain the prediction basis. The experimental results allowed to identify the genes that play a key role in prognosis prediction, demonstrating the effectiveness of this method in interpreting predicted results. This study has shown subgraphs visually and improved the reliability of prognostic predictions by clearly revealing the factors that affect patient prognosis, which helps to develop more effective clinical treatment plans and decision-making.