This technology is an electronic device that calculates influence weights between genes by training on gene data and gene networks of cancer patients with good and poor prognoses using Generative Adversarial Networks (GAN). It determines the influence index for each gene based on the weight difference between the two groups and selects top-ranking genes as biomarkers for cancer prognosis prediction.
While there have been attempts to predict cancer prognosis based on patient genetic characteristics using AI, there were limitations in constructing models that reflect all genes due to the vast number of human genes. Furthermore, including genes with little impact on prognosis often reduced the accuracy of prediction models.
This technology calculates an influence index by comparing the distribution differences of GAN training weights for each gene data type using Kullback-Leibler divergence, and creates a supervised learning prediction model using only the selected biomarkers. By focusing on a small number of genes closely related to prognosis among the vast pool of genes, prediction accuracy is improved. It can be applied to precision medicine-based cancer prognosis diagnostic software and personalized treatment strategy platforms, providing a quantitative basis for selecting treatment methods for individual patients.
N/A
