Drug Efficacy Prediction in Tumors Based on LDA Model

In order to predict the effect of tumor drug treatment in advance, this paper proposes a new method for personalized drug efficacy prediction based on Latent Dirichlet Allocation (LDA) model. Primarily, we use the LDA topic model to encode clinical symptoms in the big data of patients' clinical text, and mine the hidden clinical features and their association patterns with treatment outcomes, and then obtain the efficacy prediction model by training classifiers. The obtained model classifies patients into being responsive or non-responsive to a drug for implementing personalized tumor drug efficacy prediction. The experimental validation results on the collected clinical samples show that the proposed method combined with the Support Vector Machine (SVM) classifier achieved a drug efficacy prediction accuracy of 78%, which is higher than the 62% of the PLSA model.