RDPA:A Phenotype-disease Prediction Algorithm Based On Phenotypic Text Data

With the development of AI technology,Artificial Intelligence(AI)technology in medical field has become the most popular topic.But in the field of rare diseases,the data of rare diseases are relatively limited and the pathogenesis is diversified.No effective and reliable algorithm is available to predict the potential rare diseases and assist the diagnosis of rare diseases.Here,we propose an improved prediction algorithm for rare diseases,called Rare Disease-Phenotype Association Prediction Algorithm(RDPA).The algorithm is consisted with two parts: the first part is to improve the classic TF-IDF algorithm to deal with rare disease data and achieve the correlation between phenotype and disease;the second part is to integrate the correlation generated by the previous step.So that RDPA algorithm can predict the incidence of multi-phenotypic diseases.By comparing RDPA algorithm with other commonly used prediction algorithms(TF-CRF,TF-IDF-CHI and TF-IDF),we find that RDPA algorithm has better performance over other prediction algorithms,and achieves better accuracy and stability.In addition,based on the performance of the RDPA algorithm on the training set and independent test set,we propose recommentations for using the algorithm.In conclusion,the RDPA algorithm is an effective prediction algorithm for rare disease,and has a broad application space in both basic researches and clinical field.