| Symptoms and signs (symptoms in brief) are the essential clinical manifestations for individualized diagnosis and treatment in traditional Chinese medicine (TCM). Investigation on the underlying molecular mechanisms (e.g. associated genes and proteins) of symptoms is still an open problem, on which limited studies were conducted. There are two reasons which lead to this problem:(1)The subjectivity and individuality of the diagnosis of syndrome should be taken into account.(2)The absence of the study on molecular mechanisms of TCM four diagnostic information(symptoms), which is the basis of syndrome differentiation and treatment. To explore the molecular mechanisms of symptoms, we developed a network based analytical method for symptom gene prediction by integrating symptom-gene associations, drug side-effects, disease-gene associations and protein-protein interaction network data. We proposed two kinds of methods:the method based on symptom-disease relationship and based on drug side effects.(1)The symptom-disease relationship method, which is based on PRINCE algorithm, has three types of algorithms, namely PRICE TD, PRINCE_ST and PRINCE_SB, and all the three algorithms obtained acceptable performance. Furthermore, the improved algorithm:PRTNCE_SB gets best performance (with AUC0.843and average Ranking Score0.139) and has a10-fold increasing from the compared random selection. Due to the limit size of benchmark data set, we retrieved literature as a supplementary evaluation and confirmed good results. For instance, we find the first10genes in our ranking result, ABCB1, MTHFR, TNF, ACE, CYP3A4, IL6, TP53, TYMS, PTGS2and PTEN, may be all the causal genes of headache by document retrieval.(2) The method based on drug side effects combing with the former method, although has less AUC than the PRINCE_SB, obtained more biological meaningful results with average Ranking Score0.098, which is increased by28.95%comparing with PRINCE_SB. Herein, we investigate the symptom-gene association by integrating different phenotype-genotype association data and propose an integrated network based approach for prioritization of candidate symptom genes. The experimental results got reliable results and confirmed its feasibility and effectiveness. Our study provides a new approach and analysis method to investigation of the molecular mechanisms of TCM diagnosis. |
