Key symptoms selection for two major syndromes diagnosis of Chinese medicine in chronic hepatitis B / 中国结合医学杂志

<p><b>OBJECTIVE</b>To identify key symptoms of two major syndromes in chronic hepatitis B (CHB), which can be the clinical evidence for Chinese medicine (CM) doctors to make decisions.</p><p><b>METHODS</b>Standardization scales on diagnosis for CHB in CM were designed including physical symptoms, tongue and pulse appearance. The total of 695 CHB cases with dampness-heat (DH) syndrome or Pi (Spleen) deficiency (SD) syndrome were collected for feature selection and modeling, another 275 CHB patients were collected in different locations for validation. Key symptoms were selected based on modified information gain (IG), and 5 classifiers were applied to assist with models training and validation. Classification accuracy and area under receiver operating characteristic curves (AUC) were evaluated.</p><p><b>RESULTS</b>(1) Thirteen DH syndrome key symptoms and 13 SD syndrome key symptoms were selected from original 125 symptoms; (2) The key symptoms could achieve similar or better diagnostic accuracy than the original total symptoms; (3) In the validation phase, the key symptoms could identify syndromes effectively, especially in DH syndrome, which average prediction accuracy on 5 classifiers could achieve 0.864 with the average AUC 0.772.</p><p><b>CONCLUSION</b>The selected key symptoms could be simple DH and SD syndromes diagnostic elements applied in clinical directly. (Registration N0.: ChiCTR-DCC-10000759).</p>

Efficient Mining of Interesting Patterns in Large Biological Sequences

Pattern discovery in biological sequences (e.g., DNA sequences) is one of the most challenging tasks in computational biology and bioinformatics. So far, in most approaches, the number of occurrences is a major measure of determining whether a pattern is interesting or not. In computational biology, however, a pattern that is not frequent may still be considered very informative if its actual support frequency exceeds the prior expectation by a large margin. In this paper, we propose a new interesting measure that can provide meaningful biological information. We also propose an efficient index-based method for mining such interesting patterns. Experimental results show that our approach can find interesting patterns within an acceptable computation time.