Abstract
BACKGROUND: As a marker of Helicobacter pylori, Cytotoxin-associated gene A (cagA) has been revealed to be the major virulence factor causing gastroduodenal diseases. However, the molecular mechanisms that underlie the development of different gastroduodenal diseases caused by cagA-positive H. pylori infection remain unknown. Current studies are limited to the evaluation of the correlation between diseases and the number of Glu-Pro-Ile-Tyr-Ala (EPIYA) motifs in the CagA strain. To further understand the relationship between CagA sequence and its virulence to gastric cancer, we proposed a systematic entropy-based approach to identify the cancer-related residues in the intervening regions of CagA and employed a supervised machine learning method for cancer and non-cancer cases classification. METHODOLOGY: An entropy-based calculation was used to detect key residues of CagA intervening sequences as the gastric cancer biomarker. For each residue, both combinatorial entropy and background entropy were calculated, and the entropy difference was used as the criterion for feature residue selection. The feature values were then fed into Support Vector Machines (SVM) with the Radial Basis Function (RBF) kernel, and two parameters were tuned to obtain the optimal F value by using grid search. Two other popular sequence classification methods, the BLAST and HMMER, were also applied to the same data for comparison. CONCLUSION: Our method achieved 76% and 71% classification accuracy for Western and East Asian subtypes, respectively, which performed significantly better than BLAST and HMMER. This research indicates that small variations of amino acids in those important residues might lead to the virulence variance of CagA strains resulting in different gastroduodenal diseases. This study provides not only a useful tool to predict the correlation between the novel CagA strain and diseases, but also a general new framework for detecting biological sequence biomarkers in population studies.