Technical Aspects of Nominal Partitions on Accuracy of Data Mining Classification of Intestinal Microbiota - Comparison between 7 Restriction Enzymes

Characteristics of Gut Microbiota in Patients With Diabetes Determined by Data Mining Analysis of Terminal Restriction Fragment Length Polymorphisms

Background

This study was performed to clarify whether gut microbiota obtained from fecal samples could identify the type of diabetes in patients of each gender by using a combination of terminal restriction fragment length polymorphism (T-RFLP) analysis and data mining.

Methods

A cross-sectional study was performed at three centers. Fecal samples were collected from 12 Japanese patients with type 1 diabetes mellitus (T1D), 18 patients with type 2 diabetes mellitus (T2D), and 31 subjects without diabetes mellitus (non-DM). Amplification of fecal 16S rRNA was carried out. After digestion of the amplification products with restriction enzymes (AluI, BslI, HaeIII, and MspI), terminal restriction fragments (T-RFs) of DNA were detected. A data mining algorithm (classification and regression tree (CART) modeling system) provides a decision tree that classifies subjects into various groups according to pre-assigned characteristics.

Results

Among men, the error rate was 2.4% with MspI, while error rates were 0.0% with other restriction enzymes. Among women, the error rate was 0.0% with all restriction enzymes. The operational taxonomic units (OTUs) incorporated into the decision tree differed between men and women.

Conclusions

We were able to classify the 16SrRNA gene amplification products obtained from fecal samples of T1D patients, T2D patients, and non-DM subjects with a high level of precision by combining T-RFLP analysis and data mining. Specific gut microbiota patterns were found for T1D and T2D patients, as well as a sex difference of the patterns.

Numerical analyses of intestinal microbiota by data mining

The human intestinal microbiota has a close relationship with health control and causes of diseases, and a vast number of scientific papers on this topic have been published recently. Some progress has been made in identifying the causes or species of related microbiota, and successful results of data mining are reviewed here. Humans who are targets of a disease have their own individual characteristics, including various types of noise because of their individual life style and history. The quantitatively dominant bacterial species are not always deeply connected with a target disease. Instead of conventional simple comparisons of the statistical record, here the Gini-coefficient (i.e., evaluation of the uniformity of a group) was applied to minimize the effects of various types of noise in the data. A series of results were reviewed comparatively for normal daily life, disease and technical aspects of data mining. Some representative cases (i.e., heavy smokers, Crohn’s disease, coronary artery disease and prediction accuracy of diagnosis) are discussed in detail. In conclusion, data mining is useful for general diagnostic applications with reasonable cost and reproducibility.

Characterization of gut microbiota profiles in coronary artery disease patients using data mining analysis of terminal restriction fragment length polymorphism: gut microbiota could be a diagnostic marker of coronary artery disease

The association between atherosclerosis and gut microbiota has been attracting increased attention. We previously demonstrated a possible link between gut microbiota and coronary artery disease. Our aim of this study was to clarify the gut microbiota profiles in coronary artery disease patients using data mining analysis of terminal restriction fragment length polymorphism (T-RFLP). This study included 39 coronary artery disease (CAD) patients and 30 age- and sex- matched no-CAD controls (Ctrls) with coronary risk factors. Bacterial DNA was extracted from their fecal samples and analyzed by T-RFLP and data mining analysis using the classification and regression algorithm. Five additional CAD patients were newly recruited to confirm the reliability of this analysis. Data mining analysis could divide the composition of gut microbiota into 2 characteristic nodes. The CAD group was classified into 4 CAD pattern nodes (35/39 = 90 %), while the Ctrl group was classified into 3 Ctrl pattern nodes (28/30 = 93 %). Five additional CAD samples were applied to the same dividing model, which could validate the accuracy to predict the risk of CAD by data mining analysis. We could demonstrate that operational taxonomic unit 853 (OTU853), OTU657, and OTU990 were determined important both by the data mining method and by the usual statistical comparison. We classified the gut microbiota profiles in coronary artery disease patients using data mining analysis of T-RFLP data and demonstrated the possibility that gut microbiota is a diagnostic marker of suffering from CAD.

Use of T-RFLP and seven restriction enzymes to compare the faecal microbiota of obese and lean Japanese healthy men

The composition of the intestinal microbiota of 92 healthy Japanese men was measured following consumption of identical meals for 3 days; terminal restriction fragment length polymorphisms were then used to analyse the DNA content of their faeces. The obtained operational taxonomic units (OTUs) were further analysed using seven restriction enzymes: 516f-BslI and -HaeIII, 27f-MspI and -AluI, and 35f-HhaI, -MspI and -AluI. Subjects were classified by their body mass index (BMI) as lean (<18.5) or obese (>25.0). OTUs were then analysed using data mining software. Pearson correlation coefficients on data mining results indicated only a weak relationship between BMI and OTU diversity. Specific OTUs attributed to lean and obese subjects were further examined by data mining with six groups of enzymes and closely related accession numbers for lean and obese subjects were successfully narrowed down. 16S rRNA sequences showed Bacillus spp., Erysipelothrix spp. and Holdemania spp. to be present among 30 bacterial candidates related to the lean group. Fifteen candidates were classified Firmicutes, one was classified as Chloroflexi, and the others were not classified. 45 Microbacteriaceae, 11 uncultured Actinobacterium, and 3 other families were present among the 119 candidate OTUs related to obesity. We conclude that the presence of Firmicutes and Actinobacteria may be related to the BMI of the subject.