Biology subjects

Ruffin, M. T.

Publications and source records attributed to Ruffin, M. T..

5 recordsLinked to original sources

bioRxiv2019

Effective application of machine learning to microbiome-based classification problems

Machine learning (ML) modeling of the human microbiome has the potential to identify microbial biomarkers and aid in the diagnosis of many diseases such as inflammatory bowel disease, diabetes, and colorectal cancer. Progress has been made towards developing ML models that predict health outcomes using bacterial abundances, but inconsistent adoption of training and evaluation methods call the validity of these models into question. Furthermore, there appears to be a preference by many researchers to favor increased model complexity over interpretability. To overcome these challenges, we trained seven models that used fecal 16S rRNA sequence data to predict the presence of colonic screen relevant neoplasias (SRNs; n=490 patients, 261 controls and 229 cases). We developed a reusable open-source pipeline to train, validate, and interpret ML models. To show the effect of model selection, we assessed the predictive performance, interpretability, and training time of L2-regularized logistic regression, L1 and L2-regularized support vector machines (SVM) with linear and radial basis function kernels, decision trees, random forest, and gradient boosted trees (XGBoost). The random forest model performed best at detecting SRNs with an AUROC of 0.695 [IQR 0.651-0.739] but was slow to train (83.2 h) and not inherently interpretable. Despite its simplicity, L2-regularized logistic regression followed random forest in predictive performance with an AUROC of 0.680 [IQR 0.625-0.735], trained faster (12 min), and was inherently interpretable. Our analysis highlights the importance of choosing an ML approach based on the goal of the study, as the choice will inform expectations of performance and interpretability. ImportanceDiagnosing diseases using machine learning (ML) is rapidly being adopted in microbiome studies. However, the estimated performance associated with these models is likely over-optimistic. Moreover, there is a trend towards using black box models without a discussion of the difficulty of interpreting such models when trying to identify microbial biomarkers of disease. This work represents a step towards developing more reproducible ML practices in applying ML to microbiome research. We implement a rigorous pipeline and emphasize the importance of selecting ML models that reflect the goal of the study. These concepts are not particular to the study of human health but can also be applied to environmental microbiology studies.

microbiology↗

bioRxiv2019

Fecal short-chain fatty acids are not predictive of colonic tumor status and cannot be predicted based on bacterial community structure

Colonic bacterial populations are thought to have a role in the development of colorectal cancer with some protecting against inflammation and others exacerbating inflammation. Short-chain fatty acids (SCFAs) have been shown to have anti-inflammatory properties and are produced in large quantities by colonic bacteria which produce SCFAs by fermenting fiber. We assessed whether there was an association between fecal SCFA concentrations and the presence of colonic adenomas or carcinomas in a cohort of individuals using 16S rRNA gene and metagenomic shotgun sequence data. We measured the fecal concentrations of acetate, propionate, and butyrate within the cohort and found that there were no significant associations between SCFA concentration and tumor status. When we incorporated these concentrations into random forest classification models trained to differentiate between people with normal colons and those with adenomas or carcinomas, we found that they did not significantly improve the ability of 16S rRNA gene or metagenomic gene sequence-based models to classify individuals. Finally, we generated random forest regression models trained to predict the concentration of each SCFA based on 16S rRNA gene or metagenomic gene sequence data from the same samples. These models performed poorly and were able to explain at most 14% of the observed variation in the SCFA concentrations. These results support the broader epidemiological data that questions the value of fiber consumption for reducing the risks of colorectal cancer. Although other bacterial metabolites may serve as biomarkers to detect adenomas or carcinomas, fecal SCFA concentrations have limited predictive power.\n\nImportanceConsidering colorectal cancer is the third leading cancer-related cause of death within the United States, it is important to detect colorectal tumors early and to prevent the formation of tumors. Short-chain fatty acids (SCFAs) are often used as a surrogate for measuring gut health and for being anti-carcinogenic because of their anti-inflammatory properties. We evaluated the fecal SCFA concentration of a cohort of individuals with varying colonic tumor burden who were previously analyzed to identify microbiome-based biomarkers of tumors. We were unable to find an association between SCFA concentration and tumor burden or use SCFAs to improve our microbiome-based models of classifying people based on their tumor status. Furthermore, we were unable to find an association between the fecal community structure and SCFA concentrations. Our results indicate that the association between fecal SCFAs, the gut microbiome, and tumor burden is weak.

microbiology↗

bioRxiv2017

Spatial variation of the native colon microbiota in healthy adults

The microbiome has been implicated in the development of colorectal cancer (CRC) and inflammatory bowel diseases (IBD). The specific traits of these diseases vary along the axis of the digestive tract. Further, variation in the structure of the gut microbiota has been associated with both diseases. Here we profiled the microbiota of the healthy proximal and distal mucosa and lumen to better understand how bacterial populations vary along the colon. We used a two-colonoscope approach to sample proximal and distal mucosal and luminal contents from the colons of 20 healthy subjects that had not undergone any bowel preparation procedure. The biopsies and home-collected stool were subjected to 16S rRNA gene sequencing and Random Forest classification models were built using taxa abundance and location to identify microbiota specific to each site. The right mucosa and lumen had the most similar community structures of the five sites we considered from each subject. The distal mucosa had higher relative abundance of Finegoldia, Murdochiella, Peptoniphilus, Porphyromonas and Anaerococcus. The proximal mucosa had more of the genera Enterobacteriaceae, Bacteroides and Pseudomonas. The classification model performed well when classifying mucosal samples into proximal or distal sides (AUC = 0.808). Separating proximal and distal luminal samples proved more challenging (AUC = 0.599) and specific microbiota that differentiated the two were hard to identify. By sampling the unprepped colon, we identified distinct bacterial populations native to the proximal and distal sides. Further investigation of these bacteria may elucidate if and how these groups contribute to different disease processes on their respective sides of the colon.

microbiology↗

bioRxiv2017

Viral and Bacterial Communities of Colorectal Cancer

Human viruses (those that infect human cells) have been associated with many cancers, largely due to their mutagenic and functionally manipulative abilities. Despite this, cancer microbiome studies have almost exclusively focused on bacteria instead of viruses. We began evaluating the cancer virome by focusing on colorectal cancer, a primary cause of morbidity and mortality throughout the world, and a cancer linked to altered colonic bacterial community compositions but with an unknown association with the gut virome. We used 16S rRNA gene, whole shotgun metagenomic, and purified virus metagenomic sequencing of stool to evaluate the differences in human colorectal cancer virus and bacterial community composition. Through random forest modeling we identified differences in the healthy and colorectal cancer virome. The cancer-associated virome consisted primarily of temperate bacteriophages that were also predicted to be bacteria-virus community network hubs. These results provide foundational evidence that bacteriophage communities are associated with colorectal cancer and potentially impact cancer progression by altering the bacterial host communities.

microbiology↗

bioRxiv2017

The effect of treatment on the microbiota of patients diagnosed with colonic lesions

BackgroundColorectal cancer is a worldwide health problem. Despite growing evidence that members of the gut microbiota can drive tumorigenesis, little is known about what happens to it after treatment for an adenoma or carcinoma. This study tested the hypothesis that treatment for adenoma or carcinoma alters the abundance of bacterial populations associated with disease to those associated with a normal colon. We tested this hypothesis by sequencing the 16S rRNA genes in the feces of 67 individuals before and after treatment for adenoma (N = 22), advanced adenoma (N = 19), and carcinoma (N = 26).\n\nResultsThere were small changes to the bacterial community associated with adenoma or advanced adenoma and large changes associated with carcinoma. The communities from patients with carcinomas changed significantly more than those with adenoma following treatment (P-value < 0.001). Although treatment was associated with intrapersonal changes, the change in the abundance of individual OTUs in response to treatment was not consistent within diagnosis groups (P-value > 0.05). Because the distribution of OTUs across patients and diagnosis groups was irregular, we used the Random Forest machine learning algorithm to identify groups of OTUs that could be used to classify pre and post-treatment samples for each of the diagnosis groups. Although the three models could differentiate between the pre and post-treatment samples, there was little overlap between the OTUs that were indicative of each treatment. To determine whether individuals who underwent treatment were more likely to have OTUs associated with normal colons we used a larger cohort that contained individuals with normal colons and those with adenomas, advanced adenomas, and carcinomas. We again built Random Forest models and measured the change in the positive probability of having one of the three diagnoses to assess whether the post-treatment samples received the same classification as the pre-treatment samples. Samples from patients who had carcinomas changed towards a microbial milieu that resembles the normal colon after treatment (P-value < 0.05). Finally, we were unable to detect any significant differences in the microbiota of individuals treated with surgery alone and those treated with chemotherapy or chemotherapy and radiation (P-value > 0.05).\n\nConclusionsBy better understanding the response of the microbiota to treatment for adenomas and carcinomas, it is likely that biomarkers will eventually be validated that can be used to quantify the risk of recurrence and the likelihood of survival. Although it was difficult to identify significant differences between pre and post-treatment samples from patients with adenoma and advanced adenoma, this was not the case for carcinomas. Not only were there large changes in pre versus post-treatment samples for those with carcinoma, but these changes were towards a more normal microbiota.

microbiology↗