Machine learning fundamentally learns from the past experiences (seen data) to make predictions about future (unseen data). Predictions in nature are often uncertain. Microbiome data have unique characteristics, including high-dimensionality, over-dispersion, sparsity and zero-inflation, and heterogeneity. Thus, machine learning involving microbiome data for predicting the outcome of phenotypes is even more uncertain than learning those data from other fields. Machine Learning for Microbiome Statistics poses many challenges for evaluating the prediction performance using appropriate metrics and independent data validation.
This unique book aims to address the challenges of machine learning statistics, emphasize the importance of performance valuation by appropriate metrics and independent data, and describe several important concepts of machine learning statistics, such as feature engineering and overfitting. It comprehensively reviews commonly used and newly developed machine learning models for microbiome research. Specifically, this book provides the step-by-step procedures to perform machine learning of microbiome data, including feature engineering, algorithm selection and optimization, performance evaluation and model testing. It comments the benefits and limitations of using machine learning for microbiome statistics and remarks on the advantages and disadvantages of each machine learning algorithm.
It will be an excellent reference book for students and academics in the field.
- Presents a thorough overview of machine learning algorithms for microbiome statistics.
- Performs step-by-step procedures to perform machine learning of microbiome data, using important supervised learning algorithms, including classical, ensemble learning and tree-based models.
- Describes important concepts of machine learning, including bias and variance tradeoff, accuracy and precision, overfitting and underfitting, model complexity and interpretability, and feature engineering.
- Investigates and applies various cross-validation techniques step-by-step.
- Introduces confusion matrix and its derived measures. Comprehensively describes the properties of F1, Matthews’ correlation coefficient (MCC), area under the receiver operating characteristic curve (AUC-ROC), and area under the precision-recall curve (AUC-PR), as well as discusses their advantages and disadvantages when using them for microbiome data.
- Offers all related R codes and the datasets from the authors’ first-hand microbiome research and publicly available data.
Dr. Yinglin Xia is a Clinical Professor in the Department of Medicine at the University of Illinois Chicago. He has published six books on statistical analysis of microbiome and metabolomics data and more than 180 statistical methodology and research papers in peer-reviewed journals. He serves on the editorial boards of several scientific journals including as an Associate Editor of Gut Microbes and has served as a reviewer for over 100 scientific journals.
Dr. Jun Sun is a tenured Professor of Medicine at the University of Illinois Chicago and an internationally recognized expert on microbiome and human diseases, e.g., vitamin D receptor in inflammation, dysbiosis and intestinal dysfunction in amyotrophic lateral sclerosis (ALS). Her lab is the first to discover that chronic effects and molecular mechanisms of Salmonella infection and risk of colon cancer. Dr. Sun has published over 260 scientific articles in peer-reviewed journals and 10 books on microbiome.
