Microbiome Data Analysis Workshop 2024

We shall kick-off the workshop with some basics, starting from raw sequence reads to the generation of an ASV (Amplicon Sequence Variant) or OTU (Operational Taxonomic Unit) table, followed by data visualization, and finally differential abundance (DA) analysis. The workshop is designed to guide participants through each stage of microbiome data analysis, providing both theoretical background and practical hands-on experience. Therefore, each session on the first day will consist of two parts: a theoretical introduction to key concepts and methodologies, followed by an interactive hands-on session.

Program

09:00 – 09:30 Registration and welcome coffee

09:30 – 12:30 Profiling microbial communities with amplicon sequencing by Stijn Wittouck, Department of Bioscience Engineering, University of Antwerp, Belgium

Abstract
In this lecture, we will learn the core concepts behind the taxonomic profiling of microbial communities, with a focus on amplicon sequencing. First, we will position amplicon sequencing in the field of microbial sequencing techniques. Next, we will go over the mains steps of the technique: DNA extraction, targeted PCR, sequencing, clustering/denoising and classification. We will then discuss the 16S rRNA gene and the reasons why it is often used as amplicon region. Next, we will go into clustering vs denoising, the conceptual differences between OTUs and ASVs and the biological meaning of both. Finally, we will discuss the classification of amplicon reads: the pro’s and cons of different reference databases, the workings of the RDP classifier and the differences with metagenome read classifiers. Throughout the lecture, we will practice key steps in R, using the DADA2 package.

12:30 – 13:30 Lunch

13:30 – 15:30 Visualising the microbiome by Thies Gehrmann, Department of Bioscience Engineering, University of Antwerp, Belgium

Abstract
Microbiome composition is a high dimensional measurement modality.
Naturally, this renders visualization, a crucial tool for understanding the structure of the microbiome manifold, challenging.
Additionally, microbiome measurements are compositional (meaning that the proportions sum to 1), which further impedes standard methods.
In this session, I will address strategies to visualize the microbiome through lecture and hands-on practical exercises
This will lay the foundation for, and aid interpretation of downstream analysis.

15:30 – 16:00 Coffee Break

16:00 – 18:00 Differential Abundance Analysis by Olivier Thas, Data Science Institute, Hasselt University, Belgium

Abstract
In this lecture we will start from the OTU/ASV table and look into the problem of testing for differential abundance (DA). There is a broad spectrum of methods available, and it is hard to choose the most appropriate. We will first discuss some general principles (e.g. normalisation, FDR control, sensitivity, …) and subsequently walk through some of the more popular and better methods (e.g. ANCOM-BC, LinDA, CoDa methods, Wilcoxon-Mann-Whitney, …). In a hands-on session we will apply several of the methods to a few datasets. In addition we will spend attention to chosing the taxonomic rank to which the DA method is applied, combining results of different DA methods and reporting.

The second day will focus on additional skills in microbiome research and multi-omics integration. Participants will explore new technologies and bioinformatics tools, such as metagenome assembly, multi-table methods, and covariate-aware testing to gain deeper insights from complex datasets. We will also cover microbial networks and dynamics, integration of data sets, and development of biomarkers, showcasing modern ways of working with microbiome data. Each session will build upon the foundational knowledge from Day 1.

Program

09:00 – 10:00 Investigating the functional potential of the gut microbiome using shotgun metagenomic data by Georg Zeller, European Molecular Biology Laboratory (EMBL), Germany

10:00 – 11:00 Multi-table methods in microbiome data science by Leo Lahti, Department of Computing, University of Turku, Finland

Abstract
The diversity of various ‘omics data sets and their combinations poses challenges for explainable and verifiable analyses in microbial ecology. This talk will demonstrate how recent advances in statistical programming can help to manage relationships among diverse omics data while implementing reproducible data science workflows. The talks will conclude by highlighting current trends in computational data integration in the context of microbiome research.

11:00 – 11:30 Coffee Break

11:30 – 12:30 Identifying microbiota sub-communities using topic models by Laura Symul, Institute for Statistics, Biostatistics, and Actuarial Sciences at UCLouvain, Belgium

Abstract

Microbiota communities can be highly diverse, with hundreds of different species cohabitating in the same ecosystem (alpha diversity). Across ecosystems or samples, we also observe differences in microbiota composition, with some bacteria present in some samples and absent (or in lower abundance) in others (beta diversity). Due to interactions such as metabolic exchanges, some groups of bacteria frequently co-occur (positive interactions) or are mutually exclusive (negative interactions), forming sub-communities. These sub-communities result in samples being more similar to each other than expected by chance, if there were no interactions between bacteria.
Traditionally, clustering methods have been applied to microbiota data to define community types (CT). However, sub-communities often co-exist within samples, and most large microbiota datasets do not support clear clustering evidence. “Topic models”, initially proposed for modeling population structure and popularized in the natural language processing field, provide an elegant alternative to clustering. These models can identify sub-communities and reduce the dimensionality of microbiota data.
This presentation will explain what topic models are, how they can be applied to microbiota data, and their advantages over clustering methods. It will also introduce the alto R package, which implements a method to “align topics across resolution” (K, the number of topics) to identify both robust and spurious topics.

12:30 – 13:30 Lunch

13:30 – 14:30 Old Problems and New Approaches in Microbial Co-occurrence Network Analysis by Ksenia Guseva, Department of Microbiology and Ecosystem Science, University of Vienna, Switzerland

Abstract
Over the past decade, co-occurrence network analysis has become a popular tool for studying microbial communities across various ecosystems, including soil, gut, and marine environments. However, questions remain about the type of signals these analyses can reliably uncover. In particular, networks constructed from correlations often provide unreliable information about microbial interactions, and it is still highly debated whether these networks represent meaningful ecological relationships among microorganisms. In this talk, I will explore several key challenges that affect our ability to accurately interpret microbial associations, even when we use absolute abundances, uncover environmental confounders and use samples of small volume. I will demonstrate these issues using both real and simulated datasets. Finally, I will introduce a promising new framework for principled network reconstruction that offer a potential solutions to some of these challenges.

14:30 – 15:30 TBA

15:30 – 16:00 Coffee break

16:00 – 17:00 Detection of Joint Microbiome Biomarkers in Intervention Experiments by Ziv Shkedy, Data Science Institute, Hasselt University, Belgium

Abstract
In microbiome studies, identifying biomarkers that reflect the complex interactions between the microbiome and health outcomes is essential for advancing the understanding and improving disease management. In the past, research was focused on the detection of single feature biomarkers, this allow us to better understand the microbiome effect on the health outcome. This paper extends the prior work about microbiome biomarkers by combining information from multiple taxa to identify a joint biomarker relevant to continuous clinical outcomes.
We applied the least absolute shrinkage and selection operator (LASSO) and Elastic net methods (Hastie et al., 2015; Zou and Hastie, 2005), aiming to find a combination of taxa that can serve as biomarker(s) for the response of interest. The information theory approach (Alonso and Molenberghs, 2007) was used for biomarker(s) construction. Monte Carlo cross-validation with 1000 repetitions was used to reduce bias and improve the reliability of the prediction. A LASSO/ Elastic net and relax LASSO after MCCV were fitted to a fixed list of taxa. The methods were evaluated using mean squared error, the correlation between the observed and expected values.
Two datasets were used to illustrate the proposed method. The high salt diet study was conducted to investigate the effect of diet on both microbiome and cancer tumor size in rats. After filtering and identifying relevant genera, the analysis of the top five most relevant genera resulted in a 73.27% reduction in the uncertainty of tumor size when the microbiome predictor was known. This highlights the substantial impact of microbiome data on tumor growth. The CERTIFI study, conducted to investigate the associations between the fecal microbiota and the therapeutic response of Crohn’s disease patients used ustekinumab as a treatment. After filtering, the analysis of the top five most frequently selected taxa showed that knowing the microbiome predictor reduced the uncertainty in the change of CDAI at week 6 from baseline by 10.51%.

17:00 Closing