Working Group 1

The main objectives of Workgroup 1 are to continuously evaluate the state-of-the-art machine learning methods and respective software applied in human microbiome studies.  Based on the obtained results, WG1 is defining priority areas for novel machine learning and statistics applications for the analysis of microbiome data. The literature and software watch performed by the WG1 is concentrated on the application of machine learning in microbiome studies related to causality and clinical use for diagnostics, prognostics and therapeutics.  The literature and software review is based on the combination of the crowdsourcing approach (action participants are submitting most recent relevant publications/software to the web-based database) and input from targeted search by teams within the workgroup. The workgroup is producing review reports annually and novel machine learning methods will be continuously evaluated in terms of functionality and suitability for tackling the action’s aims. In addition, priority areas for novel machine learning and statistics applications that better address the specific challenges of human microbiome analysis will be defined.

Working Group 2

Working group 2 works to collect datasets describing microbiomes and characteristics of the underlying cohorts (from larger projects and repositories) in order to use them for elucidation which machine learning methods are most robust and comparable, to provide more optimised parameters for the use of these methods and to develop ML methodologies that could be used in open challenges such as ML-Microbiome DREAM Challenge.

Working Group 3

The main goal of Workgroup 3 is to optimize and standardize the use of state-of-the-art ML techniques, for various microbiome data types. The result will be that of defining Standard Operating Procedures (SOPs) specific to various microbiome data types, human body ecosystems, and research questions.

Specifically, the WG3 will investigate opportunities for automating the established SOPs into pipelines for translational use by clinicians and non-experts. This requires to take into account various aspects, such as different tasks, different algorithms, different combinations of algorithms, different parameters, for different data types (16S rRNA amplicons, shotgun metagenomics, and metatranscriptomics), different ecosystems e.g. high/low diversity and variability) and different research questions (e.g. diagnostics, prognostics, causality). The consideration of all these aspects also requires the adoption of automated machine learning techniques (AutoML), which are able to find the best pipeline for different data types.

The standardized use of ML/Stats methods into SOPs will be made publicly available on the Web-portal. Moreover, based on the complexity of the SOPs, we will investigate whether SOPs can be automated into user-friendly pipelines for translational use by clinicians and non-ML/microbiome experts.

Working Group 4

Workgroup 4 of our ML4Microbiome COST Action is responsible for bridging existing gaps between ML and microbiome experts. The general idea of our workgroup is to have microbiome, bioinformatics, ML, statistics and computer-science scientists share the same vocabulary as it has been done in other ‘omics’ applications, followed by deeper discussions of further research and development opportunities. In this manner, the sharing of knowledge and experience across and within fields requires frequent face-to-face meetings, with ample hands-on training and dissemination opportunities.

Thus, WG4 Dissemination and Training activities constitute an integral, horizontal component of the ML4Microbiome COST Action

To achieve this, we aim to:

• Disseminate the results of the Action in peer-reviewed journals, on the web-portal, at international conferences and through end-user workshops;
• Organise training courses and workshops in microbiome-related ML methodologies recommend by Action members;
• Combine the knowledge from the other workgroups into a journal paper describing “best practices” for using ML to analyse microbiome data;
• Share educational material both in the form of text and videos, including a YouTube channel and information in social media.