Justin van der Hooft

Justin J.J. van der Hooft is an Assistant Professor in Computational Metabolomics in the Bioinformatics Group at Wageningen University & Research, NL, and an author of >100 peer-reviewed articles in the metabolomics field. Justin is very fascinated by the ingenuity of nature in creating marvellous chemical structures.

He obtained his PhD (2012) in Systematic Metabolite Annotation and Identification at the Biochemistry and Bioscience groups in Wageningen. After a postdoctoral period in Glasgow, UK, studying both analytical and computational aspects of metabolite structure annotation, and together with Joe Wandy & Simon Rogers coining MS2LDA unsupervised substructure discovery, he returned to Wageningen. Since 2020, his team has been developing computational metabolomics strategies to decompose mass spectral data into structure and substructure information.

By linking genome and metabolome mining, his team studies plant, food, and microbiome-associated metabolites to find novel bioactive metabolites. Recently developed tools and frameworks include MS2Query to perform analogue search, FERMO to prioritize metabolite features and profiles by enabling effective and reproducible data integration and data filtering strategies, and NPLinker to handle multi-omics data for natural products discovery. Since 2022, he is also a Visiting Professor in Johannesburg. Got interested? Find out more and meet the team here: https://vdhooftcompmet.github.io.

Presentation

Boosting Mass Spectrometry-based Metabolomics with Machine-Learning Powered Analogue Search, Substructure Discovery, and Cross-Ion Mode Matching.

Introduction & Motivation

Metabolomics has often been coined as the ultimate phenotyping tool. Indeed, a full understanding of chemical profiles of complex biological extracts would lead to a deeper biochemical understanding of many biological processes. Technological advances made analytical chemical analysis ever more sensitive; however, solving metabolite structures from analytical data remains very difficult and was coined as one of the "Grand Challenges" in the metabolomics field. In this presentation, I will highlight recent advances in computational metabolomics made by my group and others that use networking and machine learning strategies to overcome this challenge.

Approach

I will briefly describe some of the motivations and concepts of a number of metabolomics mining and annotation tools to better understand the complex metabolite mixtures that specialized metabolites are typically part of. Recently, my group proposed Spec2Vec and MS2DeepScore, novel machine learning-based mass spectral similarity scores that improve library matching and analogue searching and that formed the basis for the also machine learning-based MS2Query analogue search tool. The MS2Query tool, along with many other features, is also part of the new FERMO tool that facilitates the prioritisation of novel bioactive molecules. I will also highlight our recent collaborate work on MS2LDA 2 development, a more scalable tool for unsupervised substructure pattern discovery then its predecessor. Furthermore, automated substructure annotation guidance with help of Spec2Vec was added. Finally, I will show how MS2DeepScore 2.0 now allows for cross-ion mode matching between positive and negative ionization mode profiles, resulting in novel spectral links and possibilities for integrative molecular networking. This scoring has been built into the popular tool mzmine to democratize its use within the metabolomics community.

Results

I will show examples of how the above-mentioned methodological advances boost annotation power using mass spectral librarie data as well as case studies from various natural origins including plants, bacteria, and mushrooms.

Discussion

I will finish off with a broader outlook on how modular tools and interactive dashboards will advance our understanding of the role of metabolites and their complex molecular interactions that underpin growth, development, and health.