Repository-wide screening for bioactivity and protein binding using small molecule mass spectrometry data

Projektbeschreibung

Tandem-Massenspektren auf bioaktive Stoffverbindungen und Protein-Targets untersuchen

Die Proteinbindung ist ein Schlüsselfaktor für die Bioaktivität einer Stoffverbindung, denn sie bestimmt das Stoffverhalten als Therapeutikum oder Toxin. Unbekannte kleine Moleküle in biologischen Proben werden durch die Annotation von Stoffverbindungen mittels Tandem-Massenspektrometrie (MS) identifiziert. In diesem Zusammenhang wird das ERC-finanzierte Projekt BindingShadows maschinelle Lernmodelle entwickeln, die auf der Grundlage der molekularen Tandem-Massenspektren vorhersagen, ob ein unbekanntes kleines Molekül bioaktiv ist oder an ein bestimmtes Protein bindet. Die Modelle werden dann auf das Repository-weite Screening von Millionen von Tandem-Massenspektren in den verfügbaren Datensätzen angewendet. Diese Methoden werden über eine Plattform zur Verfügung gestellt, auf der Informationen über die Bioaktivität erhältlich sind, ohne dass zusätzliche rechnerische oder experimentelle Analysen erforderlich sind.

Ziel

Bioactivity and ligand binding remain central topics for understanding protein function, for drug discovery and toxicology. Mass spectrometry (MS) allows to detect thousands of small molecules in a biological sample with a single experimental run. Compound annotation is carried out using tandem MS. I will develop machine learning models that predict whether a query small molecule has a certain bioactivity or is binding to a certain protein, where the only information we have about the query molecule is its tandem mass spectra. The identity and molecular structure of the query small molecule is unknown; it is unknown to the models, to me, and potentially even to mankind. Clearly, bioactivity and binding are two flavors of the same problem, and the differentiation is solely driven by the available training data. Notably, elucidating the structure of the small molecule is not the focus of the project, and may be postponed until after experimental confirmation. Models will not predict one particular bioactivity or protein binding, but rather tens in parallel. Evolution has, through variation and selection, optimized the structure of small molecules for tasks such as communication and warfare, and the pool of natural products is enriched with bioactive compounds. My project will allow to harvest the information.

I will apply my models at a repository scale, screening millions of tandem MS spectra in thousands of datasets for small molecules that are most likely to have bioactivity or to bind. I will also make my methods available through our well-established SIRIUS platform, allowing users to derive information about bioactivity without the need of additional experiments. I will enable the worldwide community to hunt for drug leads or unknown toxic compounds, without the need of additional experiments or computational analyses. Finally, being able to find small molecules that bind to a protein can provide helpful information to understand protein function.