Servicenavigation

Detailseite

Zurück

Projekt Druckansicht

Vermessung der molekularen Struktur komplexer Gewebe mittels räumlicher Kovarianztensoren

Antragsteller Dr. Simon Mages
Fachliche Zuordnung Bioinformatik und Theoretische Biologie
Kern- und Elementarteilchenphysik, Quantenmechanik, Relativitätstheorie, Felder
Förderung Förderung von 2020 bis 2022
Projektkennung Deutsche Forschungsgemeinschaft (DFG) - Projektnummer 442489549
 
Erstellungsjahr 2023

Zusammenfassung der Projektergebnisse

Experimental techniques to measure biological systems in increasing detail on a molecular level are steadily improved and become more available. These include ”omics” targeting different levels of molecular information in cells, like proteins (proteomics) or RNA (transcriptomics). In contrast to genomics which studies the (nearly) static content of genetic information in a cell and which is (nearly) constant across all cells in an organism, other omics also capture the activity and dynamic states of cells. They can measure molecular information at single cell resolution and in recent years also spatially resolved down to sub-cellular length scales. All these experimental omics datasets generate massive amounts of data which need to be interpreted and analyzed. A very important task is to assign cell-types to the measurements, like B cell or epithelial cell. This is necessary to not confuse differences in the activity of cells in certain conditions with a difference in cell-type composition between the conditions, which has very different biological interpretations. In many spatial transcriptomics methods there is an additional complication in this task as multiple cells can contribute to the measurement of a single profile. Surprisingly, even though this problem is of paramount importance for many applications, this basic problem was not yet solved in the beginning of the project. As all downstream applications like the analysis of spatial correlations depend on it, the projects focus shifted to take on the compositional annotation challenge. To get a compositional dataset with known ground truth, the first step was to simulate a spatial dataset with compositional annotations. The simulation runs on a torus to get boundary artefacts under control as is commonplace also in Lattice Quantum Field Theory Simulations. The compositional annotation method developed and benchmarked on this simulated data is very modular and rather a framework for developing new methods than a single annotation method. In its core it uses Optimal Transport which is very fast, can incorporate prior knowledge about the dataset, and is quite popular in machine learning. The overlap of probability amplitudes in quantum mechanics are used as generic cost matrix for the objective function of the Optimal Transport setup. This is extended by a set of wrappers to better account for the compositional nature of the annotation, for within-class heterogeneity in the reference dataset, and for differences in the per-gene capture efficiencies across different experimental methods. The final annotation method turned out to be applicable to a wide range of annotation problems apart from compositional annotation of spatial transcriptomics data, e.g. in the prediction of cell fates during cell differentiation, and the annotation of single cell data at high levels of dropout and ambient RNA. With another wrapper it can also be used to efficiently annotate single molecules in multiplexed FISH-like methods, where the positions of single RNA molecules are measured with subcellular resolution. The published framework, TACCO, also contains methods for downstream analysis and visualization of the resulting compositional annotations. During the project other tools were published as well to tackle the compositional annotation problem. In benchmarks TACCO excelled in runtime and memory requirements, while keeping or exceeding the accuracy achieved by other methods. This low resource requirement makes it possible to scale to much larger datasets or to work quasiinteractively instead of waiting for hours to see the results. Along with the ease of use of the framework, this has the potential to shorten the time from spatial omics data generation to bio-medical insights.

Projektbezogene Publikationen (Auswahl)

 
 

Zusatzinformationen

© 2025 DFG Kontakt / Impressum / Datenschutz

Textvergrößerung und Kontrastanpassung