Deconstructing Complexity: A Computational Topology Approach to Trajectory Inference in the Human Thymus withtviblindi

Abstract

Interpreting and understanding complex, organ-level mass cytometry datasets represents a formidable interdisciplinary challenge. This study aims to identify, describe, and interpret potential developmental trajectories of thymocytes and mature T cells. We developedtviblindi, a trajectory inference algorithm that integrates several autonomous modules - pseudotime inference, random walk simulations, real-time topological classification using persistence homology, and autoencoder-based 2D visualization using thevaevictisalgorithm. This integration facilitates an interactive exploration of developmental trajectories. The utility and proficiency oftviblindiare demonstrated through comprehensive analysis of the thymic and peripheral T-cell compartment. Our approach not only uncovers and elucidates the canonical CD4 and CD8 development but also offers insights into various checkpoints such as TCRβ selection and positive/negative selection, elucidating the crossroads between further development and apoptosis. Finally, we identify and thoroughly characterize thymic regulatory T cells, tracing their development from the negative selection stage to mature thymic regulatory T cells with an extensive proliferation history and an immunophenotype of activated and recirculating cells.tviblindiis a versatile and generic approach suitable for any single-cell dataset, equipping biologists with an effective tool for interpreting complex data.

Tviblindi algorithm identifies branching developmental trajectories of human B cell development

“…On the concatenated dataset we selected the developmental point of origin at CD34+ Stem cells. The tviblindi algorithm 15 was tasked to construct 5000 random walks directed away from the origin (CD34+ Stem cell) with respect to the calculated pseudotime on the nearest neighbor graph (KNNg) of all single cell events. As the KNNg is directed by the pseudotime, endpoints are automatically detected when a random walk reaches a vertex (single-cell event) with no out-going edges.…”

“…For visualization of the mass cytometry data, we used the deep learning-based dimensionality reduction technique using the vaevictis model 15 , one of the autonomous modules integrated in the tviblindi . For the projection, the healthy bone marrow (n=4) and healthy peripheral blood (n=4) samples were manually debarcoded and exported as individual FCS files.…”

“…For trajectory inference (TI), we used our recent framework called tviblindi 15 , an algorithm integrating several autonomous modules - pseudotime inference, random walk simulations, real-time topological classification using persistence homology, and autoencoder-based 2D visualization using the vaevictis model. For the TI, the same concatenated FCS file as for the vaevictis projection was used.…”

Defects in B-lymphopoiesis and B-cell maturation underlie prolonged B-cell depletion in ANCA-associated vasculitis

“…In 5/7 BM, we dissected B-cell developmental stages from HSCs to immature/transitional B cells by mass cytometry by time of flight (CyTOF) and compared it with four HD BM (online supplemental figure S5A, online supplemental table S2). Using vaevictis dimensionality reduction, 22 we observed cell clusters that could be attributed to distinct stages of B-cell development (figure 3A) as identified by manual gating 23 (online supplemental figure S5A). All clusters were present in HD BM aspirates.…”

“…Mass cytometry sample acquisition was performed on Helios instrument (Standard BioTools, CyTOF 6.7.1014 software) after preparation according to the manufacturer's recommendation. For visualisation of the mass cytometry data, we used our deep learning-based dimensionality reduction technique using the vaevictis model, 22 one of the autonomous modules integrated in the tviblindi tool.…”