Fig 1.
Pseudocell Tracer, a framework for modeling cellular trajectories in complex systems.
(A) An overview of pseudotime trajectory inference. (B) Some scenarios that may obstruct pseudotime ordering. (C) Pseudocell Tracer. Given some prior knowledge about a model system, we aim to predict expression trajectories by generating pesudocells at regular intervals along a virtual cell-state axis, even though such cells may be sparsely captured in single cell profiling data.
Fig 2.
Class switch recombination process.
(A) Overview of experimental system. (B) Relative isotype expression for all cells. N = 7,065. All isotype expressions sum to one for a given cell. (C) UMAP of RNA-seq data, colored by isotype. (D) Output from Moncole3 and Slingshot.
Fig 3.
Pseudocell Tracer efficiently integrates adjacent biological information and accurately simulates gene expression profiles in pseudocells.
(A) Overview of neural network model combining a supervised autoencoder with a conditional GAN. (B) UMAP visualization of the input and output used in the supervised autoencoder; encoder (top) and decoder (bottom). (C) Scatter plot between observed and predicted expression values on held out cells. r denotes Pearson correlation between ground truth and predicted values. Isotype expression (left) and example CSR genes (right). (D) UMAP visualization applied to cGAN prediction and subsequent output from decoder.
Fig 4.
Pseudocell Tracer models IgG1 class switching process.
(A) Pseudocells generated along the IgM to IgG1 axis. Plot of relative expression of Aicda, Ighm and Ighg1 along the IgM to IgG1 axis, where solid line indicates average expression and shading indicates 95% confidence interval. (B) Associative clustering of genes during CSR. Regions of early (left), middle (center), and late (right) transcriptional dynamics are depicted. Plots of relative expression for key genes with specific dynamics, including (C) Stat6, (D) Bach2, (E) Il4ra, and (F) Ifngr1.