Awesome

gpseudotime

This repo accompanies our preprint Bayesian Gaussian Process Latent Variable Models for pseudotime inference in single-cell RNA-seq data. To reproduce the figures in the paper:

• The Jupyter notebooks gpseudotime_all.ipynb and monocle_analysis_inversegamma.ipynb in julia_notebooks will reproduce the synthetic and 'monocle' workflows respectively
• The R scripts synthetic_plots.R and monocle_plots.R in plotting will then reproduce the figures

Note to construct the monocle representation you need to run R_notebooks/vignette.Rmd to get the Laplacian Eigenmaps representation. This all relies heavily on HDF5 (through the rhdf5 and HDF5.jl libraries).

Using bgplvm.jl

The main MH algorithm is in bgplvm.jl. Briefly, it is invoked via

B_GPLVM_MH(X, n_iter, burn, thin, 
    t, tvar, lambda, lvar, sigma, svar, 
    r = 1, return_burn = false, cell_swap_probability = 0,
    gamma = 1.0)

where

• X - cell-by-feature matrix
• t, lambda, sigma - initial values for the markov chain (note all are vectors)
• tvar, lvar, svar - variances for the proposal distributions of t, lambda and sigma respectively
• r - repulsion parameter for Corp prior
• return_burn - should the burn period of the traces be returned?
• cell_swap_probability - leave as 0
• gamma - rate parameter for exponential prior on lambda

Structure of repo

• bgplvm.jl contains the MH algorithms written in Julia for inference
• R_notebooks has R markdown notebooks to convert the embeddings to HDF5
• julia_notebooks contains Jupyter notebooks for all the analysis (synthetic + moncole)
• plotting contains R scripts to create the plots for the paper
• data contains MCMC trace data in the form of HDF5 & CSV