MGPfact.jl

A model-based, unsupervised manifold learning method that factors complex cellular trajectories into interpretable bifurcating Gaussian processes of transcription. The complete functionality of MGPfact is accessible in MGPfactR, enabling the discovery of specific biological determinants of cell fate.

installation

]
add https://github.com/renjun0324/MGPfact.jl
add RData Distributions JLD2 Mamba
using MGPfact

quick start

# load test data
nc = 1
using RData
L = 3
Q = 3
rt = 1
test_data = joinpath(dirname(pathof(MGPfact)),"..","test","test.rda")
yx = RData.load(test_data)
yx = yx["murp_matrix_pca"][:,1:Q]
iterations = 200

# import packages
using MGPfact, Mamba

# running model
model = MGPfact.modMGPpseudoT()
SC,inits,scheme = MGPfact.Initialize(yx, Q, L, rt, iterations, nc)

setinputs!(model, SC)
setinits!(model, inits)
setsamplers!(model, scheme)

@time sim1 = mcmc(model, SC, inits, iterations, burnin = 0, chains = nc)

# save
using JLD2
write(string("iter",sim.model.iter,"_bi",sim.model.burnin,".jls"), sim)
@save string("inits.jld2") inits

multiple chains

# load test data
using RData
nc = 3
L = 3
Q = 3
rt = 1

test_data = joinpath(dirname(pathof(MGPfact)),"..","test","test.rda")
yx = RData.load(test_data)
yx = yx["murp_matrix_pca"][:,1:Q]
iterations = 200

# import packages
using Distributed
addprocs(nc)
@everywhere using MGPfact, Mamba

# running model
model = MGPfact.modMGPpseudoT()
SC,inits,scheme = MGPfact.Initialize(yx, Q, L, rt, iterations, nc)

setinputs!(model, SC)
setinits!(model, inits)
setsamplers!(model, scheme)

@time sim1 = mcmc(model, SC, inits, iterations, burnin = 0, chains = nc)

# save
using JLD2
write(string("iter",sim.model.iter,"_bi",sim.model.burnin,".jls"), sim)
@save string("inits.jld2") inits

In this context, we only demonstrate how to optimize the model in the Julia environment. For specific result organization and trajectory visualization, please refer to MGPfactR.