课前准备-单细胞velocity分析（贝叶斯模型）

作者，Evil Genius

速率

• Probabilistic modeling of RNA velocity
• Direct modeling of raw spliced and unspliced read count
• Multiple uncertainty diagnostics analysis and visualizations
• Synchronized cell time estimation across genes
• Multivariate denoised gene expression and velocity prediction

实现目标

整体框架

1、前处理

import scvelo as scv
adata = scv.read("local_file.h5ad")

adata.layers['raw_spliced']   = adata.layers['spliced']
adata.layers['raw_unspliced'] = adata.layers['unspliced']
adata.obs['u_lib_size_raw'] = adata.layers['raw_unspliced'].toarray().sum(-1)
adata.obs['s_lib_size_raw'] = adata.layers['raw_spliced'].toarray().sum(-1)
scv.pp.filter_and_normalize(adata, min_shared_counts=30, n_top_genes=2000)
scv.pp.moments(adata, n_pcs=30, n_neighbors=30)

2、训练模型

from pyrovelocity.api import train_model
from pyrovelocity.plot import vector_field_uncertainty
num_epochs = 1000 # large data
# num_epochs = 4000 # small data
# Model 1
adata_model_pos = train_model(adata,
                               max_epochs=num_epochs, svi_train=True, log_every=100,
                               patient_init=45,
                               batch_size=4000, use_gpu=0, cell_state='state_info',
                               include_prior=True,
                               offset=False,
                               library_size=True,
                               patient_improve=1e-3,
                               model_type='auto',
                               guide_type='auto_t0_constraint',
                               train_size=1.0)
# Or Model 2
adata_model_pos = train_model(adata,
                               max_epochs=num_epochs, svi_train=True, log_every=100,
                               patient_init=45,
                               batch_size=4000, use_gpu=0, cell_state='state_info',
                               include_prior=True,