Skip to content

Location-specific sampling of cell type signatures and background #175

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
vitkl wants to merge 2 commits into
base: master
Choose a base branch
from
Open

Location-specific sampling of cell type signatures and background #175

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
b28453c
added location-specific sampling using gamma distribution
vitkl Jul 6, 2022
febf2e5
fix sampling:non-centered parametrisation
vitkl Jul 11, 2022
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
61 changes: 53 additions & 8 deletions cell2location/models/_cell2location_module.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -98,6 +98,7 @@ def __init__(
init_vals: Optional[dict] = None,
init_alpha=20.0,
dropout_p=0.0,
location_sampling: bool = False,
):

super().__init__()
Expand All @@ -122,6 +123,8 @@ def __init__(
if self.dropout_p is not None:
self.dropout = torch.nn.Dropout(p=self.dropout_p)

self.location_sampling = location_sampling

if (init_vals is not None) & (type(init_vals) is dict):
self.np_init_vals = init_vals
for k in init_vals.keys():
Expand Down Expand Up @@ -192,6 +195,9 @@ def __init__(
self.register_buffer("n_groups_tensor", torch.tensor(self.n_groups))

self.register_buffer("ones", torch.ones((1, 1)))
self.register_buffer("zeros", torch.zeros((1, 1)))
self.register_buffer("five", torch.tensor(5.0))
self.register_buffer("ten", torch.tensor(10.0))
self.register_buffer("ones_1_n_groups", torch.ones((1, self.n_groups)))
self.register_buffer("ones_n_batch_1", torch.ones((self.n_batch, 1)))
self.register_buffer("eps", torch.tensor(1e-8))
Expand Down Expand Up @@ -460,12 +466,47 @@ def forward(self, x_data, idx, batch_index):
# =====================Expected expression ======================= #
if not self.training_wo_observed:
# expected expression
mu = ((w_sf @ self.cell_state) * m_g + (obs2sample @ s_g_gene_add)) * detection_y_s
alpha = obs2sample @ (self.ones / alpha_g_inverse.pow(2))
# convert mean and overdispersion to total count and logits
# total_count, logits = _convert_mean_disp_to_counts_logits(
# mu, alpha, eps=self.eps
# )
if not self.location_sampling:
mu = ((w_sf @ self.cell_state) * m_g + (obs2sample @ s_g_gene_add)) * detection_y_s
alpha = obs2sample @ (self.ones / alpha_g_inverse.pow(2))
else:
# sampling location-specific cell state signatures ====
cell_state_sigma = pyro.sample(
"cell_state_sigma",
dist.Exponential(self.ten * self.ten).expand([1, 1, 1]).to_event(2),
) # (1, 1).squeeze()
cell_state = torch.exp(
torch.log(self.cell_state.unsqueeze(-3))
+ dist.Normal(self.zeros, self.ones).sample([w_sf.shape[0], self.n_factors, self.n_vars]).squeeze()
* cell_state_sigma
)
biol_mu = torch.einsum("sf,sfg->sg", w_sf, cell_state)

# sampling location-specific background counts ====
s_g_gene_add_sigma = pyro.sample(
"s_g_gene_add_sigma",
dist.Exponential(self.ten * self.ten).expand([self.n_batch, 1]).to_event(2),
) # (1, 1)
s_g_gene_add_sigma = (obs2sample @ s_g_gene_add_sigma).unsqueeze(-1)
s_g_gene_add = torch.exp(
torch.log(s_g_gene_add.unsqueeze(-3))
+ dist.Normal(self.zeros, self.ones).sample([w_sf.shape[0], self.n_batch, self.n_vars]).squeeze()
* s_g_gene_add_sigma
)
background_mu = torch.einsum("se,seg->sg", obs2sample, s_g_gene_add)

# sampling location-specific technology sensitivity ====
if False:
m_g_alpha = pyro.sample(
"m_g_alpha",
dist.Exponential(self.ten).expand([self.n_batch, 1]).to_event(2),
) # (1, 1)
m_g_alpha = obs2sample @ (self.ones / m_g_alpha.pow(2))
m_g = dist.Gamma(m_g_alpha, m_g_alpha / m_g).sample([1]).squeeze(-3)

# compute expected value ====
mu = (biol_mu * m_g + background_mu) * detection_y_s
alpha = obs2sample @ (self.ones / alpha_g_inverse.pow(2))

# =====================DATA likelihood ======================= #
# Likelihood (sampling distribution) of data_target & add overdispersion via NegativeBinomial
Expand All @@ -481,8 +522,12 @@ def forward(self, x_data, idx, batch_index):

# =====================Compute mRNA count from each factor in locations ======================= #
with obs_plate:
mRNA = w_sf * (self.cell_state * m_g).sum(-1)
pyro.deterministic("u_sf_mRNA_factors", mRNA)
if not self.location_sampling:
mRNA = w_sf * (self.cell_state * m_g).sum(-1)
pyro.deterministic("u_sf_mRNA_factors", mRNA)
else:
mRNA = torch.einsum("sf,sfg,sg->sf", w_sf, cell_state, m_g)
pyro.deterministic("u_sf_mRNA_factors", mRNA)

def compute_expected(self, samples, adata_manager, ind_x=None):
r"""Compute expected expression of each gene in each location. Useful for evaluating how well
Expand Down