scintegral.classifier module

scintegral.classifier.classify_cells(expr_mat, cov_mat, size_factor, marker_onehot, device=device(type='cpu'), e_converge=0.0001, lr=0.05, n_itr_max=1000, prior_mean=2, prior_width=0.05, disp_init=2)

The cell-type classifier.

Parameters

• expr_mat – An n (number of cells) x g (number of genes) matrix that contains the raw expression counts.

• cov_mat – An n x p (number of batches) matrix that contains the batch membership of samples.

• size_factor – An n vector of size factors (or any offset variable replacing the size factor).

• marker_onehot – A g x t (number of cell-types) matrix containing marker information for each cell-type.

• device – A device used in computation (default: cpu).

• e_converge – A real number used to determine convergence (default:1e-4).

• lr – A real number used as the initial learning-rate (default:0.05).

• n_itr_max – The maximum number of iterations in the likielihood optimization step (default:1000).

• prior_mean – The threshold parameter for initialization (default:2).

• prior_with – The width parameter for initialization (default:0.05).

• disp_init – The initial dispersion parameter of the negative binomial likelihood (default:1.5).

Returns list

A list of assigned cell-type labels.

Returns class

A torch module object containing the fitted parameters.

scintegral.classifier.classify_cells_internal(expr_mat, cov_mat, size_factor, n_type, marker_onehot, device, e_converge, lr, n_itr_max, prior_mean, prior_width, disp_init)

scintegral.classifier.nb_logpmf(k, n, p)

scintegral.classifier.phi_filter(phi)

scintegral.classifier.phi_filter_grad(phi)

scintegral.classifier.posterior_probs(delta, beta, phi, expr_mat, cov_mat, size_factor, marker_onehot)

class scintegral.classifier.scintegral_loss

Bases: torch.autograd.function.Function

static backward(ctx, grad_output)

Defines a formula for differentiating the operation.

This function is to be overridden by all subclasses.

It must accept a context ctx as the first argument, followed by as many outputs did forward() return, and it should return as many tensors, as there were inputs to forward(). Each argument is the gradient w.r.t the given output, and each returned value should be the gradient w.r.t. the corresponding input.

The context can be used to retrieve tensors saved during the forward pass. It also has an attribute ctx.needs_input_grad as a tuple of booleans representing whether each input needs gradient. E.g., backward() will have ctx.needs_input_grad[0] = True if the first input to forward() needs gradient computated w.r.t. the output.

static forward(ctx, delta, beta, phi, expr_mat, cov_mat, size_factor, marker_onehot)

Performs the operation.

This function is to be overridden by all subclasses.

It must accept a context ctx as the first argument, followed by any number of arguments (tensors or other types).

The context can be used to store tensors that can be then retrieved during the backward pass.

class scintegral.classifier.scintegral_loss_nograd

Bases: torch.autograd.function.Function

static forward(ctx, delta, beta, phi, expr_mat, cov_mat, size_factor, marker_onehot)

Performs the operation.

This function is to be overridden by all subclasses.

It must accept a context ctx as the first argument, followed by any number of arguments (tensors or other types).

The context can be used to store tensors that can be then retrieved during the backward pass.

class scintegral.classifier.scintegral_model(expr_mat, cov_mat, size_factor, n_type, marker_onehot, prior_mean, prior_width, disp_init)

Bases: torch.nn.modules.module.Module

forward(only_loss=False)

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

training: bool