descope.inference

class BaseInference(test_csv_template_fp: str, adata: str | AnnData, pretrained_model_name_or_path: str, pert_col: str = 'perturbation', ctrl_name: str = 'control')

Bases: CellEvalMixin, ABC

inference(gene_embs_file: str, device: device, batch_size: int = 128, use_generated_control_cells: bool = False, seed: int = 42)

abstract load_model(pretrained_model_name_or_path: str) → PreTrainedModel

abstract preprocess_adata(adata: AnnData) → AnnData

static write_h5ad(adata: AnnData, save_path: str)

class CellEvalMixin

Bases: object

A mixin class designed to extend evaluation capabilities for single-cell perturbation prediction tasks. This class provides methods to compute various metrics on predicted and real AnnData objects, including differential expression (DE) analysis etc.

Example:

>>> from descope.inference import CellEvalMixin

>>> evalmixin = CellEvalMixin()
>>> print(evalmixin.all_metrics)  # list all metrics in cell-eval

>>> results, agg_results, evaluator = evalmixin.compute_metrics(...)

>>> # Compute extra metrics
>>> evalmixin.extra_metrics_func.pearson(evaluator.anndata_pair, ...)  # pearson
>>> evalmixin.extra_metrics_func.pearson_delta_on_topk_de(evaluator.anndata_pair, ...)  # pearson_delta_on_topk_de
>>> evalmixin.extra_metrics_func.direction_match_on_topk_de(evaluator.anndata_pair, ...)  # direction_match_on_topk_de
>>> evalmixin.extra_metrics_func.edistance(...)  # edistance

property all_anndata_pair_metrics: list

property all_de_metrics: list

property all_metrics: list

check_whether_skip_de_metrics(metrics_to_calculate: list) → bool

compute_metrics(adata_pred: str | AnnData, adata_real: str | AnnData, metrics_to_calculate: list, de_pred: str | None = None, de_real: str | None = None, control_pert: str = 'non-targeting', pert_col: str = 'target_gene', de_method: str = 'wilcoxon', num_threads: int = -1, outdir: str = './cell-eval-outdir') → tuple[DataFrame, DataFrame, MetricsEvaluator]

property extra_metrics: list

class InferenceForATAC(test_csv_template_fp: str, adata: str | AnnData, pretrained_model_name_or_path: str, pert_col: str = 'perturbation', ctrl_name: str = 'control')

Bases: BaseInference

load_model(pretrained_model_name_or_path: str) → PreTrainedModel

preprocess_adata(adata: AnnData) → AnnData

class InferenceForRNA(test_csv_template_fp: str, adata: str | AnnData, pretrained_model_name_or_path: str, pert_col: str = 'target_gene', ctrl_name: str = 'non-targeting', target_sum: float = 10000.0, skip_raw_counts_check: bool = False)

Bases: BaseInference

load_model(pretrained_model_name_or_path: str) → PreTrainedModel

preprocess_adata(adata: AnnData) → AnnData