dilimap.models.ToxPredictor

class dilimap.models.ToxPredictor(version='v1', model=None, multioutput=None, model_params=None, local_filepath=None)

ToxPredictor pre-trained v1 or initialize new model for re-training.

It supports multi-output classification and custom hyperparameters for model training.

Parameters:

version (str) – Model version; for TGP use ‘v0.1’, for DILImap use ‘v1’.
model (str) – The model type to be used, for instance ‘RandomForestClassifier’.
multioutput (bool) – Whether to use a multi-output classifier for the model.
model_params (dict) – Dictionary of parameters to pass to the model.

Usage:

from dilimap.models import ToxPredictor

model = ToxPredictor('v1')
results = model.predict(data)
safety_margins = model.compute_safety_margin(data, 'compound_name', 'dose_uM', 'Cmax_uM')

__init__(version='v1', model=None, multioutput=None, model_params=None, local_filepath=None): Initializes the ToxPredictor model.

Methods

`__init__`([version, model, multioutput, ...])	Initializes the ToxPredictor model.
`compute_DILI_dose_regimes`([IC50_uM, dose_range])	Determines DILI dose regimes based on specified IC50.
`compute_empirical_DILI_risk`([IC50_uM, ...])	Calculates the empirical DILI risk likelihoods based on safety margin results.
`compute_safety_margin`(data[, pert_col, ...])	Computes the margin of safety (MOS) based on predicted toxicity.
`cross_validate`(X, y[, n_splits, groups, ...])	Performs cross-validation on the model using GroupKFold.
`feature_importances`([dili_pathways])	Compute feature-level AUC, mean decrease in impurity (MDI), and correlation with the true DILI label (Pearson and Spearman).
`plot_DILI_dose_regimes`(compound[, ...])	Plots the DILI likelihood dose-response curve for a given compound.
`predict`(data)	Predicts the DILI risk from pathways signatures.
`predict_cv`()	Predicts the DILI risk from pathways signatures.
`predict_proba`(data[, estimators, rescale])	Predicts class probabilities using trained estimators.
`predict_proba_across_estimators`(data[, ...])	Predict and collects class probabilities across all trained estimators.
`predict_proba_cv`([estimators, rescale])	Predicts class probabilities for cross-validated models.
`pull_data`()	Loads the training data from the model.
`save_model`(filepath[, push_to_s3])	Save the model to S3 registry.

Attributes

`DILI_pathways`	Returns the groups of pathways associated with DILI predictions.
`classes_`	Returns the classes of the classifier estimators.
`estimators`	Returns the trained estimators from cross-validation.
`indices`	Returns the training/test indices from cross-validation.
`multioutput_estimators`	Returns the estimators for each output dimension in a multi-output model.
`test_fold_assignments`