splito.simpd

splito.simpd.SIMPDSplitter

Bases: BaseShuffleSplit

The SIMPD (SImulated Medicinal chemistry Project Data) is based on a multi-objective genetic algorithm (MOGA) to split a set of compounds with bioactivity data into one or more training and test sets that differ from each other in ways resembling the differences between the temporal training/test splits observed in medicinal chemistry projects.

It's the implementation proposed in "SIMPD: an Algorithm for Generating Simulated Time Splits for Validating Machine Learning Approaches" available at https://chemrxiv.org/engage/chemrxiv/article-details/6406049e6642bf8c8f10e189.

The source code has been largely inspired by the original authors implementation available at https://github.com/rinikerlab/molecular_time_series/tree/55eb420ab0319fbb18cc00fe62a872ac568ad7f5.

__init__

__init__(
    n_splits: int = 1,
    pop_size: int = 500,
    ngens: int = 100,
    swap_fraction: float = 0.1,
    simpd_descriptors: Optional[pd.DataFrame] = None,
    target_train_frac_active: float = -1,
    target_test_frac_active: float = -1,
    target_test_set_frac: float = 0.2,
    target_delta_test_frac_active: Optional[float] = None,
    target_GF_delta_window: Tuple[int, int] = (10, 30),
    target_G_val: int = 70,
    max_population_cluster_entropy: float = 0.9,
    pareto_weight_GF_delta: float = 10,
    pareto_weight_G: float = 5,
    num_threads: int = 1,
    random_seed: Optional[int] = 19,
    verbose: bool = True,
    verbose_pymoo: bool = True,
    progress: bool = True,
    progress_leave: bool = False,
)

Creates the splitter object.

We invite the user to refer to the original paper for more details on the parameters.

Parameters:

Name	Type	Description	Default
n_splits	int	Number of splits to generate.	1
pop_size	int	The population size for the GA.	500
ngens	int	The number of generations for the GA.	100
swap_fraction	float	The swap fraction for the GA. Swap N% of the bits in each mutation.	0.1
simpd_descriptors	Optional[DataFrame]	The descriptors to use for the GA. If None, the default descriptors from the paper will be used. Load them from splito.simpd.DEFAULT_SIMPD_DESCRIPTORS.	None
target_train_frac_active	float	The target fraction of active compounds in the training set. Set to -1 to disable.	-1
target_test_frac_active	float	The target fraction of active compounds in the test set. Set to -1 to disable.	-1
target_test_set_frac	float	The target fraction of the test set.	0.2
target_delta_test_frac_active	Optional[float]	The target delta of active between the test and training set.	None
target_GF_delta_window	Tuple[int, int]	The target window for the GF delta.	(10, 30)
target_G_val	int	The target G value.	70
max_population_cluster_entropy	float	The maximum cluster entropy.	0.9
pareto_weight_GF_delta	float	The weight for the GF delta.	10
pareto_weight_G	float	The weight for the G value.	5
num_threads	int	The number of threads to use for the GA.	1
random_seed	Optional[int]	The random seed to use for the GA.	19
verbose	bool	Whether to print information about the splitter.	True
verbose_pymoo	bool	Whether to print information about the GA.	True
progress	bool	Whether to display a progress bar.	True
progress_leave	bool	Whether to leave the progress bar after completion.	False

fit

fit(X: np.ndarray, y: np.ndarray, groups: Optional[np.ndarray] = None)

Fit the splitter against a given dataset.

Parameters:

Name	Type	Description	Default
X	ndarray	An array of molecules.	required
y	ndarray	An array of activities (only 1D is supported).	required
groups	Optional[ndarray]	An array of groups.	None

---

splito.simpd.run_SIMPD

run_SIMPD(
    data: pd.DataFrame,
    mol_column: str = "mol",
    activity_column: str = "active",
    pop_size: int = 500,
    ngens: int = 100,
    swap_fraction: float = 0.1,
    simpd_descriptors: Optional[pd.DataFrame] = None,
    target_train_frac_active: float = -1,
    target_test_frac_active: float = -1,
    target_test_set_frac: float = 0.2,
    target_delta_test_frac_active: Optional[float] = None,
    target_GF_delta_window: Tuple[int, int] = (10, 30),
    target_G_val: int = 70,
    max_population_cluster_entropy: float = 0.9,
    pareto_weight_GF_delta: float = 10,
    pareto_weight_G: float = 5,
    num_threads: int = 1,
    random_seed: Optional[int] = 19,
    verbose: bool = True,
    verbose_pymoo: bool = True,
    progress: bool = True,
    progress_leave: bool = False,
)

---

splito.simpd.DEFAULT_SIMPD_DESCRIPTORS module-attribute

DEFAULT_SIMPD_DESCRIPTORS = pd.DataFrame(
    [
        {
            "name": "SA_Score",
            "function": "datamol.descriptors.sas",
            "target_delta_value": 0.1 * 2.8,
        },
        {
            "name": "HeavyAtomCount",
            "function": "datamol.descriptors.n_heavy_atoms",
            "target_delta_value": 0.1 * 31,
        },
        {
            "name": "TPSA",
            "function": "datamol.descriptors.tpsa",
            "target_delta_value": 0.15 * 88.0,
        },
        {
            "name": "fr_benzene/1000 HeavyAtoms",
            "function": "splito.simpd.descriptors.fr_benzene_1000_heavy_atoms_count",
            "target_delta_value": -0.2 * 0.44,
        },
    ]
)