test_chemspace

Functions:

• test_pipeline_1linker_1rgroup_check_h_attachment –

  During multiple mergings, we want to make sure that

• test_evaluate_scoring_function_works –

  Ensure that the passed functional form is used.

• test_evaluate_scoring_function_saves_data –

  Ensure that the passed functional form is used.

• test_evaluate_full_hijack –

  Ensure that the passed functional form is used.

• test_al –

  Ensure that the passed functional form is used.

• test_al_local –

  Run a small active learning test.

• test_umap –

  Make a map of chemistry

• test_al_full –

  Run a small active learning test.

• test_al_manual_gp –
• test_adding_enamines –

  Ensure that the passed functional form is used.

test_pipeline_1linker_1rgroup_check_h_attachment

test_pipeline_1linker_1rgroup_check_h_attachment(sars_scaffold_chunk_sdf)

During multiple mergings, we want to make sure that it is the hydrogen on the scaffold that is returned back to the dataframe

:param sars_scaffold_chunk_sdf: :return:

Source code in fegrow/testing/test_chemspace.py

def test_pipeline_1linker_1rgroup_check_h_attachment(sars_scaffold_chunk_sdf):
    """
    During multiple mergings, we want to make sure that
    it is the hydrogen on the scaffold that is returned back to the dataframe

    :param sars_scaffold_chunk_sdf:
    :return:
    """
    chemspace = ChemSpace()
    hydrogen = 8
    chemspace.add_scaffold(sars_scaffold_chunk_sdf, hydrogen)

    r_methanol = Chem.MolFromSmiles("*CO")
    linker = Chem.MolFromSmiles("[*:0]NC[*:1]")
    chemspace.add_rgroups(linker, r_methanol)

    assert chemspace.df.loc[0].h == hydrogen

test_evaluate_scoring_function_works

test_evaluate_scoring_function_works(RGroups, sars_scaffold_chunk_sdf, rec_7l10_final_path)

Ensure that the passed functional form is used.

:param RGroups: :param sars_scaffold_chunk_sdf: :param rec_7l10_final_path: :return:

Source code in fegrow/testing/test_chemspace.py

def test_evaluate_scoring_function_works(
    RGroups, sars_scaffold_chunk_sdf, rec_7l10_final_path
):
    """
    Ensure that the passed functional form is used.

    :param RGroups:
    :param sars_scaffold_chunk_sdf:
    :param rec_7l10_final_path:
    :return:
    """

    # check if two molecules were built with chemspace
    chemspace = ChemSpace()

    chemspace.add_scaffold(sars_scaffold_chunk_sdf, 8)
    chemspace.add_smiles(["[H]OC([H])([H])C([H])([H])c1c([H])nc([H])c([H])c1[H]"])
    chemspace.add_protein(rec_7l10_final_path)

    random_score = random.random()

    def scorer(rmol, pdb_filename, data):
        return random_score

    chemspace.evaluate([0], scoring_function=scorer, skip_optimisation=True)

    assert chemspace.df.iloc[0].score == random_score

test_evaluate_scoring_function_saves_data

test_evaluate_scoring_function_saves_data(RGroups, sars_scaffold_chunk_sdf, rec_7l10_final_path)

Ensure that the passed functional form is used.

:param RGroups: :param sars_scaffold_chunk_sdf: :param rec_7l10_final_path: :return:

Source code in fegrow/testing/test_chemspace.py

def test_evaluate_scoring_function_saves_data(
    RGroups, sars_scaffold_chunk_sdf, rec_7l10_final_path
):
    """
    Ensure that the passed functional form is used.

    :param RGroups:
    :param sars_scaffold_chunk_sdf:
    :param rec_7l10_final_path:
    :return:
    """

    # check if two molecules were built with chemspace
    chemspace = ChemSpace()

    chemspace.add_scaffold(sars_scaffold_chunk_sdf, 8)
    chemspace.add_smiles(["[H]OC([H])([H])C([H])([H])c1c([H])nc([H])c([H])c1[H]"])
    chemspace.add_protein(rec_7l10_final_path)

    hello_world = "Hi Frank!"

    def scorer(rmol, pdb_filename, data):
        data["hello_world"] = hello_world
        return 5

    chemspace.evaluate([0], scoring_function=scorer, skip_optimisation=True)

    assert chemspace.df.iloc[0].Mol.GetProp("hello_world") == hello_world

test_evaluate_full_hijack

test_evaluate_full_hijack(RGroups, sars_scaffold_chunk_sdf, rec_7l10_final_path)

Ensure that the passed functional form is used.

:param RGroups: :param sars_scaffold_chunk_sdf: :param rec_7l10_final_path: :return:

Source code in fegrow/testing/test_chemspace.py

def test_evaluate_full_hijack(RGroups, sars_scaffold_chunk_sdf, rec_7l10_final_path):
    """
    Ensure that the passed functional form is used.

    :param RGroups:
    :param sars_scaffold_chunk_sdf:
    :param rec_7l10_final_path:
    :return:
    """

    # check if two molecules were built with chemspace
    chemspace = ChemSpace()

    chemspace.add_scaffold(sars_scaffold_chunk_sdf, 8)
    chemspace.add_smiles(["[H]OC([H])([H])C([H])([H])c1c([H])nc([H])c([H])c1[H]"])
    chemspace.add_protein(rec_7l10_final_path)

    def full_evaluation(scaffold, h, smiles, pdb_filename, *args, **kwargs):
        # return: mol, data
        mol = copy.deepcopy(scaffold)
        return mol, {"score": 5}

    chemspace.evaluate([0], full_evaluation=full_evaluation)

    assert chemspace.df.iloc[0].score == 5

test_al

test_al(RGroups, sars_scaffold_chunk_sdf, rec_7l10_final_path)

Ensure that the passed functional form is used.

:param RGroups: :param sars_scaffold_chunk_sdf: :param rec_7l10_final_path: :return:

Source code in fegrow/testing/test_chemspace.py

def test_al(RGroups, sars_scaffold_chunk_sdf, rec_7l10_final_path):
    """
    Ensure that the passed functional form is used.

    :param RGroups:
    :param sars_scaffold_chunk_sdf:
    :param rec_7l10_final_path:
    :return:
    """

    # check if two molecules were built with chemspace
    chemspace = ChemSpace()

    chemspace.add_scaffold(sars_scaffold_chunk_sdf, 8)

    not_studied_smiles = [
        "[H]OC(=O)N([H])c1c([H])nc([H])c([H])c1[H]",
        "[H]ON([H])c1c([H])nc([H])c([H])c1[H]",
    ]
    studied_smiles = [
        "[H]OC([H])([H])c1c([H])nc([H])c([H])c1[H]",
        "[H]ON([H])C(=O)c1c([H])nc([H])c([H])c1[H]",
    ]
    chemspace.add_smiles(studied_smiles + not_studied_smiles)
    chemspace.add_protein(rec_7l10_final_path)

    # set the results for the studied smiles
    df = chemspace.df
    df.loc[df.index == 0, ["score", "Training"]] = [3.2475, True]
    df.loc[df.index == 1, ["score", "Training"]] = [3.57196, True]

    to_study = chemspace.active_learning(n=1)

    assert to_study.iloc[0].Smiles in not_studied_smiles

test_al_local

test_al_local(sars_scaffold_chunk_sdf, rec_7l10_final_path)

Run a small active learning test.

Source code in fegrow/testing/test_chemspace.py

def test_al_local(sars_scaffold_chunk_sdf, rec_7l10_final_path):
    """
    Run a small active learning test.
    """

    # check if two molecules were built with chemspace
    chemspace = ChemSpace()

    scaffold = Chem.SDMolSupplier(str(root / "data/5R83_core.sdf"), removeHs=False)[0]
    chemspace.add_scaffold(scaffold, 6)

    oracle = pandas.read_csv(root / "data/cs50k_scored49578_unique47710.csv.zip")

    # separate the Smiles to be scanned
    smiles_list = oracle.Smiles.to_list()[:40]
    chemspace.add_smiles(smiles_list, h=6)

    # the protein here does not matter as we don't use it anyway
    chemspace.add_protein(rec_7l10_final_path)

    def oracle_look_up(scaffold, h, smiles, *args, **kwargs):
        # mol, data
        return None, {"score": oracle[oracle.Smiles == smiles].iloc[0].cnnaffinity}

    # select random molecules
    random_pics = chemspace.active_learning(n=5, first_random=True)
    chemspace.evaluate(random_pics, full_evaluation=oracle_look_up)

    # set the results for the studied smiles
    for i in range(2):
        picks = chemspace.active_learning(n=5)
        res = chemspace.evaluate(picks, full_evaluation=oracle_look_up)
        assert len(res) == 5

        # filter out the penalties
        res = res[res.score != 0]
        print(
            f"AL cycle cnnaffinity. Mean: {res.score.mean():.2f}, Min: {res.score.min():.2f}, Max: {res.score.max():.2f}"
        )

test_umap

test_umap(sars_scaffold_chunk_sdf)

Make a map of chemistry

Source code in fegrow/testing/test_chemspace.py

@pytest.mark.skip(reason="in dev. ")
def test_umap(sars_scaffold_chunk_sdf):
    """
    Make a map of chemistry
    """
    # check if two molecules were built with chemspace
    chemspace = ChemSpace()

    scaffold = Chem.SDMolSupplier(str(root / "data/5R83_core.sdf"), removeHs=False)[0]
    chemspace.add_scaffold(scaffold, 6)

    oracle = pandas.read_csv(root / "data/cs50k_scored49578_unique47710.csv.zip")
    smiles_list = oracle.Smiles.to_list()[:20]
    chemspace.add_smiles(smiles_list, h=6)

    chemspace.umap()

test_al_full

test_al_full(RGroups, sars_scaffold_chunk_sdf, rec_7l10_final_path)

Run a small active learning test.

Source code in fegrow/testing/test_chemspace.py

def test_al_full(RGroups, sars_scaffold_chunk_sdf, rec_7l10_final_path):
    """
    Run a small active learning test.
    """

    # check if two molecules were built with chemspace
    chemspace = ChemSpace()

    scaffold = Chem.SDMolSupplier(str(root / "data/5R83_core.sdf"), removeHs=False)[0]
    chemspace.add_scaffold(scaffold, 6)

    oracle = pandas.read_csv(root / "data/cs50k_scored49578_unique47710.csv.zip")

    # separate the Smiles to be scanned
    chemspace.add_smiles(oracle.Smiles.to_list()[:10], h=6)

    # the protein here does not matter as we don't use it anyway
    chemspace.add_protein(rec_7l10_final_path)

    def oracle_look_up(scaffold, h, smiles, *args, **kwargs):
        # mol, data
        return Chem.MolFromSmiles(smiles), {
            "score": oracle[oracle.Smiles == smiles].iloc[0].cnnaffinity
        }

    # select random molecules
    random_pics = chemspace.active_learning(n=3, first_random=True)
    chemspace.evaluate(random_pics, full_evaluation=oracle_look_up)

    assert chemspace.df.score.count() == 3
    assert all(~chemspace.df.loc[random_pics.index].score.isna())

    # compute all
    chemspace.evaluate(full_evaluation=oracle_look_up)
    assert chemspace.df.score[chemspace.df.score.isna()].count() == 0

test_al_manual_gp

test_al_manual_gp(RGroups, sars_scaffold_chunk_sdf, rec_7l10_final_path)

Source code in fegrow/testing/test_chemspace.py

def test_al_manual_gp(RGroups, sars_scaffold_chunk_sdf, rec_7l10_final_path):
    """ """
    # check if two molecules were built with chemspace
    chemspace = ChemSpace()

    scaffold = Chem.SDMolSupplier(str(root / "data/5R83_core.sdf"), removeHs=False)[0]
    chemspace.add_scaffold(scaffold, 6)

    oracle = pandas.read_csv(root / "data/cs50k_scored49578_unique47710.csv.zip")

    # separate the Smiles to be scanned
    smiles_list = oracle.Smiles.to_list()[:50]
    chemspace.add_smiles(smiles_list, h=6)

    # the protein here does not matter as we don't use it anyway
    chemspace.add_protein(rec_7l10_final_path)

    def oracle_look_up(scaffold, h, smiles, *args, **kwargs):
        # mol, data
        return None, {"score": oracle[oracle.Smiles == smiles].iloc[0].cnnaffinity}

    # select random molecules
    random_pics = chemspace.active_learning(n=5, first_random=True)
    chemspace.evaluate(random_pics, full_evaluation=oracle_look_up)

    # configure active learning
    from fegrow.al import Model, Query

    chemspace.model = Model.gaussian_process()

    chemspace.query = Query.UCB(beta=10)
    picks = chemspace.active_learning(n=5)
    chemspace.evaluate(picks, full_evaluation=oracle_look_up)

    # another go without changing any settings
    picks = chemspace.active_learning(n=5)
    chemspace.evaluate(picks, full_evaluation=oracle_look_up)

    # use every querrying strategy
    chemspace.query = Query.Greedy()
    picks = chemspace.active_learning(n=5)
    chemspace.evaluate(picks, full_evaluation=oracle_look_up)

    chemspace.query = Query.EI(tradeoff=0.1)
    picks = chemspace.active_learning(n=5)
    chemspace.evaluate(picks, full_evaluation=oracle_look_up)

    chemspace.query = Query.PI(tradeoff=0.1)
    picks = chemspace.active_learning(n=5)
    chemspace.evaluate(picks, full_evaluation=oracle_look_up)

    chemspace.model = Model.linear()
    chemspace.query = Query.Greedy()
    picks = chemspace.active_learning(n=5)
    chemspace.evaluate(picks, full_evaluation=oracle_look_up)

test_adding_enamines

test_adding_enamines(RGroups, sars_scaffold_chunk_sdf, rec_7l10_final_path)

Ensure that the passed functional form is used.

:param RGroups: :param sars_scaffold_chunk_sdf: :param rec_7l10_final_path: :return:

Source code in fegrow/testing/test_chemspace.py

@pytest.mark.skip(reason="requires the pydockingorg interface. ")
def test_adding_enamines(RGroups, sars_scaffold_chunk_sdf, rec_7l10_final_path):
    """
    Ensure that the passed functional form is used.

    :param RGroups:
    :param sars_scaffold_chunk_sdf:
    :param rec_7l10_final_path:
    :return:
    """

    # check if two molecules were built with chemspace
    chemspace = ChemSpace()

    chemspace.add_scaffold(sars_scaffold_chunk_sdf, 8)
    chemspace.add_smiles(["[H]OC([H])([H])C([H])([H])c1c([H])nc([H])c([H])c1[H]"], h=8)
    chemspace.add_protein(rec_7l10_final_path)

    def scorer(rmol, pdb_filename, data):
        return 5

    chemspace.evaluate([0], scoring_function=scorer, skip_optimisation=True)
    assert len(chemspace) == 1

    chemspace.add_enamine_molecules(results_per_search=10)

    # at least one extra one must have made it
    assert len(chemspace) > 1