From 0329d24e0a5a949309ba7c2d7f46485ac0e6be72 Mon Sep 17 00:00:00 2001 From: Alon Grinberg Dana Date: Sun, 15 Dec 2024 09:40:04 +0200 Subject: [PATCH] engine tests mods --- arc/mapping/engine_test.py | 922 ++++++++++++++----------------------- 1 file changed, 350 insertions(+), 572 deletions(-) diff --git a/arc/mapping/engine_test.py b/arc/mapping/engine_test.py index 8952a15bf5..c6d2917546 100644 --- a/arc/mapping/engine_test.py +++ b/arc/mapping/engine_test.py @@ -10,10 +10,9 @@ from random import shuffle import itertools -from arc.common import _check_r_n_p_symbols_between_rmg_and_arc_rxns +from arc.family import get_reaction_family_products from arc.mapping.engine import * from arc.reaction import ARCReaction - from arc.species.vectors import calculate_dihedral_angle from qcelemental.models.molecule import Molecule as QCMolecule @@ -34,67 +33,62 @@ def setUpClass(cls): """ cls.maxDiff = None - smiles = ['CC(C)F', '[CH3]', 'C[CH](C)', 'CF'] - - r_1_1_xyz = {'symbols': ('C', 'C', 'C', 'F', 'H', 'H', 'H', 'H', 'H', 'H', 'H'), - 'isotopes': (12, 12, 12, 19, 1, 1, 1, 1, 1, 1, 1), - 'coords': ((1.2509680857915237, 0.00832885083067477, -0.28594855682006387), - (-0.08450322338173592, -0.5786110309038947, 0.12835305965368538), - (-1.196883483105121, 0.4516770584363101, 0.10106807955582568), - (0.03212452836861426, -1.0465351442062332, 1.402047416169314), - (1.2170230403876368, 0.39373449465586885, -1.309310880313081), - (1.5446944155971303, 0.8206316657310906, 0.38700047363833845), - (2.0327466889922805, -0.7555292157466509, -0.22527487012253536), - (-0.3397419937928473, -1.4280299782557704, -0.5129583662636836), - (-0.9791793765226446, 1.2777482351478369, 0.786037216866474), - (-1.340583396165929, 0.8569620299504027, -0.9049411765144166), - (-2.1366652861689137, -0.00037696563964776297, 0.43392760415012316))} - - r_2_1_xyz = {'symbols': ('C', 'H', 'H', 'H'), - 'isotopes': (12, 1, 1, 1), - 'coords' : ((3.3746019998564553e-09, 5.828827384106545e-09, -4.859105107686622e-09), - (1.0669051052331406, -0.17519582095514982, 0.05416492980439295), - (-0.6853171627400634, -0.8375353626879753, -0.028085652887100996), - (-0.3815879458676787, 1.0127311778142964, -0.026079272058187608))} - - p_1_1_xyz = {'symbols': ('C', 'C', 'C', 'H', 'H', 'H', 'H', 'H', 'H', 'H'), - 'isotopes': (12, 12, 12, 1, 1, 1, 1, 1, 1, 1), - 'coords': ((-1.288730238258946, 0.06292843803165035, 0.10889818910854648), - (0.01096160773224897, -0.45756396262445836, -0.3934214957819532), - (1.2841030977199492, 0.11324607936811129, 0.12206176848573647), - (-1.4984446521053447, 1.0458196461796345, -0.3223873567509909), - (-1.2824724918369017, 0.14649429503996203, 1.1995362776757934), - (-2.098384694966955, -0.616646552269074, -0.17318515188247927), - (0.027360233461550892, -1.0601383387124987, -1.2952225290380646), - (2.122551165381095, -0.534098313164123, -0.15158596254231563), - (1.2634262459696732, 0.19628891975881263, 1.2125616721427255), - (1.4596297269035956, 1.1036697883919826, -0.307255411416999))} - - p_2_1_xyz = {'symbols': ('C', 'F', 'H', 'H', 'H'), - 'isotopes': (12, 19, 1, 1, 1), - 'coords': ((-0.060384822736851786, 0.004838867136375763, -0.004814368798794687), - (1.2877092002693546, -0.10318918150563985, 0.10266661058725791), - (-0.2965861926821434, 0.9189121874074381, -0.5532990701789506), - (-0.44047773762823295, -0.8660709320146035, -0.5425894744224189), - (-0.49026044722212864, 0.04550905897643097, 0.9980363028129072))} - cls.r_1 = ARCSpecies(label='r1', smiles=smiles[0],xyz=r_1_1_xyz ) - cls.r_2 = ARCSpecies(label='r2', smiles=smiles[1],xyz=r_2_1_xyz) - cls.p_1 = ARCSpecies(label='p1', smiles=smiles[2],xyz=p_1_1_xyz) - cls.p_2 = ARCSpecies(label='p2', smiles=smiles[3],xyz=p_2_1_xyz) - - cls.r_1_1 = ARCSpecies(label='r1', smiles=smiles[0],xyz=r_1_1_xyz ) - cls.r_2_2 = ARCSpecies(label='r2', smiles=smiles[1],xyz=r_2_1_xyz) - cls.p_1_1 = ARCSpecies(label='p1', smiles=smiles[2],xyz=p_1_1_xyz) - cls.p_2_2 = ARCSpecies(label='p2', smiles=smiles[3],xyz=p_2_1_xyz) - + smiles = ['CC(=O)OCCN', '[CH3]', 'CC(=O)O[CH]CN', 'C'] + r_1_xyz = {'symbols': ('C', 'C', 'O', 'O', 'C', 'C', 'N', 'H', 'H', 'H', 'H', 'H', 'H', 'H', 'H', 'H'), + 'isotopes': (12, 12, 16, 16, 12, 12, 14, 1, 1, 1, 1, 1, 1, 1, 1, 1), + 'coords': ((-2.5517015542914723, 0.6374264685195954, 0.5845070736702861), + (-1.4280800432880785, -0.3472086949468355, 0.47616163717754817), + (-1.2629104850172335, -1.1062502734566833, -0.4679405726175041), + (-0.6371651037701046, -0.2730498235608507, 1.5802001161064552), + (0.49856500321975944, -1.1589301501340519, 1.588720559767646), + (0.09655510699468367, -2.568185592214156, 2.0411269768682176), + (-0.37892831034665264, -2.5735393338188803, 3.4281685946865346), + (-3.1735891999289865, 0.3958275025838268, 1.4500661140112607), + (-3.1697515606096327, 0.5825623462025715, -0.31624209681478366), + (-2.1504064801268794, 1.6504763745435975, 0.668903119756822), + (0.9698293410095681, -1.1864028825350514, 0.5988445097419208), + (1.2358228352693132, -0.7323092496862488, 2.278167210048777), + (0.9587848877023767, -3.2377539733728216, 1.9623392312763976), + (-0.7000074380064817, -2.961851079966572, 1.4018157719346724), + (-1.13845885143463, -1.8982400428921975, 3.519262933330831), + (-0.7771636325627056, -3.486900524639466, 3.642537332438968))} + r_2_xyz = {'symbols': ('C', 'H', 'H', 'H'), 'isotopes': (12, 1, 1, 1), + 'coords': ((3.3746019998564553e-09, 5.828827384106545e-09, -4.859105107686622e-09), + (1.0669051052331406, -0.17519582095514982, 0.05416492980439295), + (-0.6853171627400634, -0.8375353626879753, -0.028085652887100996), + (-0.3815879458676787, 1.0127311778142964, -0.026079272058187608))} + + p_1_xyz = {'symbols': ('C', 'C', 'O', 'O', 'C', 'C', 'N', 'H', 'H', 'H', 'H', 'H', 'H', 'H', 'H'), + 'isotopes': (12, 12, 16, 16, 12, 12, 14, 1, 1, 1, 1, 1, 1, 1, 1), + 'coords': ((-2.4126223050522957, 1.1184165035580413, -0.4319265772396006), + (-1.199997719487665, 0.3980835599534999, -0.9414454869911179), + (-0.8169074822909718, -0.6841962780182634, -0.516655667102441), + (-0.6161907496541585, 1.1440411378816635, -1.935905159237574), + (0.5323812182950258, 0.582654338702519, -2.503115024614697), + (1.1731175616738767, 1.2826250843604319, -3.644099004671534), + (0.19283452563266232, 1.8556238352975616, -4.553940859324175), + (-2.1214717043196956, 2.0861598326389514, -0.015875768583935898), + (-3.1344078701345897, 1.2483499240565328, -1.2422145580831452), + (-2.8804146682085983, 0.5255941351523781, 0.35936223439908505), + (1.0768000501676995, -0.09302054943485703, -1.8569589770355146), + (1.8015757121392906, 0.5761935167897022, -4.195562348469085), + (1.827573441239823, 2.0744996813087657, -3.2662834886285643), + (-0.4057151405101191, 1.1095262327310491, -4.904943644873245), + (-0.41639306744224763, 2.4770955316170746, -4.023808770383508))} + cls.ch4_xyz = {'symbols': ('C', 'H', 'H', 'H', 'H'), 'isotopes': (12, 1, 1, 1, 1), + 'coords': ((-5.45906343962835e-10, 4.233517924761169e-10, 2.9505240956083194e-10), + (-0.6505520089868748, -0.7742801979689132, -0.4125187934483119), + (-0.34927557824779626, 0.9815958255612931, -0.3276823191685369), + (-0.022337921721882443, -0.04887374527620588, 1.0908766524267022), + (1.0221655095024578, -0.15844188273952128, -0.350675540104908))} + cls.r_1 = ARCSpecies(label='r1', smiles=smiles[0], xyz=r_1_xyz) + cls.r_2 = ARCSpecies(label='r2', smiles=smiles[1], xyz=r_2_xyz) + cls.p_1 = ARCSpecies(label='p1', smiles=smiles[2], xyz=p_1_xyz) + cls.p_2 = ARCSpecies(label='p2', smiles=smiles[3], xyz=cls.ch4_xyz) cls.rxn_1 = ARCReaction(r_species=[cls.r_1, cls.r_2], p_species=[cls.p_1, cls.p_2]) - cls.rmg_reactions_rxn_1 = get_rmg_reactions_from_arc_reaction(arc_reaction=cls.rxn_1, backend="ARC") - cls.r_label_dict_rxn_1, cls.p_label_dict_rxn_1 = ( - get_atom_indices_of_labeled_atoms_in_an_rmg_reaction(arc_reaction=cls.rxn_1, - rmg_reaction=cls.rmg_reactions_rxn_1[0])) - cls.spc1 = ARCSpecies(label="Test_is_isomorphic_1",smiles="C(CO)CC") - cls.spc2 = ARCSpecies(label="Test_is_isomorphic_2",smiles="OCCCC") + cls.spc1 = ARCSpecies(label="Test_is_isomorphic_1", smiles="C(CO)CC") + cls.spc2 = ARCSpecies(label="Test_is_isomorphic_2", smiles="OCCCC") cls.c4h9o_xyz = {'symbols': ('C', 'H', 'H', 'C', 'C', 'C', 'O', 'H', 'H', 'H', 'H', 'H', 'H', 'H'), 'isotopes': (12, 1, 1, 12, 12, 12, 16, 1, 1, 1, 1, 1, 1, 1), 'coords': ((0.025711531222639566, 1.5002469234994276, -0.018809721320361607), @@ -113,73 +107,46 @@ def setUpClass(cls): (2.377638769033351, 0.43380253822255727, 0.17647842348371048))} cls.spc1_dihedral_deviation = ARCSpecies(label='[CH2]C(C)CO_a', smiles='[CH2]C(C)CO', xyz=cls.c4h9o_xyz) cls.spc2_dihedral_deviation = ARCSpecies(label='[CH2]C(C)CO_b', smiles='[CH2]C(C)CO', - xyz={'symbols': ('C', 'C', 'C', 'C', 'O', 'H', 'H', 'H', 'H', 'H', 'H', 'H', 'H', 'H'), - 'isotopes': (12, 12, 12, 12, 16, 1, 1, 1, 1, 1, 1, 1, 1, 1), - 'coords': ((-1.3857811794963277, -1.3882629357157468, -0.09505562903985151), - (-0.48149615440373633, -0.18843419821506932, -0.36867730403761334), - (-1.1615061896768615, 1.1047002102194075, 0.08616180242702906), - (0.8755815877704686, -0.37530244696805926, 0.3151370087166933), - (1.7499930104893404, 0.685479589154504, -0.04657790660423845), - (-1.5824305690669607, -1.5021839148592626, 0.9764743462618697), - (-0.9236829644275987, -2.313486599576571, -0.4551984262103633), - (-0.31894490259166897, -0.11004093787895164, -1.45123483619259), - (-0.551069637667873, 1.9799864363130495, -0.1585701723917383), - (-2.1300920179099943, 1.2282601298258158, -0.4100421867371722), - (-1.3337649482883558, 1.102945655452365, 1.1678836912458532), - (0.7709290243761263, -0.38422053705817527, 1.4054470816682596), - (1.337910696892115, -1.3171321272490044, 0.001256204546378134), - (2.595292874972368, 0.5254618254772234, 0.4066018700054956))}) - - - cls.ch4_xyz = {'symbols': ('C', 'H', 'H', 'H', 'H'), 'isotopes': (12, 1, 1, 1, 1), - 'coords': ((-5.45906343962835e-10, 4.233517924761169e-10, 2.9505240956083194e-10), - (-0.6505520089868748, -0.7742801979689132, -0.4125187934483119), - (-0.34927557824779626, 0.9815958255612931, -0.3276823191685369), - (-0.022337921721882443, -0.04887374527620588, 1.0908766524267022), - (1.0221655095024578, -0.15844188273952128, -0.350675540104908))} - - cls.ch3cl_xyz = { - "symbols": ("Cl", "C", "H", "H", "H"), - "isotopes": (35, 12, 1, 1, 1), - "coords": ( - (1.6260450732924123, 0.06950511145502204, -0.029287400909487706), - (-0.13902319255720844, -0.005942522549225256, 0.0025040105508790877), - (-0.48751519000626353, -0.5782116958350602, -0.8600292104425608), - (-0.45894137315516464, -0.4942789247294056, 0.9255869621295756), - (-0.540565317573775, 1.0089280316586664, -0.038774361328407066), - ), - } - - cls.ch3_xyz_2 = { - "symbols": ("C", "H", "H", "H"), - "isotopes": (12, 1, 1, 1), - "coords": ( - (3.3746019998564553e-09, 5.828827384106545e-09, -4.859105107686622e-09), - (1.0669051052331406, -0.17519582095514982, 0.05416492980439295), - (-0.6853171627400634, -0.8375353626879753, -0.028085652887100996), - (-0.3815879458676787, 1.0127311778142964, -0.026079272058187608), - ), - } + xyz={'symbols': ( + 'C', 'C', 'C', 'C', 'O', 'H', 'H', 'H', 'H', 'H', 'H', 'H', 'H', 'H'), + 'isotopes': (12, 12, 12, 12, 16, 1, 1, 1, 1, 1, 1, 1, 1, 1), + 'coords': ( + (-1.3857811794963277, -1.3882629357157468, -0.09505562903985151), + ( + -0.48149615440373633, -0.18843419821506932, -0.36867730403761334), + (-1.1615061896768615, 1.1047002102194075, 0.08616180242702906), + (0.8755815877704686, -0.37530244696805926, 0.3151370087166933), + (1.7499930104893404, 0.685479589154504, -0.04657790660423845), + (-1.5824305690669607, -1.5021839148592626, 0.9764743462618697), + (-0.9236829644275987, -2.313486599576571, -0.4551984262103633), + (-0.31894490259166897, -0.11004093787895164, -1.45123483619259), + (-0.551069637667873, 1.9799864363130495, -0.1585701723917383), + (-2.1300920179099943, 1.2282601298258158, -0.4100421867371722), + (-1.3337649482883558, 1.102945655452365, 1.1678836912458532), + (0.7709290243761263, -0.38422053705817527, 1.4054470816682596), + (1.337910696892115, -1.3171321272490044, 0.001256204546378134), + (2.595292874972368, 0.5254618254772234, 0.4066018700054956))}) + + cls.ch3cl_xyz = {"symbols": ("Cl", "C", "H", "H", "H"), + "isotopes": (35, 12, 1, 1, 1), + "coords": ((1.6260450732924123, 0.06950511145502204, -0.029287400909487706), + (-0.13902319255720844, -0.005942522549225256, 0.0025040105508790877), + (-0.48751519000626353, -0.5782116958350602, -0.8600292104425608), + (-0.45894137315516464, -0.4942789247294056, 0.9255869621295756), + (-0.540565317573775, 1.0089280316586664, -0.038774361328407066))} + + cls.ch3_xyz_2 = {"symbols": ("C", "H", "H", "H"), + "isotopes": (12, 1, 1, 1), + "coords": ((3.3746019998564553e-09, 5.828827384106545e-09, -4.859105107686622e-09), + (1.0669051052331406, -0.17519582095514982, 0.05416492980439295), + (-0.6853171627400634, -0.8375353626879753, -0.028085652887100996), + (-0.3815879458676787, 1.0127311778142964, -0.026079272058187608))} cls.h_rad_xyz = {"symbols": ("H",), "isotopes": (1,), "coords": ((0, 0, 0),)} - cls.hcl_xyz = { - "symbols": ("H", "Cl"), - "isotopes": (1, 35), - "coords": ((0.6878248644303301, 0.0, 0.0), (-0.6878248644303301, 0.0, 0.0)), - } - - cls.ch4_xyz = { - "symbols": ("C", "H", "H", "H", "H"), - "isotopes": (12, 1, 1, 1, 1), - "coords": ( - (-5.45906343962835e-10, 4.233517924761169e-10, 2.9505240956083194e-10), - (-0.6505520089868748, -0.7742801979689132, -0.4125187934483119), - (-0.34927557824779626, 0.9815958255612931, -0.3276823191685369), - (-0.022337921721882443, -0.04887374527620588, 1.0908766524267022), - (1.0221655095024578, -0.15844188273952128, -0.350675540104908), - ), - } + cls.hcl_xyz = {"symbols": ("H", "Cl"), + "isotopes": (1, 35), + "coords": ((0.6878248644303301, 0.0, 0.0), (-0.6878248644303301, 0.0, 0.0))} cls.ch4_xyz_diff_order = """H -0.65055201 -0.77428020 -0.41251879 H -0.34927558 0.98159583 -0.32768232 @@ -200,8 +167,8 @@ def setUpClass(cls): (-0.8942590, -0.8537420, 0.0000000)), 'isotopes': (16, 16, 1), 'symbols': ('O', 'O', 'H')} cls.nh2_xyz = """N 0.00022972 0.40059496 0.00000000 - H -0.83174214 -0.19982058 0.00000000 - H 0.83151242 -0.20077438 0.00000000""" + H -0.83174214 -0.19982058 0.00000000 + H 0.83151242 -0.20077438 0.00000000""" cls.n2h4_xyz = """N -0.67026921 -0.02117571 -0.25636419 N 0.64966276 0.05515705 0.30069593 H -1.27787600 0.74907557 0.03694453 @@ -536,383 +503,6 @@ def setUpClass(cls): H -1.04996634 -0.37234114 0.91874740 H 1.36260637 0.37153887 -0.86221771""" - - def test_assign_labels_to_products(self): - rxn_1_test = ARCReaction(r_species=[self.r_1, self.r_2], p_species=[self.p_1, self.p_2]) - assign_labels_to_products(rxn_1_test, self.p_label_dict_rxn_1) - index = 0 - for product in rxn_1_test.p_species: - for atom in product.mol.atoms: - if not isinstance(atom.label, str) or atom.label != "": - self.assertEqual(self.p_label_dict_rxn_1[atom.label], index) - index+=1 - - def test_inc_vals(self): - """Test creating an atom map via map_two_species() and incrementing all values""" - spc1 = ARCSpecies(label='CH4', smiles='C', xyz=self.ch4_xyz) - spc2 = ARCSpecies(label='CH4', smiles='C', xyz=self.ch4_xyz_diff_order) - atom_map = map_two_species(spc1, spc2, inc_vals=100) - self.assertEqual(atom_map, [102, 100, 101, 103, 104]) - - def test_label_species_atoms(self): - rxn_1_test = ARCReaction(r_species=[self.r_1, self.r_2], p_species=[self.p_1, self.p_2]) - assign_labels_to_products(rxn_1_test, self.p_label_dict_rxn_1) - - reactants, products = copy_species_list_for_mapping(rxn_1_test.r_species), copy_species_list_for_mapping(rxn_1_test.p_species) - - label_species_atoms(reactants) - label_species_atoms(products) - - index = 0 - for reactant in reactants: - for atom in reactant.mol.atoms: - self.assertEqual(atom.label,str(index)) - index +=1 - - index = 0 - for product in products: - for atom in product.mol.atoms: - self.assertEqual(atom.label,str(index)) - index +=1 - - def test_cut_species_based_on_atom_indices(self): - """test the cut_species_for_mapping function""" - rxn_1_test = ARCReaction(r_species=[self.r_1, self.r_2], p_species=[self.p_1, self.p_2]) - reactants, products = copy_species_list_for_mapping(rxn_1_test.r_species), copy_species_list_for_mapping(rxn_1_test.p_species) - label_species_atoms(reactants), label_species_atoms(products) - - r_bdes, p_bdes = find_all_bdes(rxn_1_test, self.r_label_dict_rxn_1, True), find_all_bdes(rxn_1_test, self.p_label_dict_rxn_1, False) - - r_cuts = cut_species_based_on_atom_indices(reactants, r_bdes) - p_cuts = cut_species_based_on_atom_indices(products, p_bdes) - - - self.assertIn("C[CH]C", [r_cut.mol.copy(deep=True).smiles for r_cut in r_cuts]) - self.assertIn("[F]", [r_cut.mol.copy(deep=True).smiles for r_cut in r_cuts]) - self.assertIn("[CH3]", [r_cut.mol.copy(deep=True).smiles for r_cut in r_cuts]) - self.assertIn("C[CH]C", [p_cut.mol.copy(deep=True).smiles for p_cut in p_cuts]) - self.assertIn("[F]", [p_cut.mol.copy(deep=True).smiles for p_cut in p_cuts]) - self.assertIn("[CH3]", [p_cut.mol.copy(deep=True).smiles for p_cut in p_cuts]) - - spc = ARCSpecies(label="test", smiles="CNC", bdes = [(1, 2), (2, 3)]) - for i, a in enumerate(spc.mol.atoms): - a.label=str(i) - cuts = cut_species_based_on_atom_indices([spc], [(1, 2), (2, 3)]) - self.assertEqual(len(cuts), 3) - for cut in cuts: - self.assertTrue(any([cut.mol.copy(deep=True).is_isomorphic(ARCSpecies(label="1", smiles="[CH3]").mol), - cut.mol.copy(deep=True).is_isomorphic(ARCSpecies(label="2", smiles="[NH]").mol)])) - - h2 = ARCSpecies(label="H2", smiles="[H][H]") - label_species_atoms([h2]) - - cuts = cut_species_based_on_atom_indices([h2], [(1, 2)]) - self.assertEqual(len(cuts), 2) - for cut in cuts: - self.assertEqual(cut.get_xyz()["symbols"], ('H',)) - - spcs = [ARCSpecies(label="r", smiles = 'O=C(O)CCF')] - label_species_atoms(spcs) - cuts = cut_species_based_on_atom_indices(spcs, [(6, 5), (4, 2), (3, 7)]) - self.assertEqual(len(cuts), 4) - - def test_r_cut_p_cut_isomorphic(self): - rxn_1_test = ARCReaction(r_species=[self.r_1, self.r_2], p_species=[self.p_1, self.p_2]) - reactants, products = copy_species_list_for_mapping(rxn_1_test.r_species), copy_species_list_for_mapping(rxn_1_test.p_species) - label_species_atoms(reactants), label_species_atoms(products) - - r_bdes, p_bdes = find_all_bdes(rxn_1_test, self.r_label_dict_rxn_1, True), find_all_bdes(rxn_1_test, self.p_label_dict_rxn_1, False) - - r_cuts = cut_species_based_on_atom_indices(reactants, r_bdes) - p_cuts = cut_species_based_on_atom_indices(products, p_bdes) - - self.assertTrue(r_cut_p_cut_isomorphic(self.spc1,self.spc2)) - for r_cut in r_cuts: - for p_cut in p_cuts: - if r_cut.mol.is_isomorphic(p_cut.mol): - p_cuts.remove(p_cut) - break - self.assertEqual(len(p_cuts),0) - self.assertTrue(r_cut_p_cut_isomorphic(ARCSpecies(label="r1", smiles="F[C]F", multiplicity=1), - ARCSpecies(label="r1", smiles="F[C]F", multiplicity=3))) - - def test_pairing_reactants_and_products_for_mapping(self): - smiles = ["[F]", "C[CH]C", "[CH3]"] - rxn_1_test = ARCReaction(r_species=[self.r_1, self.r_2], p_species=[self.p_1, self.p_2]) - reactants, products = copy_species_list_for_mapping(rxn_1_test.r_species), copy_species_list_for_mapping(rxn_1_test.p_species) - label_species_atoms(reactants), label_species_atoms(products) - - r_bdes, p_bdes = find_all_bdes(rxn_1_test, self.r_label_dict_rxn_1, True), find_all_bdes(rxn_1_test, self.p_label_dict_rxn_1, False) - - r_cuts = cut_species_based_on_atom_indices(reactants, r_bdes) - p_cuts = cut_species_based_on_atom_indices(products, p_bdes) - - pairs_of_reactant_and_products = pairing_reactants_and_products_for_mapping(r_cuts, p_cuts) - - for pair in pairs_of_reactant_and_products: - self.assertTrue(pair[0].mol.copy(deep=True).is_isomorphic(pair[1].mol.copy(deep=True))) - self.assertIn(str(pair[0].mol.copy(deep=True).to_smiles()), smiles) - smiles.remove(pair[0].mol.copy(deep=True).to_smiles()) - - def test_map_pairs(self): - rxn_1_test = ARCReaction(r_species=[self.r_1, self.r_2], p_species=[self.p_1, self.p_2]) - rmg_reactions = get_rmg_reactions_from_arc_reaction(arc_reaction=rxn_1_test, backend="ARC") - r_label_dict, p_label_dict = get_atom_indices_of_labeled_atoms_in_an_rmg_reaction(arc_reaction=rxn_1_test, - rmg_reaction=rmg_reactions[0]) - reactants, products = copy_species_list_for_mapping(rxn_1_test.r_species), copy_species_list_for_mapping(rxn_1_test.p_species) - label_species_atoms(reactants), label_species_atoms(products) - - r_bdes, p_bdes = find_all_bdes(rxn_1_test, r_label_dict, True), find_all_bdes(rxn_1_test, p_label_dict, False) - - r_cuts = cut_species_based_on_atom_indices(reactants, r_bdes) - p_cuts = cut_species_based_on_atom_indices(products, p_bdes) - pairs_of_reactant_and_products = pairing_reactants_and_products_for_mapping(r_cuts, p_cuts) - maps = map_pairs(pairs_of_reactant_and_products) - for map_ in maps: - if len(map_) == 1: - self.assertEqual(map_[0], 0) - elif len(map_) == 4: - self.assertEqual(map_[0], 0) - for i in map_[1:]: - self.assertIn(i, [1, 2, 3]) - self.assertEqual(len(np.unique(map_)), len(map_)) - else: - self.assertEqual(map_[:3], [0, 1, 2]) - self.assertIn(tuple(map_[3:6]), list(itertools.permutations([3, 4, 5]))) - self.assertEqual(map_[6], 6) - self.assertIn(tuple(map_[7:]), list(itertools.permutations([7, 8, 9]))) - self.assertEqual(len(np.unique(map_)), len(map_)) - - def test_glue_maps(self): - rxn_1_test = ARCReaction(r_species=[self.r_1, self.r_2], p_species=[self.p_1, self.p_2]) - rmg_reactions = get_rmg_reactions_from_arc_reaction(arc_reaction=rxn_1_test, backend="ARC") - r_label_dict, p_label_dict = get_atom_indices_of_labeled_atoms_in_an_rmg_reaction(arc_reaction=rxn_1_test, - rmg_reaction=rmg_reactions[0]) - assign_labels_to_products(rxn_1_test, p_label_dict) - reactants, products = copy_species_list_for_mapping(rxn_1_test.r_species), copy_species_list_for_mapping(rxn_1_test.p_species) - label_species_atoms(reactants), label_species_atoms(products) - - r_bdes, p_bdes = find_all_bdes(rxn_1_test, r_label_dict, True), find_all_bdes(rxn_1_test, p_label_dict, False) - - r_cuts = cut_species_based_on_atom_indices(reactants, r_bdes) - p_cuts = cut_species_based_on_atom_indices(products, p_bdes) - pairs_of_reactant_and_products = pairing_reactants_and_products_for_mapping(r_cuts, p_cuts) - maps = map_pairs(pairs_of_reactant_and_products) - atom_map = glue_maps(maps,pairs_of_reactant_and_products) - self.assertEqual(len(atom_map), self.r_1.mol.get_num_atoms() + self.r_2.mol.get_num_atoms()) - atoms_r = [atom for atom in self.r_1.mol.copy(deep=True).atoms] + [atom for atom in self.r_2.mol.copy(deep=True).atoms] - atoms_p = [atom for atom in self.p_1.mol.copy(deep=True).atoms] + [atom for atom in self.p_2.mol.copy(deep=True).atoms] - for index, value in enumerate(atom_map): - self.assertEqual(atoms_r[index].symbol, atoms_p[value].symbol) - - def test_get_atom_indices_of_labeled_atoms_in_an_rmg_reaction(self): - """Test the get_atom_indices_of_labeled_atoms_in_an_rmg_reaction() function.""" - rmg_reactions = get_rmg_reactions_from_arc_reaction(arc_reaction=self.arc_reaction_1) - r_dict, p_dict = get_atom_indices_of_labeled_atoms_in_an_rmg_reaction(arc_reaction=self.arc_reaction_1, - rmg_reaction=rmg_reactions[0]) - self.assertEqual(r_dict['*1'], 0) - self.assertIn(r_dict['*2'], [1, 2, 3, 4]) - self.assertEqual(r_dict['*3'], 5) - self.assertEqual(p_dict['*1'], 0) - self.assertIn(p_dict['*2'], [5, 6]) - self.assertEqual(p_dict['*3'], 4) - self.assertTrue(_check_r_n_p_symbols_between_rmg_and_arc_rxns(self.arc_reaction_1, rmg_reactions)) - - rmg_reactions = get_rmg_reactions_from_arc_reaction(arc_reaction=self.arc_reaction_2) - r_dict, p_dict = get_atom_indices_of_labeled_atoms_in_an_rmg_reaction(arc_reaction=self.arc_reaction_2, - rmg_reaction=rmg_reactions[0]) - self.assertIn(r_dict['*1'], [0, 2]) - self.assertIn(r_dict['*2'], [3, 4, 5, 8, 9, 10]) - self.assertEqual(r_dict['*3'], 11) - self.assertEqual(p_dict['*1'], 0) - self.assertIn(p_dict['*2'], [11, 12, 13]) - self.assertEqual(p_dict['*3'], 10) - self.assertTrue(_check_r_n_p_symbols_between_rmg_and_arc_rxns(self.arc_reaction_2, rmg_reactions)) - - rmg_reactions = get_rmg_reactions_from_arc_reaction(arc_reaction=self.arc_reaction_4) - r_dict, p_dict = get_atom_indices_of_labeled_atoms_in_an_rmg_reaction(arc_reaction=self.arc_reaction_4, - rmg_reaction=rmg_reactions[0]) - self.assertEqual(r_dict['*1'], 0) - self.assertEqual(r_dict['*2'], 2) - self.assertIn(r_dict['*3'], [7, 8]) - self.assertEqual(p_dict['*1'], 0) - self.assertEqual(p_dict['*2'], 2) - self.assertIn(p_dict['*3'], [3, 4, 5]) - self.assertTrue(_check_r_n_p_symbols_between_rmg_and_arc_rxns(self.arc_reaction_4, rmg_reactions)) - - for atom, symbol in zip(self.rxn_2a.r_species[0].mol.atoms, ['C', 'C', 'C', 'H', 'H', 'H', 'H', 'H', 'H', 'H']): - self.assertEqual(atom.symbol, symbol) - self.assertEqual(self.rxn_2a.r_species[0].mol.atoms[0].radical_electrons, 0) - self.assertEqual(self.rxn_2a.r_species[0].mol.atoms[1].radical_electrons, 1) - self.assertEqual(self.rxn_2a.r_species[0].mol.atoms[2].radical_electrons, 0) - self.assertEqual(self.rxn_2a.p_species[0].mol.atoms[0].radical_electrons, 0) - self.assertEqual(self.rxn_2a.p_species[0].mol.atoms[1].radical_electrons, 0) - self.assertEqual(self.rxn_2a.p_species[0].mol.atoms[2].radical_electrons, 1) - rmg_reactions = get_rmg_reactions_from_arc_reaction(arc_reaction=self.rxn_2a) - r_dict, p_dict = get_atom_indices_of_labeled_atoms_in_an_rmg_reaction(arc_reaction=self.rxn_2a, - rmg_reaction=rmg_reactions[0]) - self.assertEqual(r_dict['*1'], 1) - self.assertIn(r_dict['*2'], [0, 2]) - self.assertIn(r_dict['*3'], [4, 5, 6, 7, 8, 9]) - self.assertEqual(p_dict['*1'], 1) - self.assertEqual(p_dict['*2'], 2) - self.assertIn(p_dict['*3'], [3, 6]) - self.assertTrue(_check_r_n_p_symbols_between_rmg_and_arc_rxns(self.rxn_2a, rmg_reactions)) - - for atom, symbol in zip(self.rxn_2b.r_species[0].mol.atoms, ['C', 'C', 'H', 'H', 'H', 'H', 'C', 'H', 'H', 'H']): - self.assertEqual(atom.symbol, symbol) - rmg_reactions = get_rmg_reactions_from_arc_reaction(arc_reaction=self.rxn_2b) - r_dict, p_dict = get_atom_indices_of_labeled_atoms_in_an_rmg_reaction(arc_reaction=self.rxn_2b, - rmg_reaction=rmg_reactions[0]) - self.assertEqual(r_dict['*1'], 1) - self.assertIn(r_dict['*2'], [0, 6]) - self.assertIn(r_dict['*3'], [3, 4, 5, 7, 8, 9]) - self.assertEqual(p_dict['*1'], 1) - self.assertEqual(p_dict['*2'], 2) - self.assertIn(p_dict['*3'], [3, 6]) - self.assertTrue(_check_r_n_p_symbols_between_rmg_and_arc_rxns(self.rxn_2b, rmg_reactions)) - - # C3H6O + C4H9O <=> C3H5O + C4H10O - r_1 = ARCSpecies(label='C3H6O', smiles='CCC=O') - r_2 = ARCSpecies(label='C4H9O', smiles='[CH2]C(C)CO') - p_1 = ARCSpecies(label='C3H5O', smiles='C[CH]C=O') - p_2 = ARCSpecies(label='C4H10O', smiles='CC(C)CO') - rxn_1 = ARCReaction(reactants=['C3H6O', 'C4H9O'], products=['C3H5O', 'C4H10O'], - r_species=[r_1, r_2], p_species=[p_1, p_2]) - rmg_reactions = get_rmg_reactions_from_arc_reaction(arc_reaction=rxn_1) - for rmg_reaction in rmg_reactions: - r_dict, p_dict = get_atom_indices_of_labeled_atoms_in_an_rmg_reaction(arc_reaction=rxn_1, - rmg_reaction=rmg_reaction) - for d in [r_dict, p_dict]: - self.assertEqual(len(list(d.keys())), 3) - keys = list(d.keys()) - for label in ['*1', '*2', '*3']: - self.assertIn(label, keys) - self.assertTrue(_check_r_n_p_symbols_between_rmg_and_arc_rxns(rxn_1, rmg_reactions)) - - p_1 = ARCSpecies(label='C3H5O', smiles='CC=C[O]') # Use a wrong resonance structure and repeat the above. - rxn_2 = ARCReaction(reactants=['C3H6O', 'C4H9O'], products=['C3H5O', 'C4H10O'], - r_species=[r_1, r_2], p_species=[p_1, p_2]) - rmg_reactions = get_rmg_reactions_from_arc_reaction(arc_reaction=rxn_2) - for rmg_reaction in rmg_reactions: - r_dict, p_dict = get_atom_indices_of_labeled_atoms_in_an_rmg_reaction(arc_reaction=rxn_2, - rmg_reaction=rmg_reaction) - for d in [r_dict, p_dict]: - self.assertEqual(len(list(d.keys())), 3) - keys = list(d.keys()) - for label in ['*1', '*2', '*3']: - self.assertIn(label, keys) - self.assertTrue(_check_r_n_p_symbols_between_rmg_and_arc_rxns(rxn_2, rmg_reactions)) - - # C3H6O + C4H9O <=> C3H5O + C4H10O - r_1 = ARCSpecies(label='C3H6O', smiles='CCC=O', xyz=self.c3h6o_xyz) - r_2 = ARCSpecies(label='C4H9O', smiles='[CH2]C(C)CO', xyz=self.c4h9o_xyz) - p_1 = ARCSpecies(label='C3H5O', smiles='C[CH]C=O', xyz=self.c3h5o_xyz) - p_2 = ARCSpecies(label='C4H10O', smiles='CC(C)CO', xyz=self.c4h10o_xyz) - rxn_3 = ARCReaction(reactants=['C3H6O', 'C4H9O'], products=['C3H5O', 'C4H10O'], - r_species=[r_1, r_2], p_species=[p_1, p_2]) - rmg_reactions = get_rmg_reactions_from_arc_reaction(arc_reaction=rxn_3) - r_dict, p_dict = get_atom_indices_of_labeled_atoms_in_an_rmg_reaction(arc_reaction=rxn_3, - rmg_reaction=rmg_reactions[0]) - self.assertEqual(r_dict, {'*3': 10, '*1': 1, '*2': 7}) - self.assertEqual(p_dict, {'*1': 1, '*3': 9, '*2': 16}) - self.assertTrue(_check_r_n_p_symbols_between_rmg_and_arc_rxns(rxn_3, rmg_reactions)) - - - def test_map_arc_rmg_species(self): - """Test the map_arc_rmg_species() function.""" - r_map, p_map = map_arc_rmg_species(arc_reaction=ARCReaction(r_species=[ARCSpecies(label='CCjC', smiles='C[CH]C')], - p_species=[ARCSpecies(label='CjCC', smiles='[CH2]CC')]), - rmg_reaction=Reaction(reactants=[Species(smiles='C[CH]C')], - products=[Species(smiles='[CH2]CC')]), - concatenate=False) - self.assertEqual(r_map, {0: 0}) - self.assertEqual(p_map, {0: 0}) - - r_map, p_map = map_arc_rmg_species(arc_reaction=ARCReaction(r_species=[ARCSpecies(label='CCjC', smiles='C[CH]C')], - p_species=[ARCSpecies(label='CjCC', smiles='[CH2]CC')]), - rmg_reaction=Reaction(reactants=[Species(smiles='C[CH]C')], - products=[Species(smiles='[CH2]CC')])) - self.assertEqual(r_map, {0: [0]}) - self.assertEqual(p_map, {0: [0]}) - - r_map, p_map = map_arc_rmg_species(arc_reaction=self.arc_reaction_1, rmg_reaction=self.rmg_reaction_1) - self.assertEqual(r_map, {0: [0], 1: [1]}) - self.assertEqual(p_map, {0: [0], 1: [1]}) - - r_map, p_map = map_arc_rmg_species(arc_reaction=self.arc_reaction_1, rmg_reaction=self.rmg_reaction_2) - self.assertEqual(r_map, {0: [1], 1: [0]}) - self.assertEqual(p_map, {0: [0], 1: [1]}) - - r_map, p_map = map_arc_rmg_species(arc_reaction=self.arc_reaction_3, rmg_reaction=self.rmg_reaction_3) - self.assertEqual(r_map, {0: [0, 1], 1: [0, 1]}) - self.assertEqual(p_map, {0: [0]}) - - rmg_reaction_1 = Reaction(reactants=[Species(smiles='N'), Species(smiles='[H]')], - products=[Species(smiles='[NH2]'), Species(smiles='[H][H]')]) - rmg_reaction_2 = Reaction(reactants=[Species(smiles='[H]'), Species(smiles='N')], - products=[Species(smiles='[H][H]'), Species(smiles='[NH2]')]) - rmg_reaction_3 = Reaction(reactants=[Species(smiles='N'), Species(smiles='[H]')], - products=[Species(smiles='[H][H]'), Species(smiles='[NH2]')]) - arc_reaction = ARCReaction(r_species=[ARCSpecies(label='NH3', smiles='N'), ARCSpecies(label='H', smiles='[H]')], - p_species=[ARCSpecies(label='NH2', smiles='[NH2]'), ARCSpecies(label='H2', smiles='[H][H]')]) - - r_map, p_map = map_arc_rmg_species(rmg_reaction=rmg_reaction_1, arc_reaction=arc_reaction) - self.assertEqual(r_map, {0: [0], 1: [1]}) - self.assertEqual(p_map, {0: [0], 1: [1]}) - - r_map, p_map = map_arc_rmg_species(rmg_reaction=rmg_reaction_2, arc_reaction=arc_reaction) - self.assertEqual(r_map, {0: [1], 1: [0]}) - self.assertEqual(p_map, {0: [1], 1: [0]}) - - r_map, p_map = map_arc_rmg_species(rmg_reaction=rmg_reaction_3, arc_reaction=arc_reaction) - self.assertEqual(r_map, {0: [0], 1: [1]}) - self.assertEqual(p_map, {0: [1], 1: [0]}) - rmg_reaction = Reaction(reactants=[Species(smiles='[CH3]'), Species(smiles='[CH3]')], - products=[Species(smiles='CC')]) - arc_reaction = ARCReaction(r_species=[ARCSpecies(label='CH3', smiles='[CH3]'), - ARCSpecies(label='CH3', smiles='[CH3]')], - p_species=[ARCSpecies(label='C2H6', smiles='CC')]) - - r_map, p_map = map_arc_rmg_species(rmg_reaction=rmg_reaction, arc_reaction=arc_reaction) - self.assertEqual(r_map, {0: [0, 1], 1: [0, 1]}) - self.assertEqual(p_map, {0: [0]}) - - rmg_reaction = Reaction(reactants=[Species(smiles='c1ccccc1CCC=C'), Species(smiles='CCO[O]')], - products=[Species(smiles='c1ccccc1[CH]CC=C'), Species(smiles='CCOO')]) - arc_reaction = ARCReaction(r_species=[ARCSpecies(label='butenylbenzene', smiles='c1ccccc1CCC=C'), - ARCSpecies(label='peroxyl', smiles='CCO[O]')], - p_species=[ARCSpecies(label='peroxide', smiles='CCOO'), - ARCSpecies(label='butenylbenzene_rad', smiles='c1ccccc1[CH]CC=C')]) - r_map, p_map = map_arc_rmg_species(rmg_reaction=rmg_reaction, - arc_reaction=arc_reaction, - concatenate=False, - ) - self.assertEqual(r_map, {0: 0, 1: 1}) - self.assertEqual(p_map, {0: 1, 1: 0}) - - rmg_reaction = Reaction(reactants=[Species(smiles='[N]N'), Species(smiles='NN')], - products=[Species(smiles='[NH]N'), Species(smiles='[NH]N')]) - arc_reaction = ARCReaction(r_species=[ARCSpecies(label='H2NN(T)', smiles='[N]N'), - ARCSpecies(label='N2H4', smiles='NN')], - p_species=[ARCSpecies(label='N2H3', smiles='[NH]N'), - ARCSpecies(label='N2H3', smiles='[NH]N')]) - r_map, p_map = map_arc_rmg_species(rmg_reaction=rmg_reaction, - arc_reaction=arc_reaction, - concatenate=False, - ) - self.assertEqual(r_map, {0: 0, 1: 1}) - self.assertEqual(p_map, {0: 0, 1: 1}) - - - def test_find_equivalent_atoms_in_reactants_and_products(self): - """Test the find_equivalent_atoms_in_reactants_and_products() function""" - equivalence_map_1 = find_equivalent_atoms_in_reactants(arc_reaction=self.rxn_2a) - # Both C 0 and C 2 are equivalent, C 1 is unique, and H 4-9 are equivalent as well. - self.assertEqual(equivalence_map_1, [[0, 2], [1], [4, 5, 6, 7, 8, 9]]) - equivalence_map_2 = find_equivalent_atoms_in_reactants(arc_reaction=self.rxn_2b) - self.assertEqual(equivalence_map_2, [[0, 6], [1], [3, 4, 5, 7, 8, 9]]) - def test_map_two_species(self): """Test the map_two_species() function.""" # H @@ -1071,8 +661,15 @@ def test_map_two_species(self): atom_map = map_two_species(rmg_mol, rmg_spc) self.assertEqual(atom_map, [0, 1, 2, 3, 4]) + def test_inc_vals(self): + """Test creating an atom map via map_two_species() and incrementing all values""" + spc1 = ARCSpecies(label='CH4', smiles='C', xyz=self.ch4_xyz) + spc2 = ARCSpecies(label='CH4', smiles='C', xyz=self.ch4_xyz_diff_order) + atom_map = map_two_species(spc1, spc2, inc_vals=100) + self.assertEqual(atom_map, [102, 100, 101, 103, 104]) + def test_get_arc_species(self): - """Test the get_arc_species function.""" + """Test the get_arc_species() function.""" self.assertIsInstance(get_arc_species(ARCSpecies(label='S', smiles='C')), ARCSpecies) self.assertIsInstance(get_arc_species(Species(smiles='C')), ARCSpecies) self.assertIsInstance(get_arc_species(Molecule(smiles='C')), ARCSpecies) @@ -1098,7 +695,7 @@ def test_create_qc_mol(self): self.assertEqual(symbol, expected_symbol) def test_check_species_before_mapping(self): - """Test the check_species_before_mapping function.""" + """Test the check_species_before_mapping() function.""" self.assertFalse(check_species_before_mapping(spc_1=ARCSpecies(label='CH4', smiles='C'), spc_2=ARCSpecies(label='CH3', smiles='[CH3]'), verbose=True)) @@ -1122,9 +719,8 @@ def test_check_species_before_mapping(self): verbose=True)) self.assertTrue(check_species_before_mapping(spc_1=self.spc1, spc_2=self.spc2, verbose=True)) - def test_get_bonds_dict(self): - """Test the get_bonds_dict function.""" + """Test the get_bonds_dict() function.""" bond_dict = get_bonds_dict(spc=ARCSpecies(label='CH4', smiles='C')) self.assertEqual(bond_dict, {'C-H': 4}) bond_dict = get_bonds_dict(spc=self.ccjco) @@ -1135,7 +731,7 @@ def test_get_bonds_dict(self): self.assertEqual(bond_dict, {'C-C': 3, 'C-H': 10}) def test_fingerprint(self): - """Test the fingerprint function.""" + """Test the fingerprint() function.""" fp = fingerprint(ARCSpecies(label='CH4', smiles='C')) self.assertEqual(fp, {0: {'self': 'C', 'H': [1, 2, 3, 4]}}) @@ -1198,7 +794,7 @@ def test_fingerprint(self): 10: {'self': 'O', 'C': [0], 'H': [11]}}) def test_identify_superimposable_candidates(self): - """Test the identify_superimposable_candidates function.""" + """Test the identify_superimposable_candidates() function.""" candidates = identify_superimposable_candidates(fingerprint_1=self.fingerprint_1, fingerprint_2=self.fingerprint_2) self.assertEqual(candidates, [{0: 0, 3: 1, 4: 2, 5: 3, 6: 4}]) @@ -1208,7 +804,7 @@ def test_identify_superimposable_candidates(self): self.assertEqual(candidates, [{0: 0, 5: 5, 4: 4, 3: 3, 2: 2, 1: 1, 6: 6, 7: 7, 8: 8, 9: 9}]) def test_are_adj_elements_in_agreement(self): - """Test the are_adj_elements_in_agreement function.""" + """Test the are_adj_elements_in_agreement() function.""" self.assertFalse(are_adj_elements_in_agreement({'self': 'C', 'C': [3]}, {'self': 'C', 'C': [0, 3], 'H': [2]})) self.assertFalse(are_adj_elements_in_agreement({'self': 'C', 'C': [3]}, @@ -1225,7 +821,7 @@ def test_are_adj_elements_in_agreement(self): {'self': 'C', 'C': [1], 'O': [4], 'H': [8]})) def test_iterative_dfs(self): - """Test the iterative_dfs function.""" + """Test the iterative_dfs() function.""" result = iterative_dfs(fingerprint_1={0: {'self': 'C', 'C': [1]}, 1: {'self': 'C', 'H': [3, 4]}}, fingerprint_2={0: {'self': 'C', 'C': [1]}, @@ -1242,7 +838,7 @@ def test_iterative_dfs(self): self.assertEqual(result, {0: 0, 3: 1, 4: 2, 5: 3, 6: 4}) def test_prune_identical_dicts(self): - """Test the prune_identical_dicts function.""" + """Test the prune_identical_dicts() function.""" new_dicts_list = prune_identical_dicts([{0: 0}]) self.assertEqual(new_dicts_list, [{0: 0}]) new_dicts_list = prune_identical_dicts([{0: 0}, {0: 0}, {0: 0}]) @@ -1257,12 +853,12 @@ def test_prune_identical_dicts(self): self.assertEqual(new_dicts_list, [{0: 0, 3: 1, 4: 2, 5: 3, 6: 4}]) def test_remove_gaps_from_values(self): - """Test the remove_gaps_from_values function.""" + """Test the remove_gaps_from_values() function.""" self.assertEqual(remove_gaps_from_values({5: 18, 7: 502, 21: 0, 0: 55, 2: 1}), {5: 2, 7: 4, 21: 0, 0: 3, 2: 1}) def test_fix_dihedrals_by_backbone_mapping(self): - """Test the fix_dihedrals_by_backbone_mapping function.""" + """Test the fix_dihedrals_by_backbone_mapping() function.""" self.spc1.determine_rotors() self.spc2.determine_rotors() atom_map = {0: 0, 1: 6, 2: 5, 3: 1, 4: 2, 5: 3, 6: 4, 7: 7, 8: 8, 9: 10, 10: 9, 11: 11, 12: 12, 13: 13} @@ -1285,7 +881,7 @@ def test_fix_dihedrals_by_backbone_mapping(self): self.assertAlmostEqual(new_dihedrals_2[2], 121.23139016907017, places = 5) def test_get_backbone_dihedral_deviation_score(self): - """Test the get_backbone_dihedral_deviation_score function.""" + """Test the get_backbone_dihedral_deviation_score() function.""" self.spc1_dihedral_deviation.determine_rotors() self.spc2_dihedral_deviation.determine_rotors() fingerprint_1, fingerprint_2 = fingerprint(self.spc1), fingerprint(self.spc2) @@ -1294,7 +890,7 @@ def test_get_backbone_dihedral_deviation_score(self): self.assertAlmostEqual(score, 106.8417836) def test_get_backbone_dihedral_angles(self): - """Test the get_backbone_dihedral_angles function.""" + """Test the get_backbone_dihedral_angles() function.""" self.spc1.determine_rotors() self.spc2.determine_rotors() torsions = get_backbone_dihedral_angles(self.spc1, self.spc2, backbone_map={0: 0, 3: 1, 5: 3, 6: 4, 4: 2}) @@ -1313,7 +909,7 @@ def test_map_lists(self): map_lists([5.0], [3.2, 7.9]) def test_map_hydrogens(self): - """Test the map_hydrogens function.""" + """Test the map_hydrogens() function.""" # Todo: Add tests with many types of torsions and non(pseudo)-torsions, multiple bonds, cyclics. # CH4 different order spc1 = ARCSpecies(label='CH4', smiles='C', xyz=self.ch4_xyz) @@ -1331,8 +927,46 @@ def test_map_hydrogens(self): self.assertEqual(atom_map, {0: 0, 1: 6, 2: 5, 3: 1, 4: 2, 5: 3, 6: 4, 7: 7, 8: 8, 9: 10, 10: 9, 11: 11, 12: 12, 13: 13}) + def test_add_adjacent_hydrogen_atoms_to_map_based_on_a_specific_torsion(self): + """test the add_adjacent_hydrogen_atoms_to_map_based_on_a_specific_torsion() function""" + spc1 = ARCSpecies(label="c4", smiles = "CCCC") + spc2 = spc1.copy() + out_dict = add_adjacent_hydrogen_atoms_to_map_based_on_a_specific_torsion(spc1, + spc2, + spc1.mol.atoms[0], + spc2.mol.atoms[0], + [0,1,2,3], + {0: 0,1: 1,2: 2,3: 3}, + True) + for key in range(7): + self.assertEqual(out_dict[key], key) + + spc1 = ARCSpecies(label="c4", smiles = "cccc") + spc2 = spc1.copy() + out_dict = add_adjacent_hydrogen_atoms_to_map_based_on_a_specific_torsion(spc1, + spc2, + spc1.mol.atoms[0], + spc2.mol.atoms[0], + [0,1,2,3], + {0: 0,1: 1,2: 2,3: 3}, + True) + for key in range(4): + self.assertEqual(out_dict[key], key) + + spc1 = ARCSpecies(label="cn", smiles = "N1NN1") + spc2 = spc1.copy() + out_dict = add_adjacent_hydrogen_atoms_to_map_based_on_a_specific_torsion(spc1, + spc2, + spc1.mol.atoms[0], + spc2.mol.atoms[0], + [0,1,2,3], + {0: 0,1: 1,2: 2,3: 3}, + True) + for key in range(4): + self.assertEqual(out_dict[key], key) + def test_flip_map(self): - """Test the flip_map function.""" + """Test the flip_map() function.""" atom_map = [0] flipped_map = flip_map(atom_map) self.assertEqual(flipped_map, [0]) @@ -1362,7 +996,7 @@ def test_flip_map(self): flip_map(atom_map) def test_make_bond_changes(self): - """Test the make_bond_changes function""" + """Test the make_bond_changes() function""" spc1 = ARCSpecies(label="Test_bc", smiles="[CH2][CH2]") spc2 = ARCSpecies(label="Test_bc", smiles="C=C") label_species_atoms([spc1]) @@ -1381,7 +1015,7 @@ def test_make_bond_changes(self): assign_labels_to_products(rxn, p_label_dict) reactants, products = copy_species_list_for_mapping(rxn.r_species), copy_species_list_for_mapping(rxn.p_species) label_species_atoms(reactants), label_species_atoms(products) - r_bdes, _ = find_all_bdes(rxn, r_label_dict, True), find_all_bdes(rxn, p_label_dict, False) + r_bdes, _ = find_all_breaking_bonds(rxn, r_label_dict, True), find_all_breaking_bonds(rxn, p_label_dict, False) r_cuts = cut_species_based_on_atom_indices(reactants, r_bdes) self.assertFalse(r_cuts[1].mol.is_isomorphic(rxn.p_species[1].mol)) make_bond_changes(rxn=rxn, @@ -1390,7 +1024,7 @@ def test_make_bond_changes(self): self.assertTrue(r_cuts[1].mol.is_isomorphic(rxn.p_species[1].mol)) def test_update_xyz(self): - """tests the update_xyz function""" + """tests the update_xyz() function""" spc = ARCSpecies(label="test_UX",smiles = "BrC(F)Cl") shuffle(spc.mol.atoms) update_xyz([spc]) @@ -1398,7 +1032,7 @@ def test_update_xyz(self): atoms = [atom.element.symbol for atom in spc.mol.atoms] for label1, label2 in zip(atoms, xyz): self.assertEqual(label1, label2) - + spc = ARCSpecies(label="test_UX", smiles = "OCl") shuffle(spc.mol.atoms) update_xyz([spc]) @@ -1415,52 +1049,179 @@ def test_update_xyz(self): for label1,label2 in zip(atoms, xyz): self.assertEqual(label1, label2) - def test_add_adjacent_hydrogen_atoms_to_map_based_on_a_specific_torsion(self): - "test the add_adjacent_hydrogen_atoms_to_map_based_on_a_specific_torsion function" - spc1 = ARCSpecies(label="c4", smiles = "CCCC") - spc2 = spc1.copy() - out_dict = add_adjacent_hydrogen_atoms_to_map_based_on_a_specific_torsion(spc1, - spc2, - spc1.mol.atoms[0], - spc2.mol.atoms[0], - [0,1,2,3], - {0: 0,1: 1,2: 2,3: 3}, - True) - for key in range(7): - self.assertEqual(out_dict[key], key) - - spc1 = ARCSpecies(label="c4", smiles = "cccc") - spc2 = spc1.copy() - out_dict = add_adjacent_hydrogen_atoms_to_map_based_on_a_specific_torsion(spc1, - spc2, - spc1.mol.atoms[0], - spc2.mol.atoms[0], - [0,1,2,3], - {0: 0,1: 1,2: 2,3: 3}, - True) - for key in range(4): - self.assertEqual(out_dict[key], key) + def test_r_cut_p_cut_isomorphic(self): + """Test the r_cut_p_cut_isomorphic() function""" + rxn_1_test = ARCReaction(r_species=[self.r_1, self.r_2], p_species=[self.p_1, self.p_2]) + reactants, products = copy_species_list_for_mapping(rxn_1_test.r_species), copy_species_list_for_mapping(rxn_1_test.p_species) + label_species_atoms(reactants), label_species_atoms(products) - spc1 = ARCSpecies(label="cn", smiles = "N1NN1") - spc2 = spc1.copy() - out_dict = add_adjacent_hydrogen_atoms_to_map_based_on_a_specific_torsion(spc1, - spc2, - spc1.mol.atoms[0], - spc2.mol.atoms[0], - [0,1,2,3], - {0: 0,1: 1,2: 2,3: 3}, - True) - for key in range(4): - self.assertEqual(out_dict[key], key) + r_bdes, p_bdes = find_all_breaking_bonds(rxn_1_test, self.r_label_dict_rxn_1, True), find_all_breaking_bonds(rxn_1_test, self.p_label_dict_rxn_1, False) - def test_find_all_bdes(self): - """tests the find_all_bdes function""" - rxn = ARCReaction(r_species=[self.r_1, self.r_2], p_species=[self.p_1, self.p_2]) - bdes = find_all_bdes(rxn, self.r_label_dict_rxn_1, True) - self.assertEqual(bdes, [(2, 4)]) + r_cuts = cut_species_based_on_atom_indices(reactants, r_bdes) + p_cuts = cut_species_based_on_atom_indices(products, p_bdes) + + self.assertTrue(r_cut_p_cut_isomorphic(self.spc1,self.spc2)) + for r_cut in r_cuts: + for p_cut in p_cuts: + if r_cut.mol.is_isomorphic(p_cut.mol): + p_cuts.remove(p_cut) + break + self.assertEqual(len(p_cuts),0) + self.assertTrue(r_cut_p_cut_isomorphic(ARCSpecies(label="r1", smiles="F[C]F", multiplicity=1), + ARCSpecies(label="r1", smiles="F[C]F", multiplicity=3))) + + def test_pairing_reactants_and_products_for_mapping(self): + """Test the pairing_reactants_and_products_for_mapping() function""" + smiles = ["[F]", "C[CH]C", "[CH3]"] + rxn_1_test = ARCReaction(r_species=[self.r_1, self.r_2], p_species=[self.p_1, self.p_2]) + reactants, products = copy_species_list_for_mapping(rxn_1_test.r_species), copy_species_list_for_mapping(rxn_1_test.p_species) + label_species_atoms(reactants), label_species_atoms(products) + + r_bdes, p_bdes = find_all_breaking_bonds(rxn_1_test, self.r_label_dict_rxn_1, True), find_all_breaking_bonds(rxn_1_test, self.p_label_dict_rxn_1, False) + + r_cuts = cut_species_based_on_atom_indices(reactants, r_bdes) + p_cuts = cut_species_based_on_atom_indices(products, p_bdes) + + pairs_of_reactant_and_products = pairing_reactants_and_products_for_mapping(r_cuts, p_cuts) + + for pair in pairs_of_reactant_and_products: + self.assertTrue(pair[0].mol.copy(deep=True).is_isomorphic(pair[1].mol.copy(deep=True))) + self.assertIn(str(pair[0].mol.copy(deep=True).to_smiles()), smiles) + smiles.remove(pair[0].mol.copy(deep=True).to_smiles()) + + def test_map_pairs(self): + """Test the map_pairs() function""" + rxn_1_test = ARCReaction(r_species=[self.r_1, self.r_2], p_species=[self.p_1, self.p_2]) + rmg_reactions = get_rmg_reactions_from_arc_reaction(arc_reaction=rxn_1_test, backend="ARC") + r_label_dict, p_label_dict = get_atom_indices_of_labeled_atoms_in_an_rmg_reaction(arc_reaction=rxn_1_test, + rmg_reaction=rmg_reactions[0]) + reactants, products = copy_species_list_for_mapping(rxn_1_test.r_species), copy_species_list_for_mapping(rxn_1_test.p_species) + label_species_atoms(reactants), label_species_atoms(products) + + r_bdes, p_bdes = find_all_breaking_bonds(rxn_1_test, r_label_dict, True), find_all_breaking_bonds(rxn_1_test, p_label_dict, False) + + r_cuts = cut_species_based_on_atom_indices(reactants, r_bdes) + p_cuts = cut_species_based_on_atom_indices(products, p_bdes) + pairs_of_reactant_and_products = pairing_reactants_and_products_for_mapping(r_cuts, p_cuts) + maps = map_pairs(pairs_of_reactant_and_products) + for map_ in maps: + if len(map_) == 1: + self.assertEqual(map_[0], 0) + elif len(map_) == 4: + self.assertEqual(map_[0], 0) + for i in map_[1:]: + self.assertIn(i, [1, 2, 3]) + self.assertEqual(len(np.unique(map_)), len(map_)) + else: + self.assertEqual(map_[:3], [0, 1, 2]) + self.assertIn(tuple(map_[3:6]), list(itertools.permutations([3, 4, 5]))) + self.assertEqual(map_[6], 6) + self.assertIn(tuple(map_[7:]), list(itertools.permutations([7, 8, 9]))) + self.assertEqual(len(np.unique(map_)), len(map_)) + + def test_glue_maps(self): + rxn_1_test = ARCReaction(r_species=[self.r_1, self.r_2], p_species=[self.p_1, self.p_2]) + rmg_reactions = get_rmg_reactions_from_arc_reaction(arc_reaction=rxn_1_test, backend="ARC") + r_label_dict, p_label_dict = get_atom_indices_of_labeled_atoms_in_an_rmg_reaction(arc_reaction=rxn_1_test, + rmg_reaction=rmg_reactions[0]) + assign_labels_to_products(rxn_1_test, p_label_dict) + reactants, products = copy_species_list_for_mapping(rxn_1_test.r_species), copy_species_list_for_mapping(rxn_1_test.p_species) + label_species_atoms(reactants), label_species_atoms(products) + + r_bdes, p_bdes = find_all_breaking_bonds(rxn_1_test, r_label_dict, True), find_all_breaking_bonds(rxn_1_test, p_label_dict, False) + + r_cuts = cut_species_based_on_atom_indices(reactants, r_bdes) + p_cuts = cut_species_based_on_atom_indices(products, p_bdes) + pairs_of_reactant_and_products = pairing_reactants_and_products_for_mapping(r_cuts, p_cuts) + maps = map_pairs(pairs_of_reactant_and_products) + atom_map = glue_maps(maps,pairs_of_reactant_and_products) + self.assertEqual(len(atom_map), self.r_1.mol.get_num_atoms() + self.r_2.mol.get_num_atoms()) + atoms_r = [atom for atom in self.r_1.mol.copy(deep=True).atoms] + [atom for atom in self.r_2.mol.copy(deep=True).atoms] + atoms_p = [atom for atom in self.p_1.mol.copy(deep=True).atoms] + [atom for atom in self.p_2.mol.copy(deep=True).atoms] + for index, value in enumerate(atom_map): + self.assertEqual(atoms_r[index].symbol, atoms_p[value].symbol) + + def test_determine_bdes_on_spc_based_on_atom_labels(self): + """tests the determine_bdes_indices_based_on_atom_labels() function""" + spc = ARCSpecies(label="ethane", smiles="CC") + label_species_atoms([spc]) + self.assertTrue(determine_bdes_on_spc_based_on_atom_labels(spc, (1, 2))) + self.assertFalse(determine_bdes_on_spc_based_on_atom_labels(spc, (2, 3))) + + def test_cut_species_based_on_atom_indices(self): + """test the cut_species_based_on_atom_indices() function""" + reactants = copy_species_list_for_mapping(self.arc_reaction_2.r_species) + products = copy_species_list_for_mapping(self.arc_reaction_2.p_species) + label_species_atoms(reactants), label_species_atoms(products) + product_dict = get_reaction_family_products(self.arc_reaction_2) + print(product_dict) + product_dict = product_dict[0] + r_label_dict_rxn_1 = product_dict['r_label_map'] + p_label_dict_rxn_1 = product_dict['p_label_map'] + print(r_label_dict_rxn_1) + print(p_label_dict_rxn_1) + + r_bdes = find_all_breaking_bonds(rxn=self.arc_reaction_2, label_dict=r_label_dict_rxn_1, r_direction=True) + p_bdes = find_all_breaking_bonds(rxn=self.arc_reaction_2, label_dict=p_label_dict_rxn_1, r_direction=False) + + r_cuts = cut_species_based_on_atom_indices(reactants, r_bdes) + p_cuts = cut_species_based_on_atom_indices(products, p_bdes) + + print(f'r_bdes: {r_bdes}') + print(f'p_bdes: {p_bdes}') + print(f'r_cuts: {[s.mol.to_smiles() for s in r_cuts]}') + print(f'p_cuts: {[s.mol.to_smiles() for s in p_cuts]}') + + self.assertIn("[CH2]CC", [r_cut.mol.copy(deep=True).smiles for r_cut in r_cuts]) + self.assertIn("[H]", [r_cut.mol.copy(deep=True).smiles for r_cut in r_cuts]) + self.assertIn("[NH2]", [r_cut.mol.copy(deep=True).smiles for r_cut in r_cuts]) + self.assertIn("[CH2]CC", [p_cut.mol.copy(deep=True).smiles for p_cut in p_cuts]) + self.assertIn("[H]", [p_cut.mol.copy(deep=True).smiles for p_cut in p_cuts]) + self.assertIn("[NH2]", [p_cut.mol.copy(deep=True).smiles for p_cut in p_cuts]) + + spc = ARCSpecies(label="test", smiles="CNC", bdes = [(1, 2), (2, 3)]) + for i, a in enumerate(spc.mol.atoms): + a.label=str(i) + cuts = cut_species_based_on_atom_indices([spc], [(1, 2), (2, 3)]) + self.assertEqual(len(cuts), 3) + for cut in cuts: + self.assertTrue(any([cut.mol.copy(deep=True).is_isomorphic(ARCSpecies(label="1", smiles="[CH3]").mol), + cut.mol.copy(deep=True).is_isomorphic(ARCSpecies(label="2", smiles="[NH]").mol)])) + + h2 = ARCSpecies(label="H2", smiles="[H][H]") + label_species_atoms([h2]) + + cuts = cut_species_based_on_atom_indices([h2], [(1, 2)]) + self.assertEqual(len(cuts), 2) + for cut in cuts: + self.assertEqual(cut.get_xyz()["symbols"], ('H',)) + + spcs = [ARCSpecies(label="r", smiles = 'O=C(O)CCF')] + label_species_atoms(spcs) + cuts = cut_species_based_on_atom_indices(spcs, [(6, 5), (4, 2), (3, 7)]) + self.assertEqual(len(cuts), 4) + + def test_translate_indices_based_on_ref_species(self): + """tests the translate_indices_based_on_ref_species() function""" + spc_1 = ARCSpecies(label='pentanol', smiles='CCCO') + spc_2 = spc_1.copy() + new_bdes = translate_indices_based_on_ref_species(species=[spc_1], ref_species=[spc_2], bdes=[(0, 1)]) + self.assertEqual(new_bdes, [(0, 1)]) + + spc_1 = ARCSpecies(label='C3Oj', smiles='CCC[O]') + spc_2 = ARCSpecies(label='H', smiles='[H]') + spc_3 = ARCSpecies(label='OH', smiles='[OH]') + new_bdes = translate_indices_based_on_ref_species(species=[spc_3, spc_2, spc_1], + ref_species=[spc_3, spc_2, spc_1], + bdes=[(0, 2), (2, 6)]) + self.assertEqual(new_bdes, [(0, 2), (2, 6)]) + new_bdes = translate_indices_based_on_ref_species(species=[spc_1, spc_2, spc_3], + ref_species=[spc_3, spc_2, spc_1], + bdes=[(0, 2), (2, 6)]) + self.assertEqual(new_bdes, [(12, 11), (11, 3)]) def test_copy_species_list_for_mapping(self): - """tests the copy_species_list_for_mapping function""" + """tests the copy_species_list_for_mapping() function""" species = [ARCSpecies(label="test_1", smiles = "BrC(F)Cl"), ARCSpecies(label="test_2", smiles = "OOC(F)CCCONNO")] label_species_atoms(species) species_copy = copy_species_list_for_mapping(species) @@ -1471,12 +1232,29 @@ def test_copy_species_list_for_mapping(self): self.assertIsNot(atom1, atom2) self.assertEqual(atom1.label, atom2.label) - def test_determine_bdes_on_spc_based_on_atom_labels(self): - """tests the determine_bdes_indices_based_on_atom_labels function""" - spc = ARCSpecies(label="ethane", smiles="CC") - label_species_atoms([spc]) - self.assertTrue(determine_bdes_on_spc_based_on_atom_labels(spc, (1, 2))) - self.assertFalse(determine_bdes_on_spc_based_on_atom_labels(spc, (2, 3))) + def test_find_all_breaking_bonds(self): + """tests the find_all_breaking_bonds() function""" + rxn = ARCReaction(r_species=[self.r_1, self.r_2], p_species=[self.p_1, self.p_2]) + products_dicts = get_reaction_family_products(rxn) + bdes = find_all_breaking_bonds(rxn=rxn, label_dict=products_dicts[0]['r_label_map'], r_direction=True) + self.assertEqual(bdes, [(4, 10)]) + bdes = find_all_breaking_bonds(rxn=rxn, label_dict=products_dicts[0]['r_label_map'], r_direction=False) + self.assertEqual(bdes, [(10, 16)]) + + products_dicts = get_reaction_family_products(self.arc_reaction_1) + bdes = find_all_breaking_bonds(rxn=self.arc_reaction_1, label_dict=products_dicts[0]['r_label_map'], r_direction=True) + self.assertEqual(bdes, [(0, 1)]) + + products_dicts = get_reaction_family_products(self.arc_reaction_2) + bdes = find_all_breaking_bonds(rxn=self.arc_reaction_2, label_dict=products_dicts[0]['r_label_map'], r_direction=True) + self.assertEqual(bdes, [(0, 3)]) + bdes = find_all_breaking_bonds(rxn=self.arc_reaction_2, label_dict=products_dicts[0]['r_label_map'], r_direction=False) + self.assertEqual(bdes, [(3, 11)]) + + products_dicts = get_reaction_family_products(self.arc_reaction_3) + bdes = find_all_breaking_bonds(rxn=self.arc_reaction_3, label_dict=products_dicts[0]['r_label_map'], r_direction=False) + self.assertEqual(bdes, [(0, 4)]) + if __name__ == '__main__': - unittest.main(testRunner=unittest.TextTestRunner(verbosity=2)) \ No newline at end of file + unittest.main(testRunner=unittest.TextTestRunner(verbosity=2))