add ec number prediction backend with model and config

data/ec_class_names.json

@@ -0,0 +1,412 @@
+{
+"1": "Oxidoreductases",
+"1.1": "Acting on the CH-OH group of donors",
+"1.1.1": "With NAD(+) or NADP(+) as acceptor",
+"1.1.2": "With a cytochrome as acceptor",
+"1.1.3": "With oxygen as acceptor",
+"1.1.4": "With a disulfide as acceptor",
+"1.1.5": "With a quinone or similar compound as acceptor",
+"1.1.7": "With an iron-sulfur protein as acceptor",
+"1.1.9": "With a copper protein as acceptor",
+"1.1.98": "With other, known, acceptors",
+"1.1.99": "With other acceptors",
+"1.2": "Acting on the aldehyde or oxo group of donors",
+"1.2.1": "With NAD(+) or NADP(+) as acceptor",
+"1.2.2": "With a cytochrome as acceptor",
+"1.2.3": "With oxygen as acceptor",
+"1.2.4": "With a disulfide as acceptor",
+"1.2.5": "With a quinone or similar compound as acceptor",
+"1.2.7": "With an iron-sulfur protein as acceptor",
+"1.2.8": "With a flavin or flavoprotein as acceptor",
+"1.2.98": "With other, known, acceptors",
+"1.2.99": "With other acceptors",
+"1.3": "Acting on the CH-CH group of donors",
+"1.3.1": "With NAD(+) or NADP(+) as acceptor",
+"1.3.2": "With a cytochrome as acceptor",
+"1.3.3": "With oxygen as acceptor",
+"1.3.4": "With a disulfide as acceptor",
+"1.3.5": "With a quinone or related compound as acceptor",
+"1.3.7": "With an iron-sulfur protein as acceptor",
+"1.3.8": "With a flavin as acceptor",
+"1.3.98": "With other, known, acceptors",
+"1.3.99": "With other acceptors",
+"1.4": "Acting on the CH-NH2 group of donors",
+"1.4.1": "With NAD(+) or NADP(+) as acceptor",
+"1.4.2": "With a cytochrome as acceptor",
+"1.4.3": "With oxygen as acceptor",
+"1.4.4": "With a disulfide as acceptor",
+"1.4.5": "With a quinone or similar compound as acceptor",
+"1.4.7": "With an iron-sulfur protein as acceptor",
+"1.4.9": "With a copper protein as acceptor",
+"1.4.98": "With other, known, acceptors",
+"1.4.99": "With other acceptors",
+"1.5": "Acting on the CH-NH group of donors",
+"1.5.1": "With NAD(+) or NADP(+) as acceptor",
+"1.5.3": "With oxygen as acceptor",
+"1.5.4": "With a disulfide as acceptor",
+"1.5.5": "With a quinone or similar compound as acceptor",
+"1.5.7": "With an iron-sulfur protein as acceptor",
+"1.5.8": "With a flavin as acceptor",
+"1.5.98": "With other, known, acceptors",
+"1.5.99": "With other acceptors",
+"1.6": "Acting on NADH or NADPH",
+"1.6.1": "With NAD(+) or NADP(+) as acceptor",
+"1.6.2": "With a heme protein as acceptor",
+"1.6.3": "With oxygen as acceptor",
+"1.6.4": "With a disulfide as acceptor",
+"1.6.5": "With a quinone or similar compound as acceptor",
+"1.6.6": "With a nitrogenous group as acceptor",
+"1.6.7": "With a iron-sulfur protein as acceptor",
+"1.6.8": "With a flavin as acceptor",
+"1.6.99": "With other acceptors",
+"1.7": "Acting on other nitrogenous compounds as donors",
+"1.7.1": "With NAD(+) or NADP(+) as acceptor",
+"1.7.2": "With a cytochrome as acceptor",
+"1.7.3": "With oxygen as acceptor",
+"1.7.5": "With a quinone or similar compound as acceptor",
+"1.7.6": "With a nitrogenous group as acceptor",
+"1.7.7": "With an iron-sulfur protein as acceptor",
+"1.7.9": "With a copper protein as acceptor",
+"1.7.99": "With other acceptors",
+"1.8": "Acting on a sulfur group of donors",
+"1.8.1": "With NAD(+) or NADP(+) as acceptor",
+"1.8.2": "With a cytochrome as acceptor",
+"1.8.3": "With oxygen as acceptor",
+"1.8.4": "With a disulfide as acceptor",
+"1.8.5": "With a quinone or similar compound as acceptor",
+"1.8.6": "With an nitrogenous group as acceptor",
+"1.8.7": "With an iron-sulfur protein as acceptor",
+"1.8.98": "With other, known, acceptors",
+"1.8.99": "With other acceptors",
+"1.9": "Acting on a heme group of donors",
+"1.9.3": "With oxygen as acceptor",
+"1.9.6": "With a nitrogenous group as acceptor",
+"1.9.98": "With other, known, acceptors",
+"1.9.99": "With other acceptors",
+"1.10": "Acting on diphenols and related substances as donors",
+"1.10.1": "With NAD(+) or NADP(+) as acceptor",
+"1.10.2": "With a cytochrome as acceptor",
+"1.10.3": "With oxygen as acceptor",
+"1.10.5": "With a quinone or related compound as acceptor",
+"1.10.9": "With a copper protein as acceptor",
+"1.10.98": "With other, known, acceptors",
+"1.10.99": "With other acceptors",
+"1.11": "Acting on a peroxide as acceptor",
+"1.11.1": "Peroxidases",
+"1.11.2": "With H2O2 as acceptor, one oxygen atom of which is incorporated into the product",
+"1.12": "Acting on hydrogen as donors",
+"1.12.1": "With NAD(+) or NADP(+) as acceptor",
+"1.12.2": "With a cytochrome as acceptor",
+"1.12.5": "With a quinone or similar compound as acceptor",
+"1.12.7": "With an iron-sulfur protein as acceptor",
+"1.12.98": "With other, known, acceptors",
+"1.12.99": "With other acceptors",
+"1.13": "Acting on single donors with incorporation of molecular oxygen (oxygenases). The oxygen incorporated need not be derived from O2",
+"1.13.1": "With NADH or NADPH as one donor",
+"1.13.11": "With incorporation of two atoms of oxygen",
+"1.13.12": "With incorporation of one atom of oxygen (internal monooxygenases or internal mixed function oxidases)",
+"1.13.99": "Miscellaneous",
+"1.14": "Acting on paired donors, with incorporation or reduction of molecular oxygen. The oxygen incorporated need not be derived from O2",
+"1.14.1": "With NADH or NADPH as one donor",
+"1.14.2": "With ascorbate as one donor",
+"1.14.3": "With reduced pteridine as one donor",
+"1.14.11": "With 2-oxoglutarate as one donor, and incorporation of one atom each of oxygen into both donors",
+"1.14.12": "With NADH or NADPH as one donor, and incorporation of two atoms of oxygen into one donor",
+"1.14.13": "With NADH or NADPH as one donor, and incorporation of one atom of oxygen",
+"1.14.14": "With reduced flavin or flavoprotein as one donor, and incorporation of one atom of oxygen",
+"1.14.15": "With reduced iron-sulfur protein as one donor, and incorporation of one atom of oxygen",
+"1.14.16": "With reduced pteridine as one donor, and incorporation of one atom of oxygen",
+"1.14.17": "With reduced ascorbate as one donor, and incorporation of one atom of oxygen",
+"1.14.18": "With another compound as one donor, and incorporation of one atom of oxygen",
+"1.14.19": "With oxidation of a pair of donors resulting in the reduction of molecular oxygen to two molecules of water",
+"1.14.20": "With 2-oxoglutarate as one donor, and the other dehydrogenated",
+"1.14.21": "With NADH or NADPH as one donor, and the other dehydrogenated",
+"1.14.99": "Miscellaneous",
+"1.15": "Acting on superoxide as acceptor",
+"1.15.1": "Acting on superoxide as acceptor",
+"1.16": "Oxidizing metal ions",
+"1.16.1": "With NAD(+) or NADP(+) as acceptor",
+"1.16.2": "With a cytochrome as acceptor",
+"1.16.3": "With oxygen as acceptor",
+"1.16.5": "With a quinone or similar compound as acceptor",
+"1.16.8": "With a flavin as acceptor",
+"1.16.9": "With a copper protein as acceptor",
+"1.16.98": "With other known acceptors",
+"1.16.99": "With other acceptors",
+"1.17": "Acting on CH or CH2 groups",
+"1.17.1": "With NAD(+) or NADP(+) as acceptor",
+"1.17.2": "With a cytochrome as acceptor",
+"1.17.3": "With oxygen as acceptor",
+"1.17.4": "With a disulfide as acceptor",
+"1.17.5": "With a quinone or similar compound as acceptor",
+"1.17.7": "With an iron-sulfur protein as acceptor",
+"1.17.8": "With a flavin as acceptor",
+"1.17.9": "With a copper protein as acceptor",
+"1.17.98": "With other, known, acceptors",
+"1.17.99": "With other acceptors",
+"1.18": "Acting on iron-sulfur proteins as donors",
+"1.18.1": "With NAD(+) or NADP(+) as acceptor",
+"1.18.2": "With dinitrogen as acceptor",
+"1.18.3": "With H(+) as acceptor",
+"1.18.6": "With dinitrogen as acceptor",
+"1.18.96": "With other, known, acceptors",
+"1.18.99": "With H(+) as acceptor",
+"1.19": "Acting on reduced flavodoxin as donor",
+"1.19.1": "With NAD(+) or NADP(+) as acceptor",
+"1.19.6": "With dinitrogen as acceptor",
+"1.20": "Acting on phosphorus or arsenic in donors",
+"1.20.1": "With NAD(+) or NADP(+) as acceptor",
+"1.20.2": "With a cytochrome as acceptor",
+"1.20.4": "With disulfide as acceptor",
+"1.20.9": "With a copper protein as acceptor",
+"1.20.98": "With other, known acceptors",
+"1.20.99": "With other acceptors",
+"1.21": "Catalyzing the reaction X-H + Y-H = 'X-Y'",
+"1.21.1": "With NAD(+) or NADP(+) as acceptor",
+"1.21.3": "With oxygen as acceptor",
+"1.21.4": "With a disulfide as acceptor",
+"1.21.98": "With other, known acceptors",
+"1.21.99": "With other acceptors",
+"1.22": "Acting on halogen in donors",
+"1.22.1": "With NAD(+) or NADP(+) as acceptor",
+"1.23": "Reducing C-O-C group as acceptor",
+"1.23.1": "With NADH or NADPH as donor",
+"1.23.5": "With a quinone or similar compound as acceptor",
+"1.97": "Other oxidoreductases",
+"1.97.1": "Other oxidoreductases",
+"1.98": "Enzymes using H2 as reductant",
+"1.98.1": "Other oxidoreductases",
+"1.99": "Other enzymes using O2 as oxidant",
+"1.99.1": "Hydroxylases",
+"1.99.2": "Oxygenases",
+"2": "Transferases",
+"2.1": "Transferring one-carbon groups",
+"2.1.1": "Methyltransferases",
+"2.1.2": "Hydroxymethyl-, formyl- and related transferases",
+"2.1.3": "Carboxy- and carbamoyltransferases",
+"2.1.4": "Amidinotransferases",
+"2.1.5": "Methylenetransferases",
+"2.2": "Transferring aldehyde or ketonic groups",
+"2.2.1": "Transketolases and transaldolases",
+"2.3": "Acyltransferases",
+"2.3.1": "Transferring groups other than amino-acyl groups",
+"2.3.2": "Aminoacyltransferases",
+"2.3.3": "Acyl groups converted into alkyl groups on transfer",
+"2.4": "Glycosyltransferases",
+"2.4.1": "Hexosyltransferases",
+"2.4.2": "Pentosyltransferases",
+"2.4.3": "Sialyltransferases",
+"2.4.99": "Transferring other glycosyl groups",
+"2.5": "Transferring alkyl or aryl groups, other than methyl groups",
+"2.5.1": "Transferring alkyl or aryl groups, other than methyl groups",
+"2.6": "Transferring nitrogenous groups",
+"2.6.1": "Transaminases",
+"2.6.2": "Amidinotransferases",
+"2.6.3": "Oximinotransferases",
+"2.6.99": "Transferring other nitrogenous groups",
+"2.7": "Transferring phosphorus-containing groups",
+"2.7.1": "Phosphotransferases with an alcohol group as acceptor",
+"2.7.2": "Phosphotransferases with a carboxy group as acceptor",
+"2.7.3": "Phosphotransferases with a nitrogenous group as acceptor",
+"2.7.4": "Phosphotransferases with a phosphate group as acceptor",
+"2.7.5": "Phosphotransferases with regeneration of donors, apparently catalyzing intramolecular transfers",
+"2.7.6": "Diphosphotransferases",
+"2.7.7": "Nucleotidyltransferases",
+"2.7.8": "Transferases for other substituted phosphate groups",
+"2.7.9": "Phosphotransferases with paired acceptors",
+"2.7.10": "Protein-tyrosine kinases",
+"2.7.11": "Protein-serine/threonine kinases",
+"2.7.12": "Dual-specificity kinases (those acting on Ser/Thr and Tyr residues)",
+"2.7.13": "Protein-histidine kinases",
+"2.7.14": "Protein-arginine kinases",
+"2.7.99": "Other protein kinases",
+"2.8": "Transferring sulfur-containing groups",
+"2.8.1": "Sulfurtransferases",
+"2.8.2": "Sulfotransferases",
+"2.8.3": "CoA-transferases",
+"2.8.4": "Transferring alkylthio groups",
+"2.8.5": "Thiosulfotransferases",
+"2.9": "Transferring selenium-containing groups",
+"2.9.1": "Selenotransferases",
+"2.10": "Transferring molybdenum- or tungsten-containing groups",
+"2.10.1": "Molybdenumtransferases or tungstentransferases with sulfide groups as acceptors",
+"3": "Hydrolases",
+"3.1": "Acting on ester bonds",
+"3.1.1": "Carboxylic ester hydrolases",
+"3.1.2": "Thiolester hydrolases",
+"3.1.3": "Phosphoric monoester hydrolases",
+"3.1.4": "Phosphoric diester hydrolases",
+"3.1.5": "Triphosphoric monoester hydrolases",
+"3.1.6": "Sulfuric ester hydrolases",
+"3.1.7": "Diphosphoric monoester hydrolases",
+"3.1.8": "Phosphoric triester hydrolases",
+"3.1.11": "Exodeoxyribonucleases producing 5'-phosphomonoesters",
+"3.1.12": "Exodeoxyribonucleases producing 3'-phosphomonoesters",
+"3.1.13": "Exoribonucleases producing 5'-phosphomonoesters",
+"3.1.14": "Exoribonucleases producing 3'-phosphomonoesters",
+"3.1.15": "Exonucleases active with either ribo- or deoxyribonucleic acids and producing 5'-phosphomonoesters",
+"3.1.16": "Exonucleases active with either ribo- or deoxyribonucleic acids and producing 3'-phosphomonoesters",
+"3.1.21": "Endodeoxyribonucleases producing 5'-phosphomonoesters",
+"3.1.22": "Endodeoxyribonucleases producing other than 5'-phosphomonoesters",
+"3.1.23": "Site specific endodeoxyribonucleases: cleavage is sequence specific",
+"3.1.24": "Site specific endodeoxyribonucleases: cleavage is not sequence specific",
+"3.1.25": "Site-specific endodeoxyribonucleases specific for altered bases",
+"3.1.26": "Endoribonucleases producing 5'-phosphomonoesters",
+"3.1.27": "Endoribonucleases producing other than 5'-phosphomonoesters",
+"3.1.30": "Endoribonucleases active with either ribo- or deoxyribonucleic acids and producing 5'-phosphomonoesters",
+"3.1.31": "Endoribonucleases active with either ribo- or deoxyribonucleic acids and producing 3'-phosphomonoesters",
+"3.2": "Glycosylases",
+"3.2.1": "Glycosidases, i.e. enzymes hydrolyzing O- and S-glycosyl compounds",
+"3.2.2": "Hydrolyzing N-glycosyl compounds",
+"3.2.3": "Hydrolyzing S-glycosyl compounds",
+"3.3": "Acting on ether bonds",
+"3.3.1": "Thioether and trialkylsulfonium hydrolases",
+"3.3.2": "Ether hydrolases",
+"3.4": "Acting on peptide bonds (peptidases)",
+"3.4.1": "alpha-Amino-acyl-peptide hydrolases",
+"3.4.2": "Peptidyl-amino-acid hydrolases",
+"3.4.3": "Dipeptide hydrolases",
+"3.4.4": "Peptidyl peptide hydrolases",
+"3.4.11": "Aminopeptidases",
+"3.4.12": "Peptidylamino-acid hydrolases or acylamino-acid hydrolases",
+"3.4.13": "Dipeptidases",
+"3.4.14": "Dipeptidyl-peptidases and tripeptidyl-peptidases",
+"3.4.15": "Peptidyl-dipeptidases",
+"3.4.16": "Serine-type carboxypeptidases",
+"3.4.17": "Metallocarboxypeptidases",
+"3.4.18": "Cysteine-type carboxypeptidases",
+"3.4.19": "Omega peptidases",
+"3.4.21": "Serine endopeptidases",
+"3.4.22": "Cysteine endopeptidases",
+"3.4.23": "Aspartic endopeptidases",
+"3.4.24": "Metalloendopeptidases",
+"3.4.25": "Threonine endopeptidases",
+"3.4.26": "Glutamic endopeptidases",
+"3.4.99": "Endopeptidases of unknown catalytic mechanism",
+"3.5": "Acting on carbon-nitrogen bonds, other than peptide bonds",
+"3.5.1": "In linear amides",
+"3.5.2": "In cyclic amides",
+"3.5.3": "In linear amidines",
+"3.5.4": "In cyclic amidines",
+"3.5.5": "In nitriles",
+"3.5.99": "In other compounds",
+"3.6": "Acting on acid anhydrides",
+"3.6.1": "In phosphorus-containing anhydrides",
+"3.6.2": "In sulfonyl-containing anhydrides",
+"3.6.3": "Acting on acid anhydrides; catalyzing transmembrane movement of substances",
+"3.6.4": "Acting on ATP; involved in cellular and subcellular movement",
+"3.6.5": "Acting on GTP; involved in cellular and subcellular movement",
+"3.7": "Acting on carbon-carbon bonds",
+"3.7.1": "In ketonic substances",
+"3.8": "Acting on halide bonds",
+"3.8.1": "In C-halide compounds",
+"3.8.2": "In P-halide compounds",
+"3.9": "Acting on phosphorus-nitrogen bonds",
+"3.9.1": "Acting on phosphorus-nitrogen bonds",
+"3.10": "Acting on sulfur-nitrogen bonds",
+"3.10.1": "Acting on sulfur-nitrogen bonds",
+"3.11": "Acting on carbon-phosphorus bonds",
+"3.11.1": "Acting on carbon-phosphorus bonds",
+"3.12": "Acting on sulfur-sulfur bonds",
+"3.12.1": "Acting on sulfur-sulfur bonds",
+"3.13": "Acting on carbon-sulfur bonds",
+"3.13.1": "Acting on carbon-sulfur bonds",
+"3.13.2": "Thioether and trialkylsulfonium hydrolases",
+"4": "Lyases",
+"4.1": "Carbon-carbon lyases",
+"4.1.1": "Carboxy-lyases",
+"4.1.2": "Aldehyde-lyases",
+"4.1.3": "Oxo-acid-lyases",
+"4.1.99": "Other carbon-carbon lyases",
+"4.2": "Carbon-oxygen lyases",
+"4.2.1": "Hydro-lyases",
+"4.2.2": "Acting on polysaccharides",
+"4.2.3": "Acting on phosphates",
+"4.2.99": "Other carbon-oxygen lyases",
+"4.3": "Carbon-nitrogen lyases",
+"4.3.1": "Ammonia-lyases",
+"4.3.2": "Lyases acting on amides, amidines, etc",
+"4.3.3": "Amine-lyases",
+"4.3.99": "Other carbon-nitrogen lyases",
+"4.4": "Carbon-sulfur lyases",
+"4.4.1": "Carbon-sulfur lyases",
+"4.5": "Carbon-halide lyases",
+"4.5.1": "Carbon-halide lyases",
+"4.6": "Phosphorus-oxygen lyases",
+"4.6.1": "Phosphorus-oxygen lyases",
+"4.7": "Carbon-phosphorus lyases",
+"4.7.1": "Carbon-phosphorus lyases",
+"4.8": "Nitrogen-oxygen lyases",
+"4.8.1": "Hydro-lyases",
+"4.98": "ATP-independent chelatases",
+"4.98.1": "Forming coordination complexes",
+"4.99": "Other lyases",
+"4.99.1": "Other lyases",
+"5": "Isomerases",
+"5.1": "Racemases and epimerases",
+"5.1.1": "Acting on amino acids and derivatives",
+"5.1.2": "Acting on hydroxy acids and derivatives",
+"5.1.3": "Acting on carbohydrates and derivatives",
+"5.1.99": "Acting on other compounds",
+"5.2": "Cis-trans-isomerases",
+"5.2.1": "Cis-trans isomerases",
+"5.3": "Intramolecular oxidoreductases",
+"5.3.1": "Interconverting aldoses and ketoses",
+"5.3.2": "Interconverting keto- and enol-groups",
+"5.3.3": "Transposing C=C bonds",
+"5.3.4": "Transposing S-S bonds",
+"5.3.99": "Other intramolecular oxidoreductases",
+"5.4": "Intramolecular transferases",
+"5.4.1": "Transferring acyl groups",
+"5.4.2": "Phosphotransferases (phosphomutases)",
+"5.4.3": "Transferring amino groups",
+"5.4.4": "Transferring hydroxy groups",
+"5.4.99": "Transferring other groups",
+"5.5": "Intramolecular lyases",
+"5.5.1": "Intramolecular lyases",
+"5.6": "Isomerases altering macromolecular conformation",
+"5.6.1": "Enzymes altering polypeptide conformation or assembly",
+"5.6.2": "Enzymes altering nucleic acid conformation",
+"5.99": "Other isomerases",
+"5.99.1": "Other isomerases",
+"6": "Ligases",
+"6.1": "Forming carbon-oxygen bonds",
+"6.1.1": "Ligases forming aminoacyl-tRNA and related compounds",
+"6.1.2": "Acid--alcohol ligases (ester synthases)",
+"6.1.3": "Cyclo-ligases",
+"6.2": "Forming carbon-sulfur bonds",
+"6.2.1": "Acid--thiol ligases",
+"6.2.2": "Amide--thiol ligases",
+"6.3": "Forming carbon-nitrogen bonds",
+"6.3.1": "Acid--ammonia (or amine) ligases (amide synthases)",
+"6.3.2": "Acid--amino-acid ligases (peptide synthases)",
+"6.3.3": "Cyclo-ligases",
+"6.3.4": "Other carbon--nitrogen ligases",
+"6.3.5": "Carbon--nitrogen ligases with glutamine as amido-N-donor",
+"6.4": "Forming carbon-carbon bonds",
+"6.4.1": "Forming carbon-carbon bonds",
+"6.5": "Forming phosphoric ester bonds",
+"6.5.1": "Forming phosphoric ester bonds",
+"6.6": "Forming nitrogen-metal bonds",
+"6.6.1": "Forming coordination complexes",
+"6.7": "Forming nitrogen-nitrogen bonds",
+"6.7.1": "Forming diazo bonds",
+"7": "Translocases",
+"7.1": "Catalysing the translocation of hydrons",
+"7.1.1": "Hydron translocation or charge separation linked to oxidoreductase reactions",
+"7.1.2": "Hydron translocation linked to the hydrolysis of a nucleoside triphosphate",
+"7.1.3": "Hydron translocation linked to the hydrolysis of diphosphate",
+"7.2": "Catalysing the translocation of inorganic cations",
+"7.2.1": "Linked to oxidoreductase reactions",
+"7.2.2": "Linked to the hydrolysis of a nucleoside triphosphate",
+"7.2.3": "Linked to the hydrolysis of diphosphate",
+"7.2.4": "Linked to decarboxylation",
+"7.3": "Catalysing the translocation of inorganic anions and their chelates",
+"7.3.2": "Linked to the hydrolysis of a nucleoside triphosphate",
+"7.4": "Catalysing the translocation amino acids and peptides",
+"7.4.2": "Linked to the hydrolysis of a nucleoside triphosphate",
+"7.5": "Catalysing the translocation carbohydrates and their derivatives",
+"7.5.2": "Linked to the hydrolysis of a nucleoside triphosphate",
+"7.6": "Catalysing the translocation of other compounds",
+"7.6.2": "Linked to the hydrolysis of a nucleoside triphosphate"
+}

data/models/ec/prothgt-ec-config.yaml

@@ -0,0 +1,27 @@
+{
+    "lr": 0.001,
+    "mlp_lr": 0.005,
+    "wd": 5e-05,
+    "mlp_wd": 0.0,
+    "epochs": 80,
+    "tr_batch_size": 1024,
+    "hidden_channels": 256,
+    "num_heads": 16,
+    "num_layers": 2,
+    "num_neighbors": [32],
+    "disjoint_train_ratio": 0.3,
+    "mlp_dropout": 0.1,
+    "grad_clip_norm": 5.0,
+    "eval_threshold": 0.5,
+    "neg_sampling_ratio_l1": 6,
+    "pos_weight_l1": 6,
+    "neg_sampling_ratio_l2": 70,
+    "pos_weight_l2": 70,
+    "neg_sampling_ratio_l3": 100,
+    "pos_weight_l3": 100,
+    "neg_sampling_ratio_l4": 100,
+    "pos_weight_l4": 100,
+    "test_on_cpu": true,
+    "checkpoint_interval": 20,
+    "val_eval_num_proteins": 500
+}

data/models/ec/prothgt-ec-model.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e6a4f28a81b46c259c9964b9502dbece97a51ca9b6db371fce0e2684ae39272e
+size 44855705

run_prothgt_ec_app.py

@@ -0,0 +1,226 @@
+"""EC number prediction backend for the ProtHGT Space.
+
+Unlike GO term prediction (one flat decoder per sub-ontology), EC prediction uses
+a single joint model with a shared HGT backbone (plus a normalized skip
+connection) and four level-specific decoders, run as a top-down cascade so the
+predicted EC levels form a consistent hierarchy.
+"""
+import os
+import gzip
+import json
+import shutil
+from functools import lru_cache
+
+import torch
+from torch.nn import Linear, LayerNorm
+from torch_geometric.nn import HGTConv, MLP
+import pandas as pd
+import yaml
+from huggingface_hub import hf_hub_download
+
+from run_prothgt_app import load_name_info
+
+EC_MODELS_DIR = os.path.join("data", "models", "ec")
+EC_CONFIG_PATH = os.path.join(EC_MODELS_DIR, "prothgt-ec-config.yaml")
+EC_MODEL_PATH = os.path.join(EC_MODELS_DIR, "prothgt-ec-model.pt")
+EC_KG_PATH = os.path.join("data", "prothgt-ec-kg.pt")
+# EC class/subclass/sub-subclass (L1-L3) names from the ExPASy ENZYME nomenclature;
+# L4 (full EC number) names come from name_info instead.
+EC_CLASS_NAMES_PATH = os.path.join("data", "ec_class_names.json")
+
+LEVELS = ['l1', 'l2', 'l3', 'l4']
+LEVEL_TARGET_TYPE = {'l1': 'EC_class_1', 'l2': 'EC_class_2', 'l3': 'EC_class_3', 'l4': 'EC_number'}
+LEVEL_LABEL = {'l1': 'L1', 'l2': 'L2', 'l3': 'L3', 'l4': 'L4'}
+CHILDREN_KEY = {'l2': 'l1_children', 'l3': 'l2_children', 'l4': 'l3_children'}
+
+# EC taxonomy edges used to find each node's children for the cascade.
+CHILD_EDGE_TYPES = {
+    'l1_children': ('EC_class_1', 'rev_ec_parent_1', 'EC_class_2'),
+    'l2_children': ('EC_class_2', 'rev_ec_parent_2', 'EC_class_3'),
+    'l3_children': ('EC_class_3', 'rev_ec_parent_3', 'EC_number'),
+}
+
+@lru_cache(maxsize=1)
+def load_ec_class_names(path=EC_CLASS_NAMES_PATH):
+    """{EC class prefix (L1-L3) -> name}, e.g. '2.7.11' -> 'Protein-serine/threonine kinases'."""
+    with open(path, encoding='utf-8') as f:
+        return json.load(f)
+
+
+# ---------------------------------------------------------------------------
+# Model (ported from the ProtHGT src/ec package)
+# ---------------------------------------------------------------------------
+class ProtHGT(torch.nn.Module):
+    """HGT backbone with a normalized pre/post-GNN skip connection."""
+    def __init__(self, data, hidden_channels, num_heads, num_layers, mlp_hidden_layers, mlp_dropout):
+        super().__init__()
+        self.lin_dict = torch.nn.ModuleDict()
+        for node_type in data.node_types:
+            self.lin_dict[node_type] = Linear(data[node_type].x.size(1), hidden_channels)
+
+        self.convs = torch.nn.ModuleList()
+        for _ in range(num_layers):
+            self.convs.append(HGTConv(hidden_channels, hidden_channels, data.metadata(), num_heads, group='sum'))
+
+        self.initial_norm = LayerNorm(hidden_channels)
+        self.gnn_norm = LayerNorm(hidden_channels)
+        self.mlp = MLP(mlp_hidden_layers, dropout=mlp_dropout, norm=None)
+
+    def generate_embeddings(self, x_dict, edge_index_dict):
+        initial_dict = {nt: self.lin_dict[nt](x).relu_() for nt, x in x_dict.items()}
+        gnn_dict = initial_dict
+        for conv in self.convs:
+            out = conv(gnn_dict, edge_index_dict)
+            gnn_dict = {nt: (out[nt] if out.get(nt) is not None else prev)
+                        for nt, prev in gnn_dict.items()}
+        return {nt: torch.cat([self.initial_norm(initial_dict[nt]), self.gnn_norm(gnn_dict[nt])], dim=-1)
+                for nt in initial_dict}
+
+
+class ProtHGTHier(torch.nn.Module):
+    """Shared ProtHGT backbone + one MLP decoder per EC level."""
+    def __init__(self, data, hidden_channels, num_heads, num_layers, mlp_hidden_layers, mlp_dropout):
+        super().__init__()
+        self.backbone = ProtHGT(data, hidden_channels, num_heads, num_layers, mlp_hidden_layers, mlp_dropout)
+        self.backbone.mlp = torch.nn.Identity()
+        self.decoders = torch.nn.ModuleDict({
+            level: MLP(mlp_hidden_layers, dropout=mlp_dropout, norm=None) for level in LEVELS
+        })
+
+    def generate_embeddings(self, x_dict, edge_index_dict):
+        return self.backbone.generate_embeddings(x_dict, edge_index_dict)
+
+
+def load_hierarchy_masks(data):
+    """{'l1_children': {ec_class_1_idx: [ec_class_2_idx, ...]}, ...} from the EC taxonomy edges."""
+    from collections import defaultdict
+    masks = {}
+    for name, et in CHILD_EDGE_TYPES.items():
+        row, col = data[et].edge_index
+        children = defaultdict(list)
+        for r, c in zip(row.tolist(), col.tolist()):
+            children[r].append(c)
+        masks[name] = dict(children)
+    return masks
+
+
+# ---------------------------------------------------------------------------
+# Loading
+# ---------------------------------------------------------------------------
+def load_ec_kg(output=EC_KG_PATH):
+    """Download (if needed) and load the EC knowledge graph. Callers should cache the result."""
+    if not os.path.exists(output):
+        cached_path = hf_hub_download(
+            repo_id="HUBioDataLab/ProtHGT",
+            filename="knowledge_graphs/ec/prothgt-ec-kg.pt",
+            repo_type="dataset",
+        )
+        os.makedirs(os.path.dirname(output), exist_ok=True)
+        shutil.copy(cached_path, output)
+        print(f"EC KG downloaded to {output}")
+    print('Loading EC knowledge graph...')
+    return torch.load(output)
+
+
+def load_ec_model(heterodata, config_path=EC_CONFIG_PATH, model_path=EC_MODEL_PATH):
+    """Build ProtHGTHier and load its weights. Callers should cache the result."""
+    with open(config_path) as f:
+        cfg = yaml.safe_load(f)
+    h = cfg['hidden_channels']
+    model = ProtHGTHier(
+        heterodata, hidden_channels=h, num_heads=cfg['num_heads'], num_layers=cfg['num_layers'],
+        mlp_hidden_layers=[4 * h, h, 1], mlp_dropout=cfg['mlp_dropout'],
+    )
+    device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
+    model.load_state_dict(torch.load(model_path, map_location=device))
+    model.eval()
+    print(f'Loaded EC model weights from {model_path}')
+    return model
+
+
+# ---------------------------------------------------------------------------
+# Cascade prediction
+# ---------------------------------------------------------------------------
+@torch.no_grad()
+def _cascade_protein(model, x_dict, masks, protein_idx, inv_maps, threshold, device):
+    """Top-down cascade for one protein: L1 over all classes, each deeper level only
+    over the children of the kept nodes above. Returns (level_key, ec_string, prob) rows."""
+    rows = []
+    kept_prev = []
+    for level in LEVELS:
+        target_type = LEVEL_TARGET_TYPE[level]
+        if level == 'l1':
+            candidates = list(range(x_dict[target_type].size(0)))
+        else:
+            ckey = CHILDREN_KEY[level]
+            candidates = sorted({c for p in kept_prev for c in masks[ckey].get(p, [])})
+        if not candidates:
+            break
+
+        cand = torch.tensor(candidates, dtype=torch.long, device=device)
+        p_emb = x_dict['Protein'][protein_idx].unsqueeze(0).expand(len(candidates), -1)
+        z = torch.cat([p_emb, x_dict[target_type][cand]], dim=-1)
+        probs = torch.sigmoid(model.decoders[level](z).view(-1))
+
+        keep = (probs > threshold).nonzero(as_tuple=False).view(-1)
+        kept_prev = [candidates[i] for i in keep.tolist()]
+        for i in keep.tolist():
+            rows.append((level, inv_maps[target_type][candidates[i]], float(probs[i])))
+    return rows
+
+
+def generate_ec_prediction_df(protein_ids, heterodata, model, threshold=0.5, progress_callback=None):
+    """Cascade EC predictions for the given proteins. The backbone embeddings are
+    computed once over the whole graph; the decoders then run per protein."""
+    if isinstance(protein_ids, str):
+        protein_ids = [protein_ids]
+
+    # The full-graph forward over the whole KG is memory-heavy for this model
+    # (hidden=256); fall back to CPU if a GPU runs out of memory.
+    device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
+
+    def _embed(dev):
+        model.to(dev)
+        hd = heterodata.to(dev)
+        with torch.no_grad():
+            return hd, model.generate_embeddings(hd.x_dict, hd.edge_index_dict)
+
+    try:
+        heterodata, x_dict = _embed(device)
+    except RuntimeError as e:
+        if device.type == 'cuda' and 'out of memory' in str(e).lower():
+            print('GPU out of memory during the full-graph forward; falling back to CPU...')
+            torch.cuda.empty_cache()
+            device = torch.device('cpu')
+            heterodata, x_dict = _embed(device)
+        else:
+            raise
+    masks = load_hierarchy_masks(heterodata)
+    inv_maps = {
+        LEVEL_TARGET_TYPE[level]: {idx: tok for tok, idx in heterodata[LEVEL_TARGET_TYPE[level]]['id_mapping'].items()}
+        for level in LEVELS
+    }
+    id_mapping = heterodata['Protein']['id_mapping']
+    name_info = load_name_info()
+    ec_l4_names = name_info['EC_number']        # full EC number (L4) -> enzyme name
+    ec_class_names = load_ec_class_names()      # L1-L3 class prefix -> class name
+
+    rows = []
+    for i, pid in enumerate(protein_ids):
+        if progress_callback is not None:
+            progress_callback(i, len(protein_ids))
+        protein_name = name_info['Protein'].get(pid, pid)
+        for level, token, prob in _cascade_protein(
+                model, x_dict, masks, id_mapping[pid], inv_maps, threshold, device):
+            ec_name = ec_l4_names.get(token, '') if level == 'l4' else ec_class_names.get(token, '')
+            rows.append({
+                'UniProt_ID': pid,
+                'Protein': protein_name,
+                'EC': token,
+                'EC_name': ec_name,
+                'EC_level': LEVEL_LABEL[level],
+                'Probability': prob,
+            })
+
+    df = pd.DataFrame(rows, columns=['UniProt_ID', 'Protein', 'EC', 'EC_name', 'EC_level', 'Probability'])
+    return heterodata, df