{
  "TESTING": true,
  "CHEBI_VERSION": 252,
  "ENSEMBLE_CONFIG": "web",
  "ENSEMBLE_TYPE": "wmv-f1",
  "ENSEMBLE_DIR": null,
  "PR_CURVE": "data/pr_curve.csv",
  "INCONSISTENCY_RESOLUTION": "score-based",
  "BEST_MODEL": "gat-aug_chebi25-3star_v252",
  "MODEL_PRESENTATION": {
    "gat-aug_chebi25-3star_v252": {
      "name": "GAT-aug (ChEBI25-3STAR)",
      "description": "A Graph Attention Network (GAT) using a hierarchical graph structure with nodes for atoms and functional groups as well as a global node. Trained with <a href=\"https://github.com/ChEB-AI/python-chebai-graph\">ChEB-AI's graph extension</a> on the 3-STAR subset of ChEBI version 252."
    },
    "gat_chebi25-3star_v252": {
      "name": "GAT (ChEBI25-3STAR)",
      "description": "A Graph Attention Network (GAT) on the plain molecular graph. Trained with <a href=\"https://github.com/ChEB-AI/python-chebai-graph\">ChEB-AI's graph extension</a> on the 3-STAR subset of ChEBI version 252."
    },
    "resgated-aug_chebi25-3star_v252": {
      "name": "ResGated-aug (ChEBI25-3STAR)",
      "description": "A Residual-Gated Graph Convolutional Network (RGGCN) using a hierarchical graph structure with nodes for atoms and functional groups as well as a global node. Trained with <a href=\"https://github.com/ChEB-AI/python-chebai-graph\">ChEB-AI's graph extension</a> on the 3-STAR subset of ChEBI version 252."
    },
    "resgated_chebi25-3star_v252": {
      "name": "ResGated (ChEBI25-3STAR)",
      "description": "A Residual-Gated Graph Convolutional Network (RGGCN) on the plain molecular graph. Trained with <a href=\"https://github.com/ChEB-AI/python-chebai-graph\">ChEB-AI's graph extension</a> on the 3-STAR subset of ChEBI version 252."
    },
    "gat-aug_chebi25_v252": {
      "name": "GAT-aug (ChEBI25)",
      "description": "A Graph Attention Network (GAT) using a hierarchical graph structure with nodes for atoms and functional groups as well as a global node. Trained with <a href=\"https://github.com/ChEB-AI/python-chebai-graph\">ChEB-AI's graph extension</a> on all of ChEBI version 252, not only its 3-STAR subset."
    },
    "gat_chebi25_v252": {
      "name": "GAT (ChEBI25)",
      "description": "A Graph Attention Network (GAT) on the plain molecular graph. Trained with <a href=\"https://github.com/ChEB-AI/python-chebai-graph\">ChEB-AI's graph extension</a> on all of ChEBI version 252, not only its 3-STAR subset."
    },
    "c3p": {
      "name": "C3P",
      "description": "A generative-AI approach that writes Python programs for classifying ChEBI classes. See the <a href=\"https://link.springer.com/article/10.1186/s13321-025-01092-3\">C3P paper</a> for more details."
    },
    "chemlog": {
      "name": "ChemLog",
      "description": "A rule-based method for predicting peptide-related ChEBI classes, as well as classes that are defined by a given element (e.g., nitrogen molecular entity or organonitrogen compound). It uses a Monadic Second-Order Logic (MSOL) formalisation of peptides, which has been translated into a Python algorithm. More information is available on <a href=\"https://github.com/sfluegel05/chemlog-peptides\">GitHub</a>."
    },
    "lopster": {
      "name": "Lopster",
      "description": "The LOgic Programming for STructured Entities Reasoner classifies molecules into ChEBI classes using Answer Set Programming. Originally published by <a href=\"https://rdcu.be/e4etb\">Magka et al. (2014)</a>."
    },
    "lookup": {
      "name": "ChEBI Lookup",
      "description": "If the SMILES string is equivalent to an entry in ChEBI, the classification of that entry is retrieved. This model is not part of the ensemble calibration, so it votes with full trust and its influence is set by its model weight alone."
    }
  }
}