Table 2 Summary of key ATC prediction methods from the literature
From: Predicting anatomic therapeutic chemical classification codes using tiered learning
Method | Method details | |||||
|---|---|---|---|---|---|---|
Data Set Size (compounds) | Data Source(s) | Descriptors | Prediction Algorithm | Accuracy at ATC Depth 1 | Maximum Accuracy and Prediction Depth | |
SuperPred [8] | 2650 (for drug classification) | Transformer database, SuperTarget, ChEMBL, and BindingDB | 2D, fragment, and 3D Structure-based | Consensus-based | 80.90% | 75.1% at a depth of 5 |
Chen et al. [5] | 3934 | KEGG | Chemical interactions, structure and ontology | Hybrid Method | 75.70% (internal validation set) | 75.70% (internal validation set) |
Wang et al. [11] | 790 | KEGG BRITE, DrugBank | Information from chemical structures, target proteins, and ATC Codes. | Kernel method and SVM classification | 74% | 74% at depth 5. |
Gurulingappa et al. [20] | 504 (training + test) | Medline | Concepts generated from Medline terms | Naïve Bayes | 77.12% | 77.12% at depth 4 |