Prediction of drug's anatomical therapeutic chemical (ATC) code by constructing biological profiles of ATC codes

Abstract

Background: The Anatomical Therapeutic Chemical (ATC) classification system, proposed and maintained by the World Health Organization, is among the most widely used drug classification schemes. Recently, it has become a key research focus in drug repositioning. Computational models often pair drugs with ATC codes to explore drug-ATC code associations. However, the limited information available for ATC codes constrains these models, leaving significant room for improvement.

Results: This study presents an inference method to identify highly related target proteins, structural features, and side effects for each ATC code, constructing comprehensive biological profiles. Association networks for target proteins, structural features, and side effects are established, and a random walk with restart algorithm is applied to these networks to extract raw associations. A permutation test is then conducted to exclude false positives, yielding robust biological profiles for ATC codes. These profiles are used to construct new ATC code kernels, which are integrated with ATC code kernels from the existing model PDATC-NCPMKL. The recommendation matrix is subsequently generated using the procedures of PDATC-NCPMKL. Cross-validation results demonstrate that the new model achieves AUROC and AUPR values exceeding 0.96.

Conclusion: The proposed model outperforms PDATC-NCPMKL and other previous models. Analysis of the contributions of the newly added ATC code kernels confirms the value of biological profiles in enhancing the prediction of drug-ATC code associations.

Keywords: Anatomical therapeutic chemical code; Biological profiles; Drug repositioning; Network consistency projection; Random walk with restart.

Fig. 1

Procedures for inferring biological profiles of ATC codes. Six different association types are employed to construct target protein-related, fingerprint-related, and side effect-related association networks. The random walk with restart algorithm is adopted to infer related target proteins, structural features, and side effects for each ATC code, followed by a permutation test to control false positives. The final biological profiles of ATC codes include highly related target proteins, structural features, and side effects

Fig. 2

An illustration on the network in each group. A Target protein-related association network; B Fingerprint-related association network; C Side effect-related association network

Fig. 3

Procedures of the PDATC-NCPMKL-updated. This model is an updated version of PDATC-NCPMKL. Based on the biological profiles of ATC codes, include highly related target proteins, structural features, side effects, three new ATC code kernels ($K_{TP,a}^{i,\ast }$, $K_{FP,a}^{i,\ast }$, and $K_{SE,a}^{i,\ast }$) are built. These new ATC code kernels are fused with those in PDATC-NCPMKL to yield novel unified ATC code kernel. The recommendation matrix is produced with the same following procedures of PDATC-NCPMKL. Please refer to Additional file 2 for drug kernels ($K_{TP,d}^{i,\ast }$, $K_{FP,d}^{i,\ast }$, $K_{IN,d}^i$, $K_{F,d}^{i,\ast }$, and $K_{SE,d}^{i,\ast }$) and ATC code kernels ($K_{F,a}^{i,\ast }$, $K_{\mathit{SPro}}^i$, and $K_{\mathit{SM}}^i$) in PDATC-NCPMKL

Fig. 4

Bar chart to show the distribution of ATC codes at three levels based on numbers of their highly related target proteins, structural features, and side effects. A Bar chart for ATC codes at the second level; B Bar chart for ATC codes at the third level; C Bar chart for ATC codes at the fourth level. The number above each bar represents the number of ATC codes having corresponding number of highly related entities. For example, 56 in (A) suggests that there are 56 ATC codes at the second level having 1–100 highly related target proteins

Fig. 5

ROC and PR curves to show the performance of PDATC-NCPMKL-updated. A ROC curves; B PR curves. The AUROC and AUPR values are all higher than 0.96, suggesting the high performance of PDATC-NCPMKL-updated

Fig. 6

Bar chart to compare the performance of various models for the prediction of drug-ATC code associations. A Bar chart for AUROC; B Bar chart for AUPR. The new model PDATC-NCPMKL-updated yields the best performance

Fig. 7

Bar chart to compare the performance of various models only using drug structure information for the prediction of drug-ATC code associations. A Bar chart for AUROC; B Bar chart for AUPR. The model with fingerprint-related kernels, derived from PDATC-NCPMKL-updated, provides competitive performance