Predicting drug side-effect profiles: a chemical fragment-based approach

Abstract

Background: Drug side-effects, or adverse drug reactions, have become a major public health concern. It is one of the main causes of failure in the process of drug development, and of drug withdrawal once they have reached the market. Therefore, in silico prediction of potential side-effects early in the drug discovery process, before reaching the clinical stages, is of great interest to improve this long and expensive process and to provide new efficient and safe therapies for patients.

Results: In the present work, we propose a new method to predict potential side-effects of drug candidate molecules based on their chemical structures, applicable on large molecular databanks. A unique feature of the proposed method is its ability to extract correlated sets of chemical substructures (or chemical fragments) and side-effects. This is made possible using sparse canonical correlation analysis (SCCA). In the results, we show the usefulness of the proposed method by predicting 1385 side-effects in the SIDER database from the chemical structures of 888 approved drugs. These predictions are performed with simultaneous extraction of correlated ensembles formed by a set of chemical substructures shared by drugs that are likely to have a set of side-effects. We also conduct a comprehensive side-effect prediction for many uncharacterized drug molecules stored in DrugBank, and were able to confirm interesting predictions using independent source of information.

Conclusions: The proposed method is expected to be useful in various stages of the drug development process.

Figure 1

Characteristics of side-effect data. The left panel shows the index-plot of the number of associated drugs for each side-effect, and the right panel shows the histogram of the number of associated drugs for each side-effect.

Figure 2

ROC curves in the 5-fold cross-validation. Comparison of the performance between nearest neighbor (NN), support vector machine (SVM), ordinary canonical correlation analysis (OCCA) and sparse canonical correlation analysis (SCCA).

Figure 3

Boxplot of the prediction accuracy of predicted side-effects for each drug. Prediction accuract of top 10 ranked predictions (top panels) and top 100 ranked predictions (bottom panels). Comparison of the performance between nearest neighbor (NN), support vector machine (SVM), ordinary canonical correlation analysis (OCCA) and sparse canonical correlation analysis (SCCA).

Figure 4

Boxplot of the AUC (under the ROC curve) scores for individual side-effects. Comparison of the performance between nearest neighbor (NN), support vector machine (SVM), ordinary canonical correlation analysis (OCCA) and sparse canonical correlation analysis (SCCA).

Figure 5

Index-plot of weight vectors for drug substructures and side-effects in OCCA. Index-plot of weight vectors for drug substructures (left) and side-effects (right) extracted by ordinary canonical correlation analysis (OCCA).

Figure 6

Index-plot of weight vectors for drug substructures and side-effects in SCCA. Index-plot of weight vectors for drug substructures (left) and side-effects (right) extracted by sparse canonical correlation analysis (SCCA).

Figure 7

Computational cost. Total execution time of the cross-validation experiment for the four methods (log10 scale).

Figure 8

Nitrogen-containing rings of size 5. (A) Porphyrin group, (B) Proline residue, (C) Histidine residue, (D) Tryptophane residue.

Figure 9

Chemical structure of risperidone. Two dimensional graph structure of risperidone.

Figure 10

Rimonabant substructure selected by the proposed method to be a clue of psychoacticity. The substructure of Rimonabant is selected to be a clue of psychoacticity.