Systematic drug repositioning based on clinical side-effects

Abstract

Drug repositioning helps fully explore indications for marketed drugs and clinical candidates. Here we show that the clinical side-effects (SEs) provide a human phenotypic profile for the drug, and this profile can suggest additional disease indications. We extracted 3,175 SE-disease relationships by combining the SE-drug relationships from drug labels and the drug-disease relationships from PharmGKB. Many relationships provide explicit repositioning hypotheses, such as drugs causing hypoglycemia are potential candidates for diabetes. We built Naïve Bayes models to predict indications for 145 diseases using the SEs as features. The AUC was above 0.8 in 92% of these models. The method was extended to predict indications for clinical compounds, 36% of the models achieved AUC above 0.7. This suggests that closer attention should be paid to the SEs observed in trials not just to evaluate the harmful effects, but also to rationally explore the repositioning potential based on this "clinical phenotypic assay".

Figure 1. Constructing and visualization of the disease-SE associations.

a) Confusion matrix of using SE priapism to predict Parkinson Disease. PD: Parkinson Disease; MCC: Matthews correlation coefficient; TP, FP, TN and FN stand for true positives, false positives, true negatives and false negatives respectively. This confusion matrix represents one disease-SE pair. b) The overall layout of the disease-SE network. Diseases and SEs are shown in red and white circles respectively. The edge color and the width indicate the association strength as measured by MCC. The neuropsychiatric, neoplasm, circulatory- system disease dominated clusters are highlighted in yellow, red and grey rectangles respectively. c) Neuropsychiatric disease-dominated cluster. The SE tardive dyskinesia and priapism is highlighted in orange. The MCC for PD-priapism pair is 0.47 according to the confusion matrix in a) and is visualized as an orange line (see black arrow from a) to c)) . d) Sample MCC, sensitivity and specificity measures for using priapism to predict diseases. For example, 86% of the drugs that treat obsessive-compulsive disorder (OCD) list priapism as a side effect; whereas only 7% (1-sp) of the drugs not reported to treat OCD list this SE.

Figure 2. A schematic figure of mapping drug structure to side effect and then onto a disease indication.

a) Train the 566 SE models. For SE_j, the (+) and (−) were recruited from 888 SIDER molecules. b) The disease_i-molecule_k association (Θ_ik) was calculated as the dot product value of the disease-SE association vector (DS) and SE-molecule association vector (SM). The binary SE-molecule (SM) association was calculated from QSAR models. The width of the colored lines indicates the weights of the disease-SE associations. As an example, Θ_i2 is more than Θ_i1 as the association of side effect j in green to disease i is stronger.

Figure 3. Predict drugs' repositioning potential for hypertension via DRoSEf.

a) The distribution of the Θ score for the positive (red) and negative (blue) set for hypertension. The molecules with high Θ score in negative set (red square bracket) were chosen as the candidates for treating hypertension. b) The ROC curve of using Θ score to predict hypertension. The AUC is 0.74. c) Predicted relationships of the top molecules with the 12 SEs and the association of these SEs with the hypertension. The binary association among molecules and SEs is in grey lines. The association strength between SE and disease is reflected in the color and the width of the edge. Postural hypotension is highlighted as the SE explicitly linked to hypertension.