Docking Scores As Descriptors To Predict Human Serum Albumin Binding Affinity

Pharmacokinetic properties of a compound are the determinant factors for achieving its biological role in organisms. Inappropriate ADME properties were the most significant cause of drug withdrawn from development and account for 40% of all the reasons. To reduce drug attrition effectively, cut down the cost of drug discovery and development, and measure properties of drug lead in high throughput ways, in silico ADME modeling is a good choice.For the present, ADME properties are most often estimated from the structural properties of a compound by the quantitative structural activity relationship (QSAR) approach. Previous studies had shown that molecular docking, capable of analyzing compound-protein interaction, could be used to make categorical estimation of a pharmacokinetic property. However, there were few researches to make numeric estimation. Therefore, this research takes human serum albumin (HSA) binding affinity as an example to show that docking descriptors might also be useful to estimate the exact value of a pharmacokinetic property.First, we used traditional compound structure descriptors to predict Human Serum Binding Affinity (log K'HSA). The model showed a result of Q2= 0.84, MSE =0.06; r2= 0.83, mse= 0.06 and R2= 0.87, mse= 0.05. This result was a little better than previously reported model and showed our method was able to reproduce the accuracy reported before.Second, we used docking scores as descriptors to predict log K'HSA. The final result were r2= 0.78, mse= 0.09; Q2= 0.75, MSE= 0.09; R2= 0.79, mse= 0.07. This accuracy was comparable to the known QSAR model based on compound descriptors and proved that docking scores can be used as descriptors in numeric prediction. Further, our research showed that a descent account of protein flexibility was essential to get better prediction model.