Purpose
Intravaginal drug administration offers a notable alternative to traditional oral delivery methods, allowing for precise targeting of effects both locally and systemically. In response to this, there has been a notable increase in the development of advanced in silico techniques for predicting drug permeability. These methods prove to be advantageous by bypassing the lengthy and resource-intensive processes typically associated with in vitro and in vivo experiments.
Methods
This particular study delved into the creation of in silico models specifically tailored for predicting vaginal permeability. The models were meticulously constructed using SMILES descriptors and local molecular graph invariants, ensuring a conformation-independent QSAR model. Leveraging a Monte Carlo optimization strategy, the models were iteratively refined across three distinct molecular splits for training and testing purposes.
Results
For the best developed QSAR model following statistical parameters were obtained for training set r2 = 0.7152, CCC = 0.8340, IIC = 0.7572, q2 = 0.7011, RMSE = 0.0055, MAE = 0.0044 and F = 196; and for test set r2 = 0.8657, CCC = 0.8902, IIC = 0.6180, q2 = 0.8412, Rm2 = 0.6722, RMSE = 0.0040, MAE = 0.0030 and F = 168.
Conclusions
These results underscored the exceptional predictive capabilities and robustness of the QSAR models developed in this study. Furthermore, the analysis pinpointed key molecular fragments derived from SMILES descriptors that significantly influence placental permeability. Given the prevalence of SMILES notation in most molecular databases, these well-constructed QSAR models can effectively serve as a rapid and precise screening tool for evaluating vaginal permeability.