Determination of Dermal Permeability Coefficient (Kp) by Utilizing Multiple Descriptors in Artificial Neural Network Analysis and Multiple Regression Analysis

Aims: The permeability coefficient, or Kp, is an important descriptor for assessing dermal absorption of medicaments utilized for clinical treatment of various dermal accessible diseases. Determination of Kp by multiple descriptors by artificial neural network (ANN) and multiple regression is compared.
Study Design: The calculation of Kp utilizing multiple descriptors, and comparison of ANN and multiple regression is achieved.
Place and Duration of Study: Durham Science Center, Chemistry Department of the University of Nebraska, between April 2014 and July 2014.
Methodology: The calculation of Kp by previous methodologies is accomplished for a broad spectrum of medicinal and chemical compounds. The values Kp thus acquired are then compared to those obtained by ANN training and multiple regression analysis. Various other pharmaceutical based descriptors are then applied to ascertain the benefit of Kp determination by those properties.
Results: Training and determination of Kp by ANN showed that Log Ko/w and molecular weight (MW) utilized by conventional means is effective. However, ANN demonstrated the Kp determination by applying properties of Log Ko/w, MW, polar surface area, number of atoms, rotatable bonds, molecular volume, and atoms responsible for hydrogen bond donor and acceptors, are also effective and offer significant advantages. These advantages include the potential of encompassing many more molecular constitutional descriptors and molecular properties. Multiple regression showed clearly that the application of more descriptors for Kp determination increases the coefficient of determination (R2). Increased R2 shows an improved fit of the raw data to the model improved prediction.
Conclusion: Determination of Kp by applying various descriptors in addition to Log Ko/w and MW increases the model fit to the raw data. ANN prediction of Kp was more effective when using additional descriptors. Prediction of Kp by multiple regression was useful, and utilizing descriptors with Log Ko/w and MW improved the model fit to the raw data.