Using QSARs for predictions in drug delivery

Abstract

Drug delivery research is an inherently empirical process, however high-throughput approaches could take advantage of understanding drug/material interactions such as from electrostatic, hydrophobic, or other non-covalent interactions between therapeutic molecules and a drug delivery polymer. Cyclodextrin polymers have been investigated for drug delivery specifically due to their capacity to exploit this affinity interaction to change the rate of drug release. Testing drug candidates; however, for affinity is time-consuming, making computational predictions more effective. One option, molecular docking programs, provide predictions of affinity, but lack reliability, as their accuracy with cyclodextrin remains unverified experimentally. Alternatively, quantitative structure-activity relationship models (QSARs), which analyze statistical relationships between molecular properties, appear more promising. Previously constructed QSARs for cyclodextrin are not publicly available, necessitating an openly accessible model. Around 600 experimental affinities between cyclodextrin and guest molecules were cleaned and imported from published research. The software PaDEL-Descriptor calculated over 1000 chemical descriptors for each molecule, which were then analyzed in R to create several QSARs with different statistical methods. These QSARs proved highly time efficient, calculating in minutes what docking programs would take hours to accomplish. Additionally, on test sets, QSARs reached R2 values of around 0.7-0.8. The speed, accuracy, and accessibility of these QSARs improve evaluation of individual drugs and facilitate screening of large datasets for potential candidates in cyclodextrin affinity-based delivery systems. An app was built to rapidly access model predictions for end users using the shiny library in R. To demonstrate the usability for drug release planning, the QSAR predictions were coupled with a mechanistic model of diffusion within the app. Integrating new modules should provide an accessible approach to use other cheminformatic tools in the field of drug delivery.