Quantitative Nanostructure–Activity Relationship modelling of nanoparticles

Abstract

In recent years, the interest of the pharmaceutical industry to explore the use of nanoparticles for disease treatment or drug delivery has increased the need to evaluate their therapeutic efficacy or toxicity. However, evaluation of such properties using experimental means is time-consuming and costly. Thus, researchers are investigating the potential of using Quantitative Nanostructure–Activity Relationship (QNAR) models to predict the properties of nanoparticles prior to their synthesis. In this study, we developed a reliable, user-friendly and freely-accessible QNAR model to predict the cellular uptake of 105 nanoparticles with a single metal core by pancreatic cancer cells. Four modelling methods, namely Naïve Bayes, Logistic Regression, k nearest neighbour and support vector machine, were used to develop candidate models. A final consensus model was then developed using the top 5 candidate models. Validation of the final consensus model was done using a rigorous process by repeating the entire model development process five times using different combinations of training and validation sets. The final consensus model had a sensitivity of 86.7 to 98.2% and specificity of 67.3 to 76.6%. The majority of the wrong predictions were due to nanoparticles which had OC–O–CO bonding. Descriptors that were included in the final consensus models were mainly related to lipophilicity and hydrogen bonding. With the recent advances in QNAR methodology and its encouraging prediction toward virtual nanoparticles, the full potential of QNAR modelling should be exploited in the future to provide critical support to experimental studies over the design of nanomaterials.