A QSAR method for predicting biological activities of nanostructures with novel descriptors

With the rapid development of nanomedicine, the prediction and study of the biological activities of nanostructures has become increasingly important. By utilizing machine learning (ML) for modeling, significant savings in materials and time costs can be achieved, improving the efficiency of nanostructure prediction. In this paper, novel descriptors were constructed using the Delaunay triangulation method to establish a ML database. Six nanostructures' biological activities were selected as endpoint effects for investigation. Based on the Quantitative Structure-Activity Relationship (QSAR) method, a relationship model (QSAR-RF) was proposed using the random forest (RF) algorithm to establish a relationship between the novel descriptors and biological activities. Compared with experimental values, the model demonstrated high accuracy, as well as lower Root Mean Squared Error (RMSE) and Mean Absolute Error (MAE) values, proving that QSAR-RF can effectively capture the QSAR between nanostructures and their biological activities. Meanwhile, this paper conducted an importance analysis of the descriptors for six endpoint effects, resulting in an importance ranking and the extraction of the top five influencing factors. Finally, these important factors and their contributions to nanostructure properties were clearly presented through intuitive graphical representations.