Abstract
Nanoinformatics demands accurate predictive models to assess the potential hazards of nanomaterials (NMs). However, limited data availability and the diverse nature of NMs physicochemical properties and their interaction with biological media, hinder the development of robust nano-Quantitative Structure-Activity Relationship (QSAR) models. This article proposes an approach that combines artificially data generation techniques and topological projections to address the challenges of insufficient dataset sizes and their limited representativeness of the chemical space. By leveraging the rich information embedded in the topological features, this methodology enhances the representation of the chemical space, enabling a more an exploration of the structure-activity relationships. We demonstrate the efficacy of our approach through extensive experiments, employing various machine learning regression algorithms to validate the methodology. Finally, we compare two different resampling approaches based on different modeling scenarios. The results showcase a significant improved predictive performance of QSAR models demonstrating a promising strategy to overcome the limitations of small datasets in the field of nanoinformatics. The proposed approach offers noteworthy potential for advancing nanoinformatics research within the nanosafety domain by enabling the development of more accurate predictive models for assessing the potential hazards associated with NMs.
Acknowledgements
Computations were carried out using the computers of Centre of Informatics Tricity Academic Supercomputer & Network.
Disclosure statement
No potential conflict of interest was reported by the author(s).
Data availability statement
Original dataset: https://zenodo.org/record/7335039#.Y3emqnbMIQ8
Dataset with molecular descriptors: supplementary material.
Correction Statement
This article was originally published with errors, which have now been corrected in the online version. Please see Correction (http://dx.doi.org/10.1080/17435390.2023.2285561)