Rigorous data curation, enrichment and meta-analysis enable autoML prediction of plant length responses to nanoparticles powered by the Enalos Cloud platform

Abstract

The application of nanomaterials as fertilizers, biostimulants, and pesticides has been emerging as a promising approach in recent years, aiming to support sustainable and precision agriculture, while simultaneously addressing the challenges of climate change, global population growth, and the search for alternative energy sources (biofuels). In this work, to computationally assess the effects of nanoparticles (NPs) on plant growth (encoded in terms of length of e.g., root, shoot or overall plant length), we performed extensive data curation and enrichment with atomistic descriptors of an existing NP–plant interactions database, ensuring high-quality data for the development of machine learning (ML) models. To address class imbalance, data augmentation techniques were applied. An autoML workflow was developed to optimise and evaluate seven ML algorithms for predicting the plant length response class following NP exposure. The optimised XGBoost model demonstrated superior predictive performance during external validation, achieving an accuracy of 85% and a balanced accuracy of 83%, and its applicability domain was clearly defined. One of the key advantages of the plant length response model is that it requires no experimental input data to generate predictions, thus facilitating virtual screening prior to implementation of controlled experimental setups. The curated dataset has been made findable, accessible, interoperable and reusable (FAIR) via the nanoPharos database (https://db.nanopharos.eu/Queries/Datasets.zul?datasetID=np31) and the XGBoost model was documented in a standardized QSAR model report format (QMRF) to enhance its usability and FAIRness and made available as a user-friendly web-application, CeresAI-nano, via the Enalos Cloud platform (https://enaloscloud.novamechanics.com/chiasma/agrinano/).