69 In silico prediction of in vivo lung inflammation caused by nanomaterials using advanced nanomaterial characterisation

Safety assessment of engineered and incidental nanomaterials (NM) is a costly and time-consuming process, which is currently mainly relying on in vivo (animal) testing. A mechanism-aware prediction of the in vivo hazard profile from readily available material properties and in chimico assays could help avoid animal experiments and save time and resources during product design. We present two machine-learning models for the prediction of the acute lung inflammation quantified via neutrophil influx into mouse lungs at 1 d and 28 d after intratracheal NM instillation: one using advanced experimental descriptors such as the BET surface area and ROS production rate, and the other using purely in silico-derived extrinsic properties of the NMs such as interaction energies with key biomolecules (amino acids and lipids). Using the first model, we show that in vivo respiratory toxicity for a diverse set of 50 NMs, including carbon nanotubes, carbonaceous and metal oxide particles, asbestos, and graphene, can be confidently predicted using only 5 NM descriptors derived from the surface area, shape, and oxidative potency measurements. The analysis of experimental and computational data enables us to categorize NMs into toxicity classes based on their inflammogenic potential and advanced descriptors, thereby relating them to their potential mode of action. The second model, also successfully predicting in vivo outcomes, provides further insights into the toxicity mechanisms by revealing the correlation of inflammogenicity with NM-lipid interactions. The models can be used to group NMs based on in-chimico and/or in-silico characterisation and to predict inflammogenicity for novel materials outside the training set.