Titanium dioxide nanoparticles (TiO2-NPs) have widespread use in various fields. They are widely investigated for antibacterial coatings, cancer treatment solutions, sunscreens, pigments in paint matrices, and other applications. However, numerous studies have reported links between TiO2-NP inhalation and adverse pulmonary outcomes such as emphysema, lung inflammation, fibrosis, and cancer. Hybrid experimental–computational toxicogenomic approaches are increasingly used in the risk assessment of chemicals and nanomaterials. Such an approach allows researchers to decrease the need for animal studies while keeping the high accuracy of the obtained results. In our work, we aimed to quantitatively link inhaled TiO2-NP properties with a complex transcriptomic dataset comprising 621 genes measured in female mice lungs after exposure to five well‑characterized TiO2‑NPs at doses of 18–162 µg/mouse and evaluated at 1 and 28 days post‑exposure periods, resulting in 30 experimental conditions. With 30 conditions, 29 principal components (PCs) captured all transcriptomic variance before supervised modelling. The input predictors set comprised particle surface area, size, charge, dose, and post‑exposure period. The combination of the first two PCs captured 44 % of gene‑level variance, and the ridge regression model predicted this endpoint with Q² = 0.79, tested on six unseen while training conditions. Consequently, the single Machine Learning (ML) model enables approximate reconstruction of >270 genes in the response to TiO2-NP inhalation based on their loadings to the PCs. In practice, this enables rapid ML-based exploration of TiO2‑NP designs and prioritization before animal studies, accelerating safe‑by‑design iteration. By projecting the predicted gene signatures onto established Adverse Outcome Pathways (AOPs), this method can also flag early key events that mechanistically link molecular perturbations to lung outcomes. Thus, the present work extends a previously established computational paradigm of computational nanotoxicology.
This work was funded via the Polish National Science Centre in the frame of the TransNANO project (UMO-2020/37/B/ST5/01894).
