The message passing neural network (MPNN) model demonstrates strong predictive accuracy, how does its performance compare to traditional QSAR-based models in terms of generalizability and interpretability? A direct comparison could help contextualize its advantages and limitations. Additionally, the study identifies structural alerts associated with NRF2 activation, but were any efforts made to assess their toxicity profiles? Since NRF2 activation is linked to both protective and potentially oncogenic effects, evaluating off-target interactions could enhance the real-world applicability of these findings.
