Interpretable XGBoost-SHAP model predicts the nanoparticles delivery and reveals its interaction with tumor genomic profiles

Abstract

Understanding the complex interaction between nanoparticles (NPs) and tumors in vivo and how it dominates the delivery efficacy of NPs is critical for the translation of nanomedicine. Herein, we proposed an interpretable XGBoost-SHAP model by integrating the information of NPs physicochemical properties and tumor genomic profile to predict the delivery efficacy. The correlation coefficients were > 0.99 for all training sets, and 0.830, 0.839, and 0.741 for the prediction of maximum delivery efficacy (DEmax), delivery efficacy at 24 h (DE24), and delivery efficacy at 168 (DE168) for test sets. The analysis of the feature importance revealed that the tumor genomic mutations and their interaction with NPs properties played an important role in the delivery of NPs. The functional profile of the NP-delivery-related genes was further explored through gene ontology enrichment analysis. Our work provides a method to accurately predict the delivery efficacy of NPs to heterogeneous tumors and highlights the power of simultaneously using omics data and interpretable machine learning algorithms for discovering the interaction between NPs and tumors, which is important for the development of precision nanomedicine.

Converting Nanotoxicity Data to Information Using Artificial Intelligence and Simulation

“…448 The most common posthoc analysis methods used in nanotoxicology applications are permuted feature importance 134,145,368,420,443,449 and SHAP. 92,410,450,451 The permuted feature importance is a model-agnostic global explanation method that provides insights into ML model behavior. It estimates and ranks feature importance by evaluating how the prediction error increases when a feature is unavailable.…”

“…Various ML models have been developed for the in vivo behavior of NMs in mice or rats. These include modeling of reproductive and pulmonary toxicity, ,, genotoxicity, tissue-specific oxidative stress, metabolic pathways, delivery efficiency, and in vivo fate. ,,, For example, an RF model was developed to identify the most important factors determining reproductive toxicity of NPs, determined by the sperm count, percentage of motile sperm, and sperm abnormalities in rats . Feature importance analysis revealed that the reproductive toxicity was strongly related to the NP type and toxicity end points used.…”

“…Postanalysis methods can take many forms, such as permuted feature importance, partial dependence plots, local interpretable model-agnostic explanations (LIME), and Shapley additive explanations (SHAP) . The most common posthoc analysis methods used in nanotoxicology applications are permuted feature importance ,,,,, and SHAP. ,,, The permuted feature importance is a model-agnostic global explanation method that provides insights into ML model behavior. It estimates and ranks feature importance by evaluating how the prediction error increases when a feature is unavailable.…”