An agent-based learning model integrating sex differences in renal cell carcinoma.

BACKGROUND

Sex-based differences influence tumor biology, immune responses, and treatment outcomes in renal cell carcinoma (RCC), yet most computational models do not jointly incorporate sex hormones, immune composition, and tumor genetic evolution. Agent-based models (ABMs) effectively simulate tumor-immune interactions but are rarely extended to include sex-specific modulation or machine learning-based optimization. This study enhanced an agent-based learning model (ALM) to simulate RCC progression and treatment response by integrating hormonal effects, immune interactions, and tumor genetic adaptation with data-driven tuning.

METHODS

An RCC-specific ALM was developed incorporating immune agents (CD8+, NK, Treg, dendritic cells), hormone-sensitive mechanisms, tumor genetic modules, and effects of immune checkpoint inhibitors (ICIs) and tyrosine kinase inhibitors (TKIs). Tumor evolution was modeled using a genetic algorithm simulating promoter and gene mutations, with fitness defined by immune evasion and proliferation advantages. Model parameters were optimized using clinical outcomes from the ARON dataset via the Optuna framework, and performance was assessed using concordance index (CI) and mean squared error (MSE).

RESULTS

Simulations reproduced sex-specific treatment responses. Female models showed delayed initial responses but stronger late immune activation and rapid tumor regression, whereas male models exhibited more stable early responses but greater tumor resilience driven by genetic adaptations. Adaptive learning showed capability of reducing prediction error with both fitness functions.

CONCLUSIONS

This ALM offers an exploratory framework to provide preliminary insights into how sex hormones, immune dynamics, and tumor genetics may jointly contribute to shaping RCC treatment outcomes. Although the limited sample size constrains validation, the results suggest the potential of combining ABMs with biological data-driven optimization to support patient prediction and call for further investigation in larger cohorts.