Clinical-oriented tacrolimus dosing algorithms in kidney transplant based on genetic algorithm and deep forest

Background The immunosuppressant tacrolimus (TAC) plays a crucial role in preventing rejection reactions after organ transplant. Due to a narrow therapeutic window, it is one of the long-term challenges in postoperative care, increasingly requiring a precise management due to individual variability. To alleviate the burden on clinicians and achieve an automatic and precise drug dosing, the AI-assisted personalized dosing of TAC is a promising predictive method. Methods This study presents a clinical-oriented TAC dosing algorithm that integrates genetic algorithm (GA) with deep forest (DF) to predict both initial and follow-up doses for kidney transplant recipients. The optimized candidate variables were first conducted from numerous clinical factors by GA using support vector regression based on radial basis function. Then a smaller number of key clinical variables were confirmed for clinical relevance and ease of use by an exhaustive feature selection method. Results Validated in a cohort of 288 recipients, the DF model combined with a few clinical variables ultimately achieved an average accuracy of 84.5% and 91.7% in the initial and follow-up dosage prediction. Conclusion The proposed approach can provide a potential reference to algorithm-based automatic pipeline methods for drug dosing prediction and analysis in clinical practice.