MLPA: A Multi-scale Digital Twin Framework for Personalized Cancer Simulation and Treatment Optimization

Abstract

This research proposes a new personalized cancer modeling and treatment strategy: the Multi-level Parameterized Automata (MLPA), an innovative digital twin framework. The MLPA framework fully integrates macroscopic electronic health records (EHR) and microscopic genomic data for the first time, employing stochastic cellular automata to model tumor progression and treatment efficacy dynamically. The multi-scale strategy effectively enables MLPA to simulate complex cancer behaviors, including metastasis and pharmacological responses, with remarkable precision. The validation using bioluminescent imaging from mice demonstrates MLPAs exceptional predictive power, achieving a state-of-the-art R{superscript 2} value of 0.93, improvement over previously reported conventional models, and fitting within 95% confidence intervals. The proposed MLPA framework accurately captures tumors characteristic S-shaped growth curve and shows high fidelity in simulating various scenarios, from natural progression to aggressive growth and drug treatment responses. MLPAs ability to simulate drug effects through gene pathway perturbation, with predicted tumor cell counts being within a 2.5% equivalence margin interval, underscores its potential as a powerful method for precision oncology. The proposed MLPA framework not only offers a more reliable platform for exploring personalized treatment strategies, but also potentially transforms patient outcomes by optimizing therapy based on individual human biological profiles. As clinically adopted technology progresses towards precision medicine, MLPA stands at the forefront, offering new possibilities in cancer simulation and treatment optimization. The code and imaging dataset used is available at https://github.com/alphamind-club/MLPA.