Pycno: An Open-Source Python Framework for Computational Nuclear Oncology and Virtual Theranostic Trials
Abstract
Purpose
Computational nuclear oncology (CNO) methods are advancing rapidly but often developed in isolated tools and workflows that are difficult to reproduce, compare, and share. This fragmentation limits virtual studies and slows translation of physiologically based pharmacokinetic (PBPK) and pharmacodynamic (PD) models into optimized theranostic imaging and radiopharmaceutical therapy (RPT) protocols. To address this gap, we present PyCNO, an open-source framework for developing, implementing, and sharing CNO methods. Built on PBPK/PD modeling, PyCNO provides an integrated environment to simulate imaging and RPTs, perform model fitting, and conduct virtual patient- and population-level analyses. By unifying these capabilities in a single platform, PyCNO enables virtual theranostic trials (VTTs), supports RPT protocol optimization, and provides a foundation for constructing theranostic digital twins (TDTs).
Methods
PyCNO is implemented as an open-source Python library supporting reproducible CNO workflows. Mechanistic PBPK/PD models are encoded in Systems Biology Markup Language (SBML), enabling standardized model specification and efficient simulation of complex biological systems. Additionally, PyCNO translates SBML-defined models into JAX, enabling automatic differentiation and flexible inverse-problem formulations. This dual representation combines rapid simulations (SBML) with gradient-based parameter estimation (JAX) supporting user-defined optimization pipelines. We evaluated the performance and numerical accuracy of PyCNO using both SBML- and JAX-based model implementations and compared these results against equivalent MATLAB SimBiology models.
Results
PyCNO produced results consistent with MATLAB SimBiology for both SBML and JAX implementations. In large VTT workloads, PyCNO achieved computational performance comparable to SimBiology, as runtime was dominated by repeated model solves rather than one-time setup costs. For single-patient simulations, PyCNO was substantially faster due to reduced model initialization overhead.
Conclusion
PyCNO is a flexible, efficient, open-source platform for sharing, simulating, and fitting mechanistic models in CNO. By combining standardized model representations with modern Python-based optimization, PyCNO enables reproducible workflows and supports VTTs, RPT protocol optimization, and digital twinning.