CrossLabFit: A Novel Framework for Integrating Qualitative and Quantitative Data Across Multiple Labs for Model Calibration

Abstract

The integration of computational models with experimental data is a cornerstone for gaining insight into biomedical applications. However, parameter fitting procedures often require a vast availability and frequency of data that are challenging to obtain from a single source. Here, we present a novel methodology "CrossLabFit" designed to integrate qualitative data from multiple laboratories, overcoming the constraints of single-lab data collection. Our approach harmonizes disparate qualitative assessments--ranging from different experimental labs to categorical observations--into a unified framework for parameter estimation. By using machine learning algorithms, these qualitative constraints are represented as dynamic "qualitative windows" that capture significant trends to which models must adhere. For numerical implementation, we developed a GPU-accelerated version of differential evolution to navigate in the cost function that integrated quantitative and qualitative data. We validate our approach across a series of case studies, demonstrating significant improvements in model accuracy and parameter identifiability. This work opens a new paradigm for collaborative science, enabling a methodological road to combine and compare findings between studies to improve our understanding of biological systems and beyond.