Hybrid Modeling Toolbox

The Hybrid Modeling Toolbox by Novasign is a standalone desktop application designed to empower bioprocess development teams by combining advanced machine learning with mechanistic modeling. Installable directly on a PC, it enables scientists and engineers to create, train, evaluate, and deploy hybrid models that accelerate bioprocess development and optimization.

The toolbox is purpose-built for life sciences and bioprocessing workflows, offering an intuitive guided interface that removes the complexity typically associated with building and tuning hybrid AI/ML models. It supports the full modeling lifecycle — from raw data import through model training, selection, analysis, and final export for integration into broader digital infrastructure.

Data Import and Exploration

• Import datasets directly from your PC with a single click, supporting widely-used formats including Excel (with multiple tabs) and CSV.
• Imported datasets are displayed in a responsive data grid that provides a comprehensive overview and highlights any import errors for quick rectification.
• Explore data visually using sophisticated plotting tools that automatically split runs within a dataset, allowing independent time-series visualization without pre-processing.
• Dynamically select variables of interest and plot them against the time axis or against each other to uncover visual correlations.
• Overlay multiple plots within a single figure using the Multiplot feature, and apply transparencies to compare overlaid figures.
• Customize figure axes, titles, legends, and margins, and export generated figures to PDF, SVG, PNG, C#, or text formats.

Model Setup and Customization

• Create hybrid models using a guided stepper that walks users through selecting datasets, defining model outputs and inputs, and configuring neural network architecture.
• Fine-tune models after creation by adjusting configurations, equations, and datasets, with full support for retraining and optimization.
• Choose from multiple model types to suit different bioprocess scenarios:
  • Integrative models — predict outcomes over time by integrating successive predictions, ideal for continuous monitoring and forecasting.
  • Autoregressive models — use previous outputs as inputs for future predictions, enhancing accuracy in time-dependent scenarios.
  • Mass Balance models — automatically adjust for changes in mass or concentration due to additions or removals, ensuring accurate mass conservation.
  • Custom ANN — flexible neural networks where inputs, hidden layers, transfer functions, and individual neuron connections can be fully tailored.
  • Bootstrapping models — trained on various distributions of training, validation, and testing datasets to increase robustness and reduce overfitting.
  • Mechanistic models — combine the predictability of parametric models with the adaptability of non-parametric neural networks for applications requiring both mechanistic understanding and predictive power.

Model Training and Selection

• Initiate model training with a single click; the toolbox handles complex calculations using multi-threaded parallelization to utilize all available computing resources.
• Track training progress in real time through live statistics, learning progress updates, and fitting and prediction plots.
• Use the Model Selection Helper to choose which trained models are carried forward for reporting and analysis, with options for manual selection, automated pre-selection, or a combination of both.

Model Analysis and Reporting

• Perform in-depth analysis of selected models, including plotting individual or aggregated models with average outputs and confidence intervals.
• Customize report plots interactively — filter by specific models, training or test datasets, and individual output variables.
• Export analysis results and plots directly to Excel for external use and documentation.

Model Export and Integration

• Exported models are royalty-free and fully owned by the user, ready for deployment in proprietary projects or third-party applications.
• Integrate models with Novasign's Digital Twin platform for advanced simulations and real-time analytics.
• Access the freely available Novasign Model API, packaged as a Docker image, which supports HTTP REST calls for universal compatibility. OPENAPI documentation enables SDK generation in numerous programming languages including Python, C#, C++, Java, R, Ruby, and Rust.
• Comprehensive integration guides and code examples are provided for Python, MATLAB, and other platforms to ensure seamless connectivity.
• Enhance .NET applications by directly incorporating compiled .NET DLLs, enabling dynamic on-the-fly integration with other .NET-based systems.

The Hybrid Modeling Toolbox is available as a downloadable application and is backed by Novasign's broader bioprocessing digitalization ecosystem, including a strategic partnership with Repligen to further advance bioprocessing digitalization.