How to Document Predictive Models Using OpenTox QMRF Editor Documenting predictive models is essential for regulatory acceptance, transparency, and scientific reproducibility. In the field of computational toxicology and quantitative structure-activity relationship (QSAR) modeling, the QSAR Model Reporting Format (QMRF) is the standard template used to summarize model validity.
The OpenTox QMRF Editor is a specialized, open-source tool designed to help researchers, toxicologists, and data scientists document their predictive models efficiently according to European Union Reference Laboratory for Alternatives to Animal Testing (EURL ECVAM) guidelines.
Here is a comprehensive guide on how to use the OpenTox QMRF Editor to document your predictive models. Understanding the QMRF Framework
Before diving into the software, it is crucial to understand what a QMRF document requires. The framework consists of nine structured chapters that ensure your model is robust, validated, and transparent:
User and Model Identification: Details about the authors, developers, and the model name.
Classification and External Databases: Chemical categories and database links.
Prediction Endpoint: The biological or toxicological effect being predicted.
Quantitative Validation – Council of Europe (CoE) / OECD Principles: Alignment with regulatory validation standards.
Defining the Applicability Domain: The chemical space where predictions are reliable.
Internal Validation: Goodness-of-fit and robustness statistics.
External Validation: Performance metrics on an independent test dataset.
Mechanistic Interpretation: The biological or chemical rationale behind the model.
Miscellaneous Information: Additional references and documentation. Step-by-Step Guide to Using the OpenTox QMRF Editor Step 1: Installation and Setup
The OpenTox QMRF Editor is typically available as a standalone desktop application or an executable Java archive (JAR), ensuring it can run across different operating systems like Windows, macOS, and Linux.
Download the latest release from the official OpenTox repository or associated GitHub pages.
Ensure you have the required Java Runtime Environment (JRE) installed on your system. Launch the application to open the main user interface. Step 2: Creating a New Document
Upon launching the editor, you will be greeted with an option to start a new project. Click on File > New. Select the standard QMRF template.
The left-hand panel will populate with a tree-view structure reflecting the nine mandatory regulatory chapters. Step 3: Inputting Metadata and Endpoint Data
Start by filling out the foundational information in Chapters 1, 2, and 3.
Chapter 1: Input your name, organization, contact info, and the date of model creation.
Chapter 3: Clearly define the endpoint. Specify the species, organ system, exposure duration, and endpoint units (e.g., LC50cap L cap C sub 50 Step 4: Outlining Model Training and Algorithms In Chapter 4, you must detail how the model was built.
Specify the software used to generate chemical descriptors (e.g., PaDEL, CDK, Dragon).
Describe the mathematical or machine learning algorithm applied (e.g., Random Forest, Support Vector Machines, Multiple Linear Regression).
Provide the exact mathematical equation or a detailed architectural description if a deep learning approach was used. Step 5: Defining the Applicability Domain (AD)
A model cannot safely predict every chemical structure. In Chapter 5, explicitly define the boundaries of your model.
Describe the method used to determine the AD (e.g., bounding box, Euclidean distance, leverage/Williams plot).
Provide the statistical thresholds or cutoff values that determine whether a query chemical falls inside or outside the model’s prediction space. Step 6: Entering Validation Statistics
Chapters 6 and 7 require the statistical outputs of your model’s performance. The OpenTox QMRF Editor provides structured tables to input these variables. Internal Validation (Chapter 6): Enter metrics like R2cap R squared (coefficient of determination), Cross-Validated Q2cap Q squared
, Root Mean Squared Error (RMSE), or sensitivity/specificity for classification models.
External Validation (Chapter 7): Input the statistical performance against a completely independent test set that was not involved in model training. Regulators heavily prioritize this section. Step 7: Adding Mechanistic Justification
In Chapter 8, describe the causal link between the descriptors used and the toxicological endpoint. For example, if your model predicts skin sensitization, explain how the selected descriptors relate to protein reactivity or molecular electrophilicity. Step 8: Reviewing and Exporting
Once all the sections turn from incomplete indicators to validated states within the application UI:
Run the built-in validation check to ensure no mandatory fields are missing. Click on File > Export.
Save your document in the preferred formats. The editor allows exporting to XML (for electronic submission and registry databases) and PDF/Word (for human-readable reports and publication supplements). Best Practices for QMRF Documentation
Be Specific with Software Versions: Do not just list “Python” or “R.” State the exact package versions used to ensure full reproducibility.
Quantify the Training Set: Clearly state the number of compounds used to train the model, along with their structural diversity.
Keep it Concise but Comprehensive: Write clearly using universal scientific terminology so that regulatory reviewers without a deep data-science background can understand your model’s utility and limitations.
Using the OpenTox QMRF Editor standardizes your workflow, ensuring your predictive computational models gain the trust and compliance verification required in modern toxicology.
If you want, I can help you expand this article. Please let me know:
What target audience is this article for? (e.g., academic students, regulatory scientists, software developers)
Leave a Reply