Motivation: Artificial intelligence (AI) applications require explainability (XAI) for FAIR, ethical deployment, whether in the clinic or in the laboratory. Richly descriptive XAI metadata representing how pre-model data were obtained, characterized, transformed, and distributed, should be available along with the data prior to training and application of AI models.
Results: The FAIRSCAPE framework generates, packages, and integrates critical pre-model XAI descriptive metadata, including deep provenance graphs and data dictionaries with feature validation on uploaded data, software, and computations, with special reference to biomedical datasets. It provides ethical and semantic characterization of the dataset along with licensing and availability information, and integrates seamlessly with NIH-recommended generalist repositories. The server is cloud-compliant and implemented in Python3. Client software in Python3 is callable from the command line or directly as python functions. We provide a REST API, and a GUI-based client in javascript is also available.
Availability and implementation: The code is freely available under MIT license and is hosted at https://fairscape.github.io/, along with comprehensive documentation and tutorials.