FAIRSCAPE: An Evolving AI-readiness Framework for Biomedical Research

Sadnan Al Manir ¹, Maxwell Adam Levinson ¹, Justin Niestroy ¹

⁰University of Virginia School of Medicine.

Biorxiv : the Preprint Server for Biology +

|

January 7, 2025

View abstract on PubMed

Summary

The FAIRSCAPE framework generates explainable AI (XAI) metadata for FAIR data, ensuring ethical AI deployment in research. This framework supports data provenance and characterization for robust AI model development.

Area Of Science

Biomedical Informatics
Artificial Intelligence
Data Science

Background

Artificial intelligence (AI) applications necessitate explainability (XAI) for ethical and FAIR (Findable, Accessible, Interoperable, Reusable) deployment in clinical and laboratory settings.
Comprehensive XAI metadata detailing data acquisition, characterization, transformation, and distribution is crucial before AI model training and application.

Purpose Of The Study

To introduce the FAIRSCAPE framework for generating, packaging, and integrating essential pre-model XAI descriptive metadata.
To enhance the FAIRness and ethical considerations of biomedical datasets used in AI.

Main Methods

The FAIRSCAPE framework generates deep provenance graphs and data dictionaries with feature validation for uploaded data, software, and computations.
It provides ethical and semantic dataset characterization, including licensing and availability information.
The framework integrates with NIH-recommended generalist repositories and is cloud-compliant, implemented in Python 3 with both server and client software, a REST API, and a JavaScript GUI.

Main Results

FAIRSCAPE successfully generates and integrates critical pre-model XAI metadata, including provenance and data dictionaries.
The framework ensures ethical and semantic characterization of datasets, enhancing their FAIRness.
It offers flexible access via command-line, Python functions, a REST API, and a GUI.

Conclusions

FAIRSCAPE provides a robust solution for generating and managing XAI metadata, crucial for FAIR and ethical AI deployment in biomedical research.
The framework's comprehensive features and flexible implementation facilitate seamless integration with existing data infrastructure and repositories.