# SynthCraft: An AI partner for synthetic data generation to support data access and augmentation in healthcare

## TL;DR

SynthCraft is an AI tool that generates synthetic medical data through natural language, helping researchers access data while preserving privacy.

## Contribution

SynthCraft introduces a human-in-the-loop AI system combining reinforcement learning and LLMs for accessible synthetic data generation.

## Key findings

- SynthCraft generates synthetic data that maintains statistical fidelity and downstream utility comparable to real data.
- Different synthetic data generators perform variably across datasets and use cases, with no single method being optimal.
- SynthCraft can be used to explore data augmentation for ethnic representation, though it did not improve model performance in these analyses.

## Abstract

Access to high-quality data provides the foundation for biomedical research. But data access is often limited or challenging due to privacy constraints, whilst the data themselves may be unrepresentative or sparse. Synthetic data can support both privacy-preserving data access and advanced analytical workflows, including data augmentation or the development of digital twins. However, the use of synthetic data remains limited due to the complexity of the methods themselves, their use, and their evaluation. To address this, we developed SynthCraft, an AI tool to support the principled, transparent, application of state-of-the-art synthetic data generation methods. SynthCraft couples a reinforcement learning-based reasoning engine with large language models (LLMs) to orchestrate the workflow necessary for the generation of synthetic data based on dynamic interaction with the user through natural language. We demonstrate the capability of SynthCraft with both tabular and genomic datasets: the National Health and Nutrition Examination Survey (NHANES) and the Cancer Genome Atlas (TCGA). Using SynthCraft, we analysed the privacy, statistical fidelity, and downstream utility of four different synthetic data generators both with and without explicit privacy-preserving designs when applied to both the NHANES and TCGA datasets. We show that how different generators perform differently – and that no single method was optimal – across varying use-cases and datasets. Furthermore, we demonstrate how SynthCraft can be used for data augmentation as part of a workflow to attempt to mitigate imbalances in the proportion of individuals from different ethnic backgrounds. In conclusion, a human-in-the-loop AI partner using LLMs can support the generation of synthetic datasets. Such tools could improve the quality, reproducibility, and transparency of research methods, whilst increasing their accessibility. Research into their use across different methodological areas is warranted.

Medical research depends on access to patient data, but legitimate privacy concerns often mean access is restricted. We created SynthCraft to address this challenge. SynthCraft is an AI partner designed to help researchers generate synthetic versions of medical datasets entirely through natural language, without requiring programming skills. Synthetic data mimic the patterns seen in real datasets, but without containing actual patient data. However, creating and evaluating synthetic data is technically complex, requiring specialised knowledge that limits its accessibility. SynthCraft supports users through each step in the generation of synthetic data: analysing the original data, selecting appropriate generation methods, creating synthetic data itself, before finally rigorously evaluating the results. All actions and code used by SynthCraft are recorded throughout. We demonstrated SynthCraft’s capabilities using a national health survey and cancer genomics dataset. Models trained on our synthetic data performed comparably to those trained on real data. We also explored using synthetic data to address imbalances in ethnic representation, though we did not find that this improved model performance in these analyses. By making advanced methods accessible through natural language and ensuring transparent, reproducible workflows, such tools could transform how researchers apply state-of-the-art methods across biomedical research.