Generative AI for Enhancing Reproducibility in Mpox Outbreak Models: Code Reconstruction, Synthetic Data Generation and Automated Documentation

Computational reproducibility remains a significant barrier in artificial intelligence (AI) for infectious diseases due to incomplete documentation, unavailable code, and underspecified workflows, hindering independent verification of findings. This is particularly critical in monkeypox (mpox) modelling, where rapid outbreaks demand transparent and trustworthy analytical tools. This study investigates the ability of generative AI (GenAI to measurably improve reproducibility in mpox detection and prediction models by embedding GenAI-driven interventions within a CRISP-ML(Q) workflow. Five representative mpox studies, including image-based deep learning (ViT-B/16, ResNet-18, DenseNet201) and symptom-based machine learning, were systematically reproduced. GenAI supported three critical dimensions: synthetic data generation for scarcity/overfitting, LLM-assisted code reconstruction for missing pipelines, and automated documentation for standardisation. Results show that GenAI-enhanced workflows substantially improved reproducibility, increasing end-to-end replication success from 20% to 80%. In image-based models with missing code, GenAI-enabled reconstruction reduced performance gaps relative to reported results. Synthetic data improved model stability, and automated documentation achieved high completeness scores. While GenAI aids, foundational scientific practices such as open data and transparent reporting remain paramount. This study concludes that GenAI is an enabling technology that strengthens methodological transparency and auditability but is not a substitute for rigorous scientific conduct.