ProgHAM 🧬

Machine Learning-based Prediction Tool for HAM/TSP Development Risk

📋 About

ProgHAM is a trained Gradient Boosting machine learning model designed to predict the risk of HAM/TSP (HTLV-1-associated myelopathy/tropical spastic paraparesis) development in people living with HTLV-1 (PLwHTLV).

The model was developed using clinical data from approximately 121 PLwHTLV patients followed at the Oswaldo Cruz University Hospital (HUOC) in Pernambuco, Brazil, including clinical data, symptoms, and neurological assessment regarding the presence or absence of HAM/TSP. To address data imbalance, we employed undersampling via Tomek Links, ensuring robust model training and validation.

This prognosis tool is a non-profit project in partnership between the Laboratory of Immunobiologicals and Pathology (LIPa) and the Laboratory of Bioinformatics from the University of Pernambuco (UPE). In order to improve the predictive capacity of the models, the insertion of more samples is essential, so we encourage collaboration and data sharing.

🎯 Objectives

ProgHAM aims to support medical professionals in:

• Risk stratification of HAM/TSP progression
• Enhanced patient follow-up and monitoring
• Data-driven clinical decision support
• Prioritization of patients with higher indicative risk

🧠 Classification Logic

This notebook aims to classify PLwHTLV according to the relative risk of developing HAM/TSP:

• Indicative of HAM
  Possible neurological impairment; patients with priority for receiving care (more severe cases).

• Indicative of PLwHTLV NON-HAM
  No signs of neurological impairment associated with HAM/TSP (less severe cases).

The classification is generated from previously trained machine learning models based on a cohort of around 121 PLwHTLV, followed up at HUOC (Pernambuco, Brazil), using clinical data, reported symptoms, and a neurologist’s evaluation regarding the presence or absence of HAM/TSP.

🚀 Quick Start

Access the tool directly through our Google Colab notebook – no installation required:

Launch ProgHAM

📊 Model Performance

The following visualization demonstrates ProgHAM's predictive performance across multiple evaluation metrics:

⚠️ Important Disclaimer

Clinical Use Notice: ProgHAM requires validation for clinical application and should be used exclusively as a support tool for healthcare professionals. This tool is NOT a diagnostic device and should not replace clinical judgment or standard diagnostic procedures.

🤝 Collaboration and Data Contribution

This prognosis tool is a non-profit initiative created in partnership between:

• Laboratory of Immunobiologicals and Pathology (LIPa), University of Pernambuco (UPE)
• Laboratory of Bioinformatics, UPE

To improve the predictive capacity of the models, increasing the number of samples is essential. If you would like to collaborate, contribute data, or discuss partnerships to improve the prognosis of PLwHTLV, please contact us using the emails below.

👥 Authors

• Matheus Azevedo Bomfim (Bioinformatics Specialist)
  ORCID: 0000-0001-9073-0530
  E-mail: matheusazevedo98@gmail.com

• José Anchieta (Professor at UPE)
  ORCID: 0000-0001-7441-9013

• Paula Magalhães (Professor at UPE)
  ORCID: 0000-0003-1202-6008

• João Pacífico Bezerra Neto (Professor at UPE)
  ORCID: 0000-0003-3861-4879
  E-mail: pacifico.joao@upe.br

• Patrícia Moura (Professor at UPE)
  ORCID: 0000-0003-4259-388X
  E-mail: patricia.moura@upe.br

Registered under the Brazilian Patent and Trademark Office (INPI), Process No. BR512025003608-6.

📧 Contact

Feel free to contact us at: matheusazevedo98@gmail.com

📖 Citation

If you use ProgHAM in your research, please cite (example format):

@software{Bomfim_ProgHAM_2026,
  title  = {ProgHAM: Machine Learning Prediction of HAM/TSP Development Risk in People Living with HTLV-1},
  author = {Bomfim, Matheus Azevedo and Anchieta, José and Magalhães, Paula and Bezerra Neto, João Pacífico and Moura, Patrícia},
  year   = {2026},
  url    = {https://github.com/MatheusABomfim/ProgHAM}
}