ADVANCED STEM CELL DONOR MATCHING USING A HYBRID RANDOM FOREST AND VARIATIONAL AUTOENCODER FRAMEWORK

Abstract

Stem cell transplantation is a vital treatment for various hematologic diseases, where finding a compatible donor is crucial for the success of the procedure. Traditionally, the donor-recipient matching process involves complex evaluation of genetic markers and Human Leukocyte Antigen (HLA) typings, which requires advanced analytical techniques to ensure compatibility. The proposed research presents an advanced stem cell donor matching methodology using a Hybrid Random Forest and Variational Autoencoder (VAE) framework. The model leverages the VAE for complex feature extraction, compressing high-dimensional donor-recipient characteristics into an informative latent space, and integrates this with a Random Forest classifier for predicting compatibility. The enriched feature set, derived by combining latent features and original data, enables the model to capture nuanced relationships between genetic markers, HLA typings, and other biological factors. The model was implemented in Jupyter Notebook and achieved a remarkable accuracy of 80.17%, outperforming nine existing models, including Standard Random Forest, XGBoost, and LightGBM, by an average margin of 4%. Additionally, the model demonstrated high precision, recall, F1-score, and AUC-ROC values, indicating its robustness in correctly identifying compatible donor-recipient pairs. The effectiveness of this approach suggests its potential to enhance decision-making in clinical settings, providing a reliable and efficient solution for stem cell donor matching.