Derin öğrenme temelli ilaç yeniden konumlandırma: Kelime temsilleri ve siyam ikizi ağları kullanılarak literatüre dayalı bir çerçeve

The high cost, long timelines, and risks of traditional drug development have sparked interest in drug repositioning—finding new uses for drugs we already have. My thesis introduces a deep learning tool that digs into biomedical data from SemMedDB to spot potential new treatment connections between drugs and diseases. It's built to be a practical, efficient way to come up with ideas for early drug discovery. The tool uses a Siamese Neural Network (SNN), trained on word patterns from the FastText model. The study tested two subnetworks—one dense, one convolutional—to see which worked best for pulling out useful features. After running over 570 setups, the best configuration hit a validation accuracy of 87.66% and a test accuracy of about 83%. It performed well across precision, recall, and F1-scores, too. This work shows how deep learning paired with organized biomedical literature can power smarter drug discovery. It's not perfect yet—relying on one data source, using made-up negative samples, and missing contextual embeddings are some drawbacks. Overall, this system aims to support smarter decisions in drug research and fits into broader efforts for affordable and accessible healthcare.