Estefanos Kebebew

📑 Thesis Abstract

Antimicrobial resistance (AMR) is a growing global threat, and traditional resistance testing methods are often slow and labor-intensive, delaying clinical decisions. Machine learning (ML) offers promise for faster AMR phenotype prediction from bacterial genomes. However, many models remain under-tested, depend on multiple genomic inputs, and show limited performance on key metrics like the F1-score. This study investigates the use of deep learning—specifically Convolutional Neural Networks (CNNs)—to predict AMR phenotypes in Escherichia coli (E. coli) using single-nucleotide polymorphism (SNP) data, comparing results to Random Forest and XGBoost models. E. coli serves as an ideal model due to its clinical relevance and rich genomic resources. We hypothesize that SNP matrices from reference genomes and bacterial samples are effective ML inputs, and that CNNs will outperform traditional methods. Phenotypic profiles were analyzed for four key antibiotics: ciprofloxacin (CIP), cefotaxime (CTX), ceftazidime (CTZ), and gentamicin (GEN), all of which are increasingly impacted by resistance.

⚙️ Thesis Pipeline

🎓 Education

🛠️ Technical Skills