Electronic Theses and Dissertation

Penyakit alzheimer merupakan gangguan neurodegeneratif progresif yang menjadi penyebab utama demensia pada lansia. salah satu target terapi yang banyak diteliti adalah beta-secretase 1 (bace1), yaitu enzim yang berperan dalam pembentukan peptida amiloid-beta. penelitian ini bertujuan mengembangkan model klasifikasi aktivitas inhibitor bace1 menggunakan deep neural network (dnn) berbasis deskriptor molekuler dua dimensi untuk mendukung proses penyaringan awal kandidat senyawa obat. dataset diperoleh dari basis data bioaktivitas chembl versi 36 (chembl4822) dan melalui tahapan pra-pemrosesan menghasilkan 7.829 senyawa unik. deskriptor molekuler dihitung menggunakan library mordredcommunity, kemudian dilakukan seleksi fitur menggunakan variance threshold dan filter korelasi pearson sebelum proses pelatihan model. kinerja model dnn dibandingkan dengan tiga algoritma ensemble machine learning, yaitu random forest, extra trees, dan gradient boosting. hasil pengujian menunjukkan bahwa model dnn memperoleh accuracy sebesar 0,8263, precision sebesar 0,9215, f1- score sebesar 0,8706, dan roc-auc sebesar 0,9066, serta memiliki specificity tertinggi di antara seluruh model yang dibandingkan. namun, secara keseluruhan gradient boosting memberikan performa terbaik dengan accuracy sebesar 0,8633 dan f1-score sebesar 0,9061. uji statistik mcnemar menunjukkan bahwa perbedaan performa antara dnn dan ketiga model ensemble signifikan secara statistik (p < 0,05), sedangkan perbedaan antar model ensemble tidak signifikan (p > 0,05). analisis feature importance mengidentifikasi deskriptor ic2 sebagai fitur yang paling berkontribusi dalam prediksi aktivitas inhibitor bace1. hasil penelitian menunjukkan bahwa meskipun dnn mampu menghasilkan performa yang kompetitif, model ensemble, khususnya gradient boosting, masih memberikan kinerja yang lebih baik pada dataset tabular berbasis deskriptor molekuler yang digunakan dalam penelitian ini. kata kunci: bace1, alzheimer, deep neural network, qsar, deskriptor molekuler mordred, klasifikasi aktivitas.

Alzheimer's disease is a progressive neurodegenerative disorder and the leading cause of dementia among older adults. One of the most widely investigated therapeutic targets is Beta-Secretase 1 (BACE1), an enzyme involved in the production of amyloid-beta peptides. This study aims to develop a Deep Neural Network (DNN) model for classifying BACE1 inhibitor activity based on two-dimensional molecular descriptors to support the early-stage virtual screening of drug candidates. The dataset was obtained from the ChEMBL bioactivity database version 36 (CHEMBL4822) and, after preprocessing, consisted of 7,829 unique compounds. Molecular descriptors were calculated using the mordredcommunitylibrary, followed by feature selection using Variance Threshold and Pearson correlation filtering prior to model training. The performance of the DNN model was compared with three ensemble machine learning algorithms: Random Forest, Extra Trees, and Gradient Boosting. The DNN achieved an accuracy of 0.8263, a precision of 0.9215, an F1-score of 0.8706, and a ROC-AUC of 0.9066, while also obtaining the highest specificity among all evaluated models. However, Gradient Boosting demonstrated the best overall performance, achieving an accuracy of 0.8633 and an F1-score of 0.9061. McNemar's statistical test indicated significant performance differences between the DNN and the three ensemble models (p < 0.05), whereas no statistically significant differences were found among the ensemble models themselves (p > 0.05). Feature importance analysis identified the IC2 descriptor as the most influential feature for predicting BACE1 inhibitor activity. These findings indicate that although the DNN provides competitive predictive performance, ensemble methods, particularly Gradient Boosting, remain more effective for tabular datasets based on molecular descriptors. Keywords: BACE1, Alzheimer, Deep Neural Network, QSAR, Mordred molecular descriptors, activity classification.