Pengembangan Quantum Machine Learning untuk Prediksi Efek Off-Target CRISPR/Cas9

Ramadhan, Gilang Kista (2026) Pengembangan Quantum Machine Learning untuk Prediksi Efek Off-Target CRISPR/Cas9. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5025221173-Undergraduate_thesis.pdf - Accepted Version Restricted to Repository staff only Download (7MB) | Request a copy

Abstract

Efek off-target pada sistem penyuntingan gen CRISPR-Cas9, yaitu pemotongan DNA di luar lokasi sasaran, menimbulkan risiko keamanan yang menuntut sistem prediksi akurat. Penelitian ini bertujuan membangun model prediksi off-target berbasis Quantum Machine Learning dengan sirkuit kuantum usulan untuk data sekuensial berdimensi tinggi dan sangat tidak seimbang. Tahap praproses meliputi ekstraksi fitur hibrida DNA-BERT dan C-RNN, pembersihan dan standardisasi, reduksi dimensi PCA, serta subsampling terstratifikasi untuk menangani ketidakseimbangan kelas, lalu dievaluasi dengan validasi silang lima lipat. Sirkuit usulan Soft-Entanglement CPMap memodifikasi arsitektur dasar CPMap dengan mengganti gerbang CNOT menjadi CRX parametrik yang dioptimasi memakai Algoritma Genetika. Percobaan berbasis kernel SVM dengan Projected Quantum Kernel, SE-CPMap menghasilkan MCC 0,72, PR-AUC 0,84, dan AUROC 0,92. Selanjutnya, ekspansi fitur hibrida SE-CPMap pada model QXGB yang menggabungkan proyeksi kuantum dengan XGBoost menghasilkan MCC 0,74, PR-AUC 0,90, dan AUROC 0,95.
========================================================================================================================
The off-target effect in the CRISPR-Cas9 gene editing system, which is the cutting of DNA outside the target location, poses safety risks that demand an accurate prediction system. This research aims to build a Quantum Machine Learning-based off-target prediction model with a proposed quantum circuit for high-dimensional and highly imbalanced sequential data. The preprocessing stage includes hybrid feature extraction of DNA-BERT and C-RNN, cleaning and standardization, PCA dimensionality reduction, and stratified subsampling to handle class imbalance, then evaluated with five-fold cross-validation. The proposed Soft-Entanglement CPMap circuit modifies the basic CPMap architecture by replacing the CNOT gate with a parametric CRX optimized using a Genetic Algorithm. In the SVM kernel-based experiment with Projected Quantum Kernel, SE-CPMap produced an MCC of 0.72, PR-AUC of 0.84, and AUROC of 0.92. Furthermore, the hybrid feature expansion of SE-CPMap on the QXGB model that combines quantum projection with XGBoost produced an MCC of 0.74, PR-AUC of 0.90, and AUROC of 0.95.

Item Type:	Thesis (Other)
Subjects:	Q Science
Divisions:	Faculty of Information Technology > Informatics Engineering > 55201-(S1) Undergraduate Thesis
Depositing User:	Gilang Kista Ramadhan
Date Deposited:	26 Jul 2026 14:45
Last Modified:	26 Jul 2026 14:45
URI:	http://repository.its.ac.id/id/eprint/138109

Actions (login required)

View Item