Pengembangan Hybrid Quantum Neural Network untuk Prediksi Efisiensi On-target CRISPR/Cas9

Andaru, Farel Hanif (2026) Pengembangan Hybrid Quantum Neural Network untuk Prediksi Efisiensi On-target CRISPR/Cas9. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5025221253-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (8MB) | Request a copy

Abstract

Teknologi penyuntingan genom CRISPR/Cas9 membutuhkan prediksi efisiensi on target yang akurat untuk menekan biaya dan kegagalan eksperimen di laboratorium. Pendekatan deep learning telah menjadi metode unggulan untuk tugas ini, sementara paradigma Quantum Machine Learning menawarkan potensis baru melalui pemetaan fitur ke dalam ruang Hilbert berdimensi tinggi. Tugas akhir ini mengembangkan model Hybrid Quantum Neural Network yang mengintegrasikan backbone klasikal CrisprDA sebagai pengekstraksi fitur dengan sebuah sirkuit kuantum variasional sebagai lapisan yang menggantikan lapisan outpuit klasikal. Backbone diadaptasi dari Tensorflow ke Pytorch agar gradien dapat mengalir melalui sirkuit kuantum yang disimulasikan menggunakan library Pennylane. Konfigurasi sirkuit kuantum dioptimalkan menggunakan Grid Search dengan lima axis parameter, dan model dievaluasi pada sembilan dataset varian Cas9 menggunakan korelasi Spearman, korelasi Pearson, MAE, dan RMSE. Untuk menguji secara adil apakah keunggulan yang teramaati benar benar bersumber dari komputasi kuantum, model kuantum dibandingan dengan model konktrol klasik yang disetarakan secara ketat mengikuti rancangan arsitektur kuantum. Hasil menunjukkan konfigurasi kuantum terbaik mencapai rata rata korelasi Spearman sebesar 0,8553 sedangkan model baseline memiliki rata rata korelasi Spearman 0,8518 sementara model kontrol sedikit di bawah model hibrida yaitu sebesar 0,8550.
======================================================================================================================================
CRISPR/Cas9 genome editing technology requires accurate on-target efficiency prediction to reduce costs and experimental failures in the laboratory. Deep learning approaches have become the leading method for this task, while the Quantum Machine Learning paradigm offers new potential through feature mapping into a high-dimensional Hilbert space. This final project develops a Hybrid Quantum Neural Network model that integrates the classical CrisprDA backbone as a feature extractor with a variational quantum circuit as a layer replacing the classical output layer. The backbone was adapted from TensorFlow to PyTorch so that gradients could flow through the quantum circuit simulated using the PennyLane library. The quantum circuit configuration was optimized using Grid Search with five parameter axes, and the model was evaluated across nine Cas9 variant datasets using Spearman correlation, Pearson correlation, MAE, and RMSE. To fairly test whether the observed advantages truly stem from quantum computing, the quantum model was compared against a strictly matched classical control model that follows the quantum architectural design. The result show that the best quantum configuration achieves an avarage Spearman correlation of 0,8553 while the baseline achieve 0,8517, the classical control model achieve slightly below the best quantum configuration at 0,8550.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	CRISPR/Cas9, Hybrid Quantum Neural Network, Prediksi Efisiensi Ontarget, Quantum Machine Learning, Sirkuit Kuantum Variasional, Variational Quantum Circuit
Subjects:	Q Science > QA Mathematics > QA336 Artificial Intelligence Q Science > QA Mathematics > QA76.87 Neural networks (Computer Science)
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Informatics Engineering > 55201-(S1) Undergraduate Thesis
Depositing User:	Farel Hanif Andaru
Date Deposited:	27 Jul 2026 13:48
Last Modified:	27 Jul 2026 13:48
URI:	http://repository.its.ac.id/id/eprint/137518

Actions (login required)

View Item