DEEP LEARNING-ENHANCED CRISPR/CAS9 GENE EDITING FOR CHRONIC MYELOGENOUS LEUKEMIA: TARGETED THERAPEUTICS AND OFF-TARGET PREDICTION
Keywords:
CRISPR/Cas9, Chronic Myelogenous Leukemia, CML, deep learning, LSTM, GRU, RNN, guide RNA, off-target prediction, genomic sequence modeling.Abstract
Chronic Myelogenous Leukemia (CML) is driven by the BCR-ABL1 fusion gene and remains an important model for precision cancer therapy. Although tyrosine kinase inhibitors have substantially improved clinical outcomes, resistance, persistent leukemic stem cells, cumulative toxicity, and long-term treatment requirements continue to motivate investigation of potentially curative strategies. CRISPR/Cas9 offers a direct route for targeting disease-associated genetic alterations, but unintended cleavage remains a major obstacle to safe therapeutic development. This study evaluates sequence-based deep learning methods for predicting CRISPR/Cas9 off-target activity in a CML-oriented setting. The computational framework incorporates one-hot sequence encoding, mismatch profiling, GC-content analysis, normalized genomic features, and ten recurrent, convolutional, feedforward, and attention-based architectures. CRISPR-DIPOFF-LSTM produced the highest reported accuracy (0.985), precision (0.983), recall (0.984), F1-score (0.984), AUROC (0.992), and AUPRC (0.991), with CRISPR-DIPOFF-GRU providing the closest recurrent-model performance. These findings support recurrent sequence modeling as a useful basis for gRNA prioritization, while experimental validation remains necessary before therapeutic application.
