TABLE 4.

Summary of the contributions of deep learning studies in CRISPR-Cas9 editing outcomes.

Author	Contributions
Li et al. (2021)	Proposed an end-to-end deep learning framework named CROTON, an approach used deep multi-task convolutional neural networks and neural architecture search (NAS) to automate both feature and model engineering
Liu et al. (2022)	Developed Apindel, a deep learning model employing BiLSTM and Attention mechanism, to predict a comprehensive range of Cas9-generated mutational outcomes, surpassing previous models in accuracy
Wang et al. (2021)	Introduced EditPredict, a CNN-based model that accurately predicts RNA editing events in humans, including those resulting from CRISPR-Cas9 knockout of the ADAR1 enzyme
Li et al. (2022b)	Devised SeqGAN, a model combining CNN and an adversarial network, to predict CRISPR off-target cleavage sites, achieving superior performance in cross-validation tests
Li et al. (2022c)	Proposed machine learning approaches, including deep learning, for identifying human blood cells with CRISPR-mediated fetal chromatin domain (FCD) ablations, with promising results in the prediction of edited cells
Naert et al. (2020)	Applied predictive modeling of editing outcomes to maximize CRISPR-Cas9 phenotype penetrance in Xenopus and zebrafish embryos, improving phenotype penetrance in the F0 generation
Naert et al. (2021)	Introduced CRISPR-SID, a method using deep learning to predict double-strand break repair patterns for identifying genes essential for tumorigenesis in a Xenopus tropicalis desmoid tumor model
Marquart et al. (2021)	Developed BE-DICT, a deep learning model incorporating attention mechanisms, to predict base editing outcomes, providing a versatile tool for genome editing
Zhang et al. (2021b)	Conducted a comprehensive evaluation of the editing efficiency of several Cas9 variants and utilized a deep learning model to verify and predict SpRY off-target sites, informing the refinement of Cas9 variants for precise editing