CRISPR-Embedding: CRISPR/Cas9 off-target activity prediction using DNA <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" altimg="si30.svg" display="inline" id="d1e659"> <mml:mi>k</mml:mi> </mml:math> -mer embedding

CRISPR/Cas9 has revolutionized gene editing, offering biologists and researchers a powerful tool for targeted genome modification. Despite its promise, the technique carries the risk of unintended off-target edits, which can disrupt normal cellular functions. To enhance the accuracy of off-target predictions, various computational approaches have been explored; however, traditional methods often face challenges such as data imbalance and high model complexity. In this paper, we present CRISPR-Embedding, a deep learning model based on a 9-layer Convolutional Neural Network (CNN) designed to predict CRISPR/Cas9 off-target activity. Our model utilizes DNA k -mer embeddings for effective sequence representation. To address the issue of data imbalance, we applied data augmentation and under-sampling strategies, resulting in a cleaner, more balanced dataset. Through 5-fold cross-validation, CRISPR-Embedding achieved a notable average accuracy of 94.07%. Furthermore, it demonstrated superior performances over existing state-of-the-art methods in off-target activity prediction. The code and resources for CRISPR-Embedding are available at: https://github.com/swakkhar/crispr-embedding .