In silico design of single guide RNA (sgRNA) plays a critical role in clustered regularly interspaced, short palindromic repeats/CRISPR-associated protein 9 (CRISPR/Cas9) system. Continuous efforts are aimed at improving sgRNA design with efficient on-target activity and reduced off-target mutations. In the last 5 years, an increasing number of deep learning-based methods have achieved breakthrough performance in predicting sgRNA on- and off-target activities. Nevertheless, it is worthwhile to systematically evaluate these methods for their predictive abilities. In this review, we conducted a systematic survey on the progress in prediction of on- and off-target editing. We investigated the performances of 10 mainstream deep learning-based on-target predictors using nine public datasets with different sample sizes. We found that in most scenarios, these methods showed superior predictive power on large- and medium-scale datasets than on small-scale datasets. In addition, we performed unbiased experiments to provide in-depth comparison of eight representative approaches for off-target prediction on 12 publicly available datasets with various imbalanced ratios of positive/negative samples. Most methods showed excellent performance on balanced datasets but have much room for improvement on moderate- and severe-imbalanced datasets. This study provides comprehensive perspectives on CRISPR/Cas9 sgRNA on- and off-target activity prediction and improvement for method development.

Off-target effects of CRISPR/Cas9 can lead to suboptimal genome editing outcomes. Numerous deep learning-based approaches have achieved excellent performance for off-target prediction; however, few can predict the off-target activities with both mismatches and indels between single guide RNA (sgRNA) and target DNA sequence pair. In addition, data imbalance is a common pitfall for off-target prediction. Moreover, due to the complexity of genomic contexts, generating an interpretable model also remains challenged. To address these issues, firstly we developed a BERT-based model called CRISPR-BERT for enhancing the prediction of off-target activities with both mismatches and indels. Secondly, we proposed an adaptive batch-wise class balancing strategy to combat the noise exists in imbalanced off-target data. Finally, we applied a visualization approach for investigating the generalizable nucleotide position-dependent patterns of sgRNA-DNA pair for off-target activity. In our comprehensive comparison to existing methods on five mismatches-only datasets and two mismatches-and-indels datasets, CRISPR-BERT achieved the best performance in terms of AUROC and PRAUC. Besides, the visualization analysis demonstrated how implicit knowledge learned by CRISPR-BERT facilitates off-target prediction, which shows potential in model interpretability. Collectively, CRISPR-BERT provides an accurate and interpretable framework for off-target prediction, further contributes to sgRNA optimization in practical use for improved target specificity in CRISPR/Cas9 genome editing. The source code is available at https://github.com/BrokenStringx/CRISPR-BERT.

CRISPR/Cas9 is a powerful genome-editing tool in biology, but its wide applications are challenged by a lack of knowledge governing single-guide RNA (sgRNA) activity. Several deep-learning-based methods have been developed for the prediction of on-target activity. However, there is still room for improvement. Here, we proposed a hybrid neural network named CrnnCrispr, which integrates a convolutional neural network and a recurrent neural network for on-target activity prediction. We performed unbiased experiments with four mainstream methods on nine public datasets with varying sample sizes. Additionally, we incorporated a transfer learning strategy to boost the prediction power on small-scale datasets. Our results showed that CrnnCrispr outperformed existing methods in terms of accuracy and generalizability. Finally, we applied a visualization approach to investigate the generalizable nucleotide-position-dependent patterns of sgRNAs for on-target activity, which shows potential in terms of model interpretability and further helps in understanding the principles of sgRNA design.

CRISPR/Cas9 is a popular genome editing technology, yet its clinical application is hindered by off-target effects. Many deep learning-based methods are available for off-target prediction. However, few can predict off-target activities with insertions or deletions (indels) between single guide RNA and DNA sequence pairs. Additionally, the analysis of off-target data is challenged due to a data imbalance issue. Moreover, the prediction accuracy and interpretability remain to be improved. Here, we introduce a deep learning-based framework, named Crispr-SGRU, to predict off-target activities with mismatches and indels. This model is based on Inception and stacked BiGRU. It adopts a dice loss function to solve the inherent imbalance issue. Experimental results show our model outperforms existing methods for off-target prediction in terms of accuracy and robustness. Finally, we study the interpretability of this model through Deep SHAP and teacher–student-based knowledge distillation, and find it can provide meaningful explanations for sequence patterns regarding off-target activity.

CRISPR/Cas9 has been applied to edit the genome of various organisms, but our understanding of editing outcomes at specific sites after Cas9-mediated DNA cleavage is still limited. Several deep learning-based methods have been proposed for repair outcome prediction; however, there is still room for improvement in terms of performance regarding frameshifts and model interpretability. Here, we present DeepIndel, an end-to-end multi-label regression model for predicting repair outcomes based on the BERT-base module. We demonstrate that our model outperforms existing methods in terms of accuracy and generalizability across various metrics. Furthermore, we utilized Deep SHAP to visualize the importance of nucleotides at various positions for DNA sequence and found that mononucleotides and trinucleotides in DNA sequences surrounding the cut site play a significant role in repair outcome prediction.

Antiviral peptides (AVPs) have potential to enhance our response to various viral diseases. One of the key determinants of their diverse antiviral properties lies in specific functional motifs, making them a valuable addition to therapeutic options. While most studies do not analyze the focus of their models or lack predictions for subclasses. Therefore, it is necessary to have a functional prediction of AVPs with a motif-based interpretable model. However, this remains a challenge because of restricted peptides representation and data imbalance. Here, we introduce BERTAVP, an interpretable deep learning framework that employs multi-task learning to identify AVPs and simultaneously characterize their functional activities (eight species and six families). First, it utilizes the BERT branch to extract peptide features and CNN branch to extract amino acid and physicochemical features. It then applies the focal loss to mitigate the negative impact caused by imbalanced datasets. Finally, we further analyze key motifs learned by BERTAVP, finding ELDKWA and SLWNWF motifs may serve as critical structural and functional modules that define the overall physicochemical character of AVPs. Furthermore, within the tryptophan-rich motif (TRM), WMEWDREI motif exhibits the most potent anti-HIV activity, while the H in KxHxx motif exhibits the most distinctive coronavirus-specific characteristics. Experimental results on public datasets show our model achieves superior performance in identification and functional prediction of AVPs. Source codes are available at https://github.com/755a755/BERTAVP.