CRISPR‐Net: A Recurrent Convolutional Network Quantifies CRISPR Off‐Target Activities with Mismatches and Indels

Abstract

The off‐target effects induced by guide RNAs in the CRISPR/Cas9 gene‐editing system have raised substantial concerns in recent years. Many in silico predictive models have been developed for predicting the off‐target activities; however, few are capable of predicting the off‐target activities with insertions or deletions between guide RNA and target DNA sequence pair. In order to fill this gap, a recurrent convolutional network named CRISPR‐Net is developed for scoring the gRNA‐target pairs with mismatches and indels; and a machine‐learning based model named CRISPR‐Net‐Aggregate is also developed for aggregating the scores as the consensus off‐target score for each potential guide RNA. It is demonstrated that CRISPR‐Net achieves competitive performance on CIRCLE‐Seq and GUIDE‐seq datasets with indels and mismatches, outperforming the state‐of‐the‐art off‐target prediction methods on two independent mismatch‐only datasets. The CRISPR‐Net‐Aggregate also surpasses a competing method on the aggregation task. Moreover, a two‐stage sensitivity analysis is introduced to visualize the CRISPR‐Net prediction on the gRNA‐target pair of interest, demonstrating how implicit knowledge encoded in CRISPR‐Net contributes to the accurate off‐target activity quantification. Finally, the source code is made available at the Code Ocean repository (https://codeocean.com/capsule/9553651/tree/v1).

Prediction of CRISPR/Cas9 single guide RNA cleavage efficiency and specificity by attention-based convolutional neural networks

“…Abadi et al proposed a machine learning-based method named CRISTA for CRISPR cleavage propensity prediction and found that DNA/RNA bulges are predictive features for boosting the performance [16] . Recently, a RNN-based CRISPR-Net has been developed to predict off-target activities with insertions and deletions from sgRNA-DNA pairs [27] . However, the improvement in the predictive power comes at the expense of training time due to the RNN-based model structure.…”

“…In CnnCrispr, bidirectional LSTM (BLSTM) [26] is also followed by CNN to predict off-target activity. CRISPR-Net utilizes RNN to quantify off-target activities with insertions or deletions between sgRNA and target DNA sequence pair [27] . Despite the progress made so far, there is still need for developing more accurate and interpretable methods for sgRNA on- and off-target activities prediction.…”

Systematic decomposition of sequence determinants governing CRISPR/Cas9 specificity

“…Thus, we adopted the Spearman correlation in log-scale for quantitative evaluations. We compared the performance of MOFF-target and MOFF-aggregate to 5 off-target prediction methods, including the benchmark CFD score 9 and its improved version Elevation 24 , two recent deep learning-based methods CNN_std 27 and CRISPR-Net 29 , and an energy-based model CRISPRoff 31 (Figures 4b-4c and Supplementary Tables 12-13). Among them, three machine learning-based methods (Elevation, CNN_std and CRISPR-Net) achieved good predictive power on their training dataset generated by GUIDE-seq, but the performances degraded when tested on datasets from the other two platforms, suggesting overfitting.…”

“…Although experimental techniques are capable of quantifying Cas9 off-target activities, in silico prediction of the off-target effects remains the most efficient and cost-effective method for designing and optimizing CRISPR-based applications. The advancement of machine learning approaches has fueled the progressive improvement of off-target prediction over the past several years [24][25][26][27][28][29] . Moreover, biophysical modeling has provided new insights into the prediction of offtargeting from a bottom-up perspective [30][31][32] .…”

“…machine learning approaches29 . The source codes to implement CRISPR-Net were downloaded from https://codeocean.com/capsule/9553651/tree/v1.…”