DeepMEns: an ensemble model for predicting sgRNA on-target activity based on multiple features.

Area of Science:

• Genomics and Bioinformatics
• Molecular Biology
• Computational Biology

Background:

• The CRISPR/Cas9 system, particularly Streptococcus pyogenes Cas9 (SpCas9), shows great promise for gene editing.
• Variability in single-guide RNA (sgRNA) on-target efficiencies limits CRISPR/Cas9's successful application.
• Existing deep learning models for predicting sgRNA activity lack interpretability and have room for performance improvement.

Purpose of the Study:

• To develop an interpretable deep learning model, DeepMEns, for predicting sgRNA on-target activity.
• To overcome the limitations of existing models in terms of interpretability and prediction accuracy.
• To enhance the efficiency and reliability of CRISPR/Cas9 gene editing through improved sgRNA design.

Main Methods:

• Developed DeepMEns, an ensemble interpretable model utilizing deep learning.
• Integrated three distinct input feature types: one-hot encoding of secondary structure, DNA shape features, and positional encoding features.
• Employed convolutional neural networks (CNNs) with Transformer encoders and long short-term memory (LSTM) networks with attention mechanisms.
• Constructed five sub-regressors, averaging their predictions for the final output.

Main Results:

• DeepMEns achieved state-of-the-art performance, demonstrating superior prediction accuracy.
• The model attained the highest Spearman correlation coefficient on 6 out of 10 independent test datasets compared to previous predictors.
• Ablation analysis confirmed that the ensemble strategy significantly enhances prediction model performance.

Conclusions:

• DeepMEns offers a significant advancement in predicting sgRNA on-target activity, improving CRISPR/Cas9 gene editing.
• The model's interpretability allows for better understanding of the factors influencing sgRNA efficiency.
• DeepMEns provides a reliable tool for designing more effective sgRNAs, paving the way for more precise gene editing outcomes.