Predicting CRISPR-Cas9 off-target effects in human primary cells using bidirectional LSTM with BERT embedding.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• The Clustered Regularly Interspaced Short Palindromic Repeats (CRISPR)-Cas9 system is a revolutionary genome editing technology.
• Optimizing single-guide RNA (sgRNA) design is crucial for high on-target efficiency and minimal off-target effects in CRISPR-Cas9 applications.
• Empirical testing of sgRNA designs is resource-intensive, necessitating advanced computational prediction methods.

Purpose of the Study:

• To develop a high-performance deep learning model for predicting off-target effects of sgRNAs in CRISPR-Cas9 genome editing.
• To improve the accuracy and efficiency of sgRNA design for therapeutic and research applications.

Main Methods:

• Developed CrisprBERT, a deep learning model utilizing Bidirectional Encoder Representations from Transformers (BERT) and Bidirectional Long Short-term Memory (LSTM) networks.
• Employed doublet stack encoding to represent local energy configurations of Cas9 binding.
• Utilized paired sgRNA and DNA sequences for predicting off-target effects.

Main Results:

• CrisprBERT demonstrated superior performance compared to existing state-of-the-art deep learning models.
• The model achieved high accuracy in single split, leave-one-sgRNA-out cross-validations, and independent testing.
• The deep learning approach effectively captured contextual embeddings for accurate off-target prediction.

Conclusions:

• CrisprBERT offers a powerful and accurate in silico tool for predicting CRISPR-Cas9 off-target effects.
• The model facilitates optimized sgRNA design, potentially accelerating the development of cell and gene therapies.
• The CrisprBERT model is publicly available for research use.