DeepCRISTL: deep transfer learning to predict CRISPR/Cas9 on-target editing efficiency in specific cellular contexts.

Area of Science:

• Genomics
• Molecular Biology
• Bioinformatics

Background:

• CRISPR/Cas9 gene editing relies on guide RNAs (gRNAs) for targeted DNA modification.
• Predicting gRNA editing efficiency is crucial for optimizing CRISPR experiments.
• Existing computational methods struggle with accuracy due to limitations in training data.

Purpose of the Study:

• To develop a deep-learning model for accurate prediction of CRISPR/Cas9 editing efficiency in specific cellular contexts.
• To leverage both high-throughput and functional/endogenous datasets for improved model training.

Main Methods:

• Developed DeepCRISTL, a deep-learning model utilizing transfer learning.
• Trained DeepCRISTL on large high-throughput datasets and fine-tuned on specific cellular context data.
• Compared DeepCRISTL with state-of-the-art models like DeepHF and CRISPRon using various transfer-learning strategies.

Main Results:

• DeepCRISTL, particularly with the CRISPRon model and full weight fine-tuning, significantly outperformed existing methods.
• The model demonstrated superior accuracy in predicting editing efficiency on functional and endogenous datasets.
• Saliency maps were used to identify key features influencing predictions across different cellular contexts.

Conclusions:

• DeepCRISTL enhances the prediction of CRISPR/Cas9 editing efficiency in specific cellular environments.
• The model's transfer learning approach effectively utilizes diverse datasets for improved accuracy.
• DeepCRISTL offers a valuable tool for optimizing CRISPR/Cas9 applications in various biological contexts.