CrnnCrispr: An Interpretable Deep Learning Method for CRISPR/Cas9 sgRNA On-Target Activity Prediction.

Area of Science:

• Molecular Biology
• Bioinformatics
• Genomics

Background:

• CRISPR/Cas9 is a revolutionary genome-editing technology.
• Accurate prediction of single-guide RNA (sgRNA) activity is essential for its effective application.
• Current deep-learning methods for sgRNA activity prediction have limitations.

Purpose of the Study:

• To develop an improved deep-learning model for predicting CRISPR/Cas9 sgRNA on-target activity.
• To enhance the accuracy and generalizability of sgRNA activity prediction.
• To investigate nucleotide-position-dependent patterns influencing sgRNA activity.

Main Methods:

• Proposed CrnnCrispr, a hybrid neural network integrating convolutional neural networks (CNN) and recurrent neural networks (RNN).
• Conducted unbiased experiments comparing CrnnCrispr with four mainstream methods on nine public datasets.
• Implemented a transfer learning strategy to enhance performance on small datasets.

Main Results:

• CrnnCrispr demonstrated superior performance over existing methods in predicting sgRNA on-target activity.
• The model exhibited improved accuracy and generalizability across diverse datasets.
• Visualization revealed generalizable nucleotide-position-dependent patterns crucial for sgRNA function.

Conclusions:

• CrnnCrispr offers a more accurate and generalizable approach to predicting sgRNA activity.
• The model's interpretability through visualization aids in understanding sgRNA design principles.
• This advancement facilitates more effective and precise applications of CRISPR/Cas9 genome editing.