Prediction of CRISPR sgRNA Activity Using a Deep Convolutional Neural Network.

Area of Science:

• Molecular Biology
• Genomics
• Bioinformatics

Background:

• The CRISPR-Cas9 system is a revolutionary tool for genome engineering, derived from bacterial adaptive immunity.
• Optimizing single-guide RNA (sgRNA) design is crucial for enhancing the efficiency and accuracy of CRISPR-Cas9 gene editing.
• Current computational tools struggle to accurately predict functional sgRNAs from large-scale sequence data.

Purpose of the Study:

• To develop a deep-learning framework, DeepCas9, for accurate prediction of functional sgRNAs.
• To automatically learn sequence determinants of sgRNA activity using convolutional neural networks (CNNs).
• To improve the efficiency of genome engineering and genetic screens by reducing experimental validation of sgRNAs.

Main Methods:

• Implementation of a deep-learning framework, DeepCas9, utilizing convolutional neural networks (CNNs).
• Training and validation of the CNN model on experimental data to predict sgRNA activity.
• Visualization of convolutional kernels to identify sequence signatures and nucleotide preferences.

Main Results:

• DeepCas9 demonstrated superior performance compared to previous methods in identifying highly active sgRNAs.
• The framework accurately predicted sgRNA target efficacies across different organisms.
• Identified sequence signatures correlated with known nucleotide preferences, validating the model's learning.

Conclusions:

• DeepCas9 offers a powerful and accurate method for predicting functional sgRNAs, significantly advancing CRISPR-Cas9 applications.
• The framework facilitates the design of more effective genome-scale CRISPR interference (CRISPRi) and CRISPR activation (CRISPRa) libraries.
• DeepCas9 is expected to streamline genetic screens and genome engineering workflows by minimizing experimental validation efforts.