Deep learning improves prediction of CRISPR-Cpf1 guide RNA activity

Area of Science:

• Molecular Biology
• Bioinformatics
• Genomics

Background:

• Predicting CRISPR-Cas enzyme activity is crucial for efficient genome editing.
• AsCpf1 (Acidaminococcus sp. Cpf1) is a versatile CRISPR-Cas system for gene editing.
• Current prediction algorithms have limitations in accuracy.

Purpose of the Study:

• To develop and evaluate novel deep-learning algorithms for predicting AsCpf1 guide RNA activity.
• To improve the accuracy of predicting on-target and off-target editing efficiency.
• To assess the impact of chromatin accessibility on AsCpf1 activity prediction.

Main Methods:

• Trained Seq-deepCpf1 using a convolutional neural network framework with indel frequencies from 15,000 target sequences.
• Developed DeepCpf1 by incorporating chromatin accessibility data into the Seq-deepCpf1 model.
• Validated algorithm performance against existing machine learning methods using independent datasets.

Main Results:

• Seq-deepCpf1 accurately predicts AsCpf1 guide RNA activity.
• DeepCpf1 demonstrates enhanced prediction accuracy by integrating chromatin accessibility information.
• Both algorithms outperform previously established machine learning approaches.

Conclusions:

• Deep learning models, particularly DeepCpf1, offer a significant advancement in predicting AsCpf1 guide RNA efficacy.
• Chromatin accessibility is a key factor for improving CRISPR-AsCpf1 activity prediction.
• These algorithms provide valuable tools for optimizing CRISPR-based genome engineering strategies.