Improved Machine Learning-Based Model for the Classification of Off-Targets in the CRISPR/Cpf1 System

Area of Science:

• Genomics
• Molecular Biology
• Bioinformatics

Background:

• Targeted nucleases enable precise genome alteration using guide RNAs (gRNAs).
• Off-target activity in CRISPR-Cas systems, caused by mismatch tolerance, hinders clinical applications.
• Minimizing off-target effects while ensuring on-target efficiency is crucial for safe genome editing.

Purpose of the Study:

• To design a computational pipeline for predicting off-target sites in the human genome.
• To develop a multilayer perceptron (MLP) model for predicting cleavage efficiency at off-target sites.
• To identify sequence- and thermodynamics-associated features distinguishing high-activity from low-activity off-targets.

Main Methods:

• A pipeline was created to predict potential off-target sites for a given target sequence.
• A multilayer perceptron (MLP) model was trained using sequence- and binding energy-related features for AsCpf1 and LbCpf1.
• Analysis included nucleotide positional preferences, mismatch distribution, and feature importance for off-target classification.

Main Results:

• Thymine is disfavored near the Protospacer Adjacent Motif (PAM), while guanine is favored in high-activity off-targets.
• Mismatches in the trunk region (positions 16-18 from PAM) and transition mismatches were characteristic of high-activity off-targets.
• MLP identified low-to-moderate melting temperature and base-dependent PAM binding energy as key predictors for high-activity off-targets.

Conclusions:

• The study provides insights into sequence features governing CRISPR-Cas off-target activity.
• The developed MLP model and identified features can help minimize off-target mutations in genome editing.
• This work contributes to enhancing the safety and efficacy of CRISPR-Cas-based therapeutic strategies.