High Activity Target-Site Identification Using Phenotypic Independent CRISPR-Cas9 Core Functionality.

Area of Science:

• Molecular Biology
• Bioinformatics
• Genomics

Background:

• CRISPR-Cas9 target site activity is crucial for gene editing applications.
• Existing computational models for predicting CRISPR-Cas9 activity often perform poorly outside their training conditions.
• There is a need for robust models that can accurately predict CRISPR-Cas9 activity across diverse datasets.

Purpose of the Study:

• To investigate how different data sources influence the predictive power of CRISPR-Cas9 activity models.
• To identify the optimal dataset for building the most robust predictive model.
• To develop a highly accurate and scalable computational model for predicting CRISPR-Cas9 activity.

Main Methods:

• Utilized a machine learning approach to train a predictive model using activity data from 28,606 target sites.
• Compared the performance of the developed model against previously published methods.
• Evaluated the impact of different data sources, particularly sequencing-based measurements, on prediction accuracy.

Main Results:

• The TUSCAN model achieved an 80% average increase in accuracy for predicting the degree of CRISPR-Cas9 activity and a 13% increase for classifying activity into active/inactive categories.
• Models trained on sequencing data demonstrated more accurate predictions of CRISPR-Cas9 activity.
• The TUSCAN model is highly scalable, predicting activity for 5000 target sites in under 7 seconds, suitable for genome-wide screens.

Conclusions:

• Sophisticated machine learning methods can effectively classify binary CRISPR-Cas9 activity.
• Predicting fine-scale activity scores necessitates larger datasets that directly measure indel insertion rates.
• The TUSCAN model offers a significant advancement in predicting CRISPR-Cas9 activity, particularly for large-scale genomic applications.