CRISPRpred(SEQ): a sequence-based method for sgRNA on target activity prediction using traditional machine learning.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• CRISPR genome editing tools increasingly rely on deep learning.
• Deep learning models present challenges in explainability and reproducibility.
• There is a need for accurate, interpretable genome editing tools.

Purpose of the Study:

• To develop an accurate genome editing tool using traditional machine learning and sequence-based features.
• To create a method that can compete with existing deep learning models in performance.
• To enhance the explainability and reproducibility of CRISPR tool development.

Main Methods:

• Developed CRISPRpred(SEQ), a novel method for sgRNA on-target activity prediction.
• Utilized traditional machine learning algorithms.
• Employed hand-crafted features extracted directly from sgRNA sequences.

Main Results:

• CRISPRpred(SEQ) achieved superior performance compared to DeepCRISPR, a deep learning model.
• Outperformed DeepCRISPR in three out of four cell lines in benchmark datasets.
• Demonstrated significant performance improvements of 2.174%, 6.905%, and 8.119% in the tested cell lines.

Conclusions:

• CRISPRpred(SEQ) convincingly outperformed DeepCRISPR in 3 out of 4 cell lines.
• Suggests that traditional machine learning with optimized sequence-based features can rival deep learning.
• Highlights potential for further development of traditional machine learning methods for CRISPR applications.