RNAS-sgRNA: Recurrent Neural Architecture Search for Detection of On-Target Effects in Single Guide RNA

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• CRISPR/Cas9 is a powerful genomic editing tool.
• Single guide RNA (sgRNA) efficiency varies significantly, limiting CRISPR/Cas9 effectiveness.
• Accurate prediction of sgRNA on-target efficacy is crucial for successful gene editing.

Purpose of the Study:

• To develop an automated model for evaluating CRISPR/Cas9 sgRNA on-target efficacy.
• To improve the accuracy and efficiency of sgRNA sequence categorization.
• To provide a tool that enhances personalized medicine and genetic research.

Main Methods:

• Integration of neural architecture search (NAS) with recurrent neural networks (RNN) to create the RNAS-sgRNA model.
• Automated architectural discovery for optimizing RNN structure.
• Analysis of sgRNA sequences represented as binary matrices to generate classification scores.

Main Results:

• RNAS-sgRNA demonstrated substantial performance enhancements across multiple cell lines (HCT116, HEK293T, HeLa, HL60).
• Achieved significant improvements in area under the receiver operating characteristic curve (AUROC) compared to CRISPRpred(SEQ) and DeepCRISPR models.
• Showcased superior performance on smaller datasets through transfer learning, with an overall AUROC improvement of 13.46%.

Conclusions:

• RNAS-sgRNA offers a novel, automated approach to evaluating CRISPR/Cas9 sgRNA efficacy.
• The model significantly outperforms existing methods, advancing genomic editing research.
• Its ability to leverage transfer learning positions it as a valuable tool for personalized medicine and genetic applications.