Deep learning predicts CRISPR off-target effects

Off-target effects remain a significant challenge in CRISPR-CAS9 therapeutics, as the system can tolerate sequence mismatches and structural variations at unintended genomic sites. While CRISPR-CAS9 offers remarkable potential for treating monogenic diseases through single-intervention therapies, existing computational methods for predicting these unwanted effects are typically restricted to specifically designed single-guide RNAs and show limited performance when applied to novel sequences.

The researchers developed CCLMoff by incorporating a pre-trained RNA language model from RNAcentral, enabling the framework to capture complex sequence relationships between guide RNAs and their potential target sites. Training on a comprehensive, updated dataset allowed CCLMoff to demonstrate superior generalization capabilities across multiple next-generation sequencing-based detection datasets.

The model's interpretability features confirmed the biological significance of the seed region in guide RNA targeting, validating an established understanding of CRISPR-CAS9 mechanics whilst providing enhanced analytical capabilities.

This advancement represents progress towards developing comprehensive, end-to-end guide RNA design platforms that could improve both precision and efficiency in CRISPR-CAS9 therapeutic applications.

The study was led by Dongsheng Tang at Foshan University, China, and it was published on 6 July in Communications Biology.