Fig. 4: The CCLMoff performance on uncanonical length sgRNA (len = 19, 21).
From: A versatile CRISPR/Cas9 system off-target prediction tool using language model
Despite being trained solely on canonical 20- nt sgRNAs, CCLMoff achieved an AUROC of 0.81 on this unseen dataset, highlighting its strong generalization ability. This result underscores the advantage of the underlying language model in handling variable-length inputs, which is crucial for real-world sgRNA design where non-canonical lengths are frequently employed to optimize CRISPR targeting efficiency.
