Volume 19, Issue 44, July - December, 2025

CRISPR-Cas13d-mediated transcriptome editing and in silico off-target landscape prediction for RNA therapeutics: A review

David Chukwuma Nwikwe^1♦, Chidozie Augustine Nwanedo², Reza Ghasemi³

¹Chemical Sciences Department (Biochemistry Unit), Faculty of Science, Kings University, P.M.B 555, Km 7, Gbongan-Osogbo Expressway, Odeomu, Osun State, Nigeria
²Department of Sustainable Industrial Pharmaceutical Biotechnology, University of Siena, Italy.
³Department of Biology, Faculty of Converging Sciences and Technologies, Islamic Azad University, Roudehen Branch, Tehran, Iran.

^♦Corresponding author
David Chukwuma Nwikwe, Chemical Sciences Department (Biochemistry Unit), Faculty of Science, Kings University, P.M.B 555, Km 7, Gbongan-Osogbo Road, Odeomu, Osun State, Nigeria.

ABSTRACT

CRISPR-Cas13d is a kind of smaller RNA-targeting nuclease. It shows an increasing number of uses in RNA therapeutics and transcriptome engineering. However, its main advantage over DNA-editing systems is its relatively low activity towards the genome. This study therefore assessed the effectiveness of Cas13d for predicting offtarget effects using modern machine learning tools to target of mammalian transcriptome. The CRISPR-Cas13d guides RNAs (crRNAs) were design for the Cas13d (RfxCas13d/CasRx) system with the intention of targeting the human transcriptome. Thereafter, high-throughput cellular assessments were done to measure the performance of the cells and their viability. The data was used to both construct and verify machine learning models, such as the CNN-based TIGER predictor and DeepCas13, a hybrid CNN–RNN framework. We gauged crucial quantitative indices, which included the AUC, sensitivity, accuracy, and specificity parameters. The result of the prediction model was crosschecked for accuracy with the experimental result, with the use of qRT-PCR technique. It was thereafter compared with nonessential gene controls to reduce the false predictions. The comparative analysis revealed a remarkable improvement in predictive power when transitioning from an empirical rule-based model to a machine learning model. It was observed that the DeepCas13 model outperformed the TIGER model. The former recorded an AUC of ~0.90 and an accuracy of ~85%, while the latter had an AUC of ~0.82 and an accuracy of ~78%. Also, it improved interpretability by combining sequence, structural, and contextual RNA characteristics than the TIGER model. The ROC and radar curve analyses consistently demonstrated that advanced deep learning frameworks reduced off-target errors while maintaining high sensitivity. We used both CRISPR-Cas13d and modern deep learning models to predict the outcomes of direct-to-edit editing in the gene sequence. Hence, modern machine learning tools, especially DeepCas13, have shown promising prospective in the development of RNA therapeutics for research and clinical applications by proffering solutions to the limitations that come with using computational approaches in CRISPR-Cas13d research.

Keywords: CRISPR-Cas13d, RNA editing, transcriptome, off-target prediction, deep learning

Drug Discovery, 2025, 19(44), e25dd3021

PDF

DOI: https://doi.org/10.54905/disssi.v19i44.e25dd3021

Published: 25 October 2025