AI Accurately Predicts On- and Off-Target Activity of CRISPR Tools

A new study published last week in Nature Biotechnology demonstrated that a deep learning (DL) tool can predict on- and off-target activity of RNA-targeting CRISPR tools, which may spur the development of new CRISPR-based therapies.

CRISPR genome-editing technology has garnered significant interest from the precision medicine community in recent years, as it allows genetic engineers to alter an organism’s DNA. The tech is being utilized in multiple research applications, including studies investigating CRISPR-based treatments for cancer, HIV/AIDS, and COVID-19.

Typically, CRISPR tools target DNA using an enzyme known as Cas9, but another type of CRISPR discovered recently can be used to target RNA using the enzyme Cas13.

RNA-targeting CRISPR tools can be used to edit RNA or knock down RNA to block gene expression, and researchers are studying how CRISPR-Cas13 could be applied to cancer diagnosis and virus-host interactions.

However, the use of CRISPR tools can alter not only the targeted DNA or RNA, but also other DNA and RNA molecules. This unintended activity can have detrimental side effects on the cell being edited, making predicting both on- and off-target activity necessary to advance CRISPR research.

However, the study authors noted that most studies on RNA-targeting CRISPRs focused only on evaluating on-target activity, while off-target activity, such as mismatches between the guide and target RNA or insertion and deletion mutations, has been less well-studied.

“Similar to DNA-targeting CRISPRs such as Cas9, we anticipate that RNA-targeting CRISPRs such as Cas13 will have an outsized impact in molecular biology and biomedical applications in the coming years,” said co-senior author of the study Neville Sanjana, PhD, associate professor of biology at New York University (NYU) and associate professor of neuroscience and physiology at NYU Grossman School of Medicine, in a press release discussing the research. “Accurate guide prediction and off-target identification will be of immense value for this newly developing field and therapeutics.”

In the study, the research team performed a series of pooled RNA-targeting CRISPR screens in human cells to measure the activity of 200,000 guide RNAs targeting essential genes.

This activity included both “perfect match” guide RNAs and off-target mismatches, insertions, and deletions.

From there, the researchers developed the Targeted Inhibition of Gene Expression via guide RNA design (TIGER) model, which leveraged the CRISPR screen data to generate predictions about on- and off-target activity.

When TIGER’s predictions were compared with laboratory tests on human cells, the DL tool was found to outperform previous models developed for Cas13 on-target guide design.

The research team indicated that TIGER is the first tool for predicting off-target activity of RNA-targeting CRISPRs and stated that DL and machine learning have an important role to play in genomics research.

“Machine learning and deep learning are showing their strength in genomics because they can take advantage of the huge datasets that can now be generated by modern high-throughput experiments. Importantly, we were also able to use 'interpretable machine learning' to understand why the model predicts that a specific guide will work well,” explained David Knowles, PhD, co-senior author on the study who serves as an assistant professor of computer science and systems biology at Columbia Engineering.

The researchers also stated that TIGER’s predictions can be used to modulate gene dosage, which may be applicable for diseases linked to the presence of too many copies of a gene, like Down syndrome and various cancers.

Moving forward, the research team hopes that TIGER may help avoid undesired off-target CRISPR activity and lead to the development of new RNA-targeting therapies.