Integrating Deep Learning and CRISPR

Researchers from New York University, Columbia Engineering, and the New York Genome Centre demonstrated the integration of a deep learning model and Clustered Regularly Interspaced Short Palindromic Repeats (CRISPR) screens to manipulate the expression of human genes.

This innovative approach allows for precise control over gene activity, like adjusting a light switch or a dimmer knob, opening possibilities for developing novel CRISPR-based therapies.

CRISPR, a gene editing technology, has various applications in biomedicine and beyond, including treating diseases like sickle cell anaemia and enhancing agricultural produce. While traditional CRISPR targets DNA using the Cas9 enzyme, recent advancements have unveiled a variant targeting RNA utilising the Cas13 enzyme.

Researchers at NYU and the New York Genome Centre have developed an innovative platform for RNA-targeting CRISPR screens using Cas13. This technology offers versatile applications, including RNA editing, gene expression suppression, and high-throughput screening for drug discovery. Scientists aim to gain a deeper understanding of cellular processes by studying RNA regulation and non-coding RNAs.

Additionally, RNA-targeting CRISPRs can potentially combat viral infections, as RNA serves as the primary genetic material in viruses like SARS-CoV-2 and influenza. Furthermore, investigating the creation of RNA from DNA in the genome sheds light on fundamental gene expression mechanisms in human cells.

The study’s objective is to optimise the specificity of RNA-targeting CRISPRs, ensuring they effectively target the intended RNA while minimising unintended effects on other RNAs within the cell. Previous research primarily focused on target activity and mismatches, overlooking the exploration of off-target activity, specifically insertion and deletion mutations.

Given that approximately one in five mutations in human populations involve insertions or deletions, it is crucial to consider these types of potential off-target effects when designing CRISPR strategies.

“RNA-targeting CRISPRs, including Cas13, are poised to make a profound impact on molecular biology and biomedical applications, similar to DNA-targeting CRISPRs like Cas9,” said Neville Sanjana, co-senior author of the study. “Accurate guide prediction and identification of off-target effects will be invaluable for this emerging field and its therapeutic implications,” he added.

Sanjana and his colleagues conducted a series of pooled RNA-targeting CRISPR screens in human cells. They assessed the effectiveness of 200,000 guide RNAs targeting essential genes, including those with precise matches and those with off-target mismatches, insertions, and deletions.

Collaborating with a machine learning expert, Sanjana’s team developed a deep learning model called Targeted Inhibition of Gene Expression via guide RNA design (TIGER). Using the data from the CRISPR screens, TIGER was trained to make predictions.

Compared to laboratory tests conducted on human cells, TIGER demonstrated superior performance in predicting both on-target and off-target activity. It surpasses previous models designed for Cas13 on-target guide design and establishes the first tool for predicting the off-target effects of RNA-targeting CRISPRs.

The researchers successfully designed Cas13 guides using TIGER, balancing on-target knockdown and minimising off-target effects. This advance follows their earlier research on designing Cas13 guides for specific RNA knockdown. The team also demonstrated TIGER’s ability to modulate gene dosage precisely, offering potential applications for gene copy number variations, certain diseases, and cancer.

“Our deep learning model provides guidance not only for complete knockdown of a transcript but also for fine-tuning it, such as achieving 70% expression of a specific gene,” explained Andrew Stirn, co-first author of the study.

The researchers believe that integrating artificial intelligence with RNA-targeting CRISPR screens will reduce off-target CRISPR activity and advance the development of RNA-targeting therapies.

Sanjana believes with larger CRISPR screen datasets, the application of machine learning models is expanding rapidly. He added that TIGER’s ability to predict off-target effects and fine-tune gene dosage opens promising prospects for RNA-targeting CRISPRs in biomedicine.