Russian Bioinformaticians Have Created A Neural Network Architecture That Can Evaluate How Well An RNA Guide Has Been Chosen For Gene Editing

A novel neural network design to assess how successfully a guide RNA has been chosen for a gene-editing procedure. This methodology will allow for more efficient DNA alteration using the popular CRISPR/Cas system, which will aid in the development of new tactics for making genetically modified creatures and finding ways to cure severe hereditary illnesses. The study, funded by the Russian Science Foundation, was published in Nucleic Acids Research.


Genomic editing, particularly the CRISPR/Cas technique, is widely employed in experimental biology, agriculture, and biotechnology. CRISPR/Cas is one of several weapons used by bacteria to resist viruses. As the pathogen’s DNA enters the cell, Cas proteins detect it as foreign hereditary material and break it because its sequences differ from those of the bacteria. To respond to the virus quicker, the bacterium saves pieces of the pathogen’s DNA—much like a computer antivirus retains a collection of viral signatures—and passes them on to subsequent generations so that its Cas can prevent future attacks.

Teams from different laboratories independently adapted the CRISPR/Cas system to introduce arbitrary changes into DNA sequences in human and animal cells. It made genomic editing much easier and more efficient. The critical components of the mechanism are guide RNA, which “marks the site,” and the Cas9 protein, which cleaves DNA at that location. The cell subsequently “heals the wound,” but the genetic code has already been altered.

The issue is that guide RNA targeting is not always accurate, leading to Cas9 misinterpretation. It is critical to transforming CRISPR/Cas technology into a useful high-precision tool, especially for medical treatments.


Deep learning, Gaussian processes, and other approaches were utilized by Skoltech researchers led to improve the accuracy of identifying suitable guide RNAs. The researchers created a collection of neural networks, trainable mathematical models represented as sequential multiplication of matrices, which are enormous arrays of numbers with complicated underlying structures. A neural network can learn because it contains “memory” in numbers updated in a certain way each time the system performs the computation in training mode. The models were trained on datasets including tens of thousands of experimentally confirmed guide RNAs that have demonstrated great accuracy in human and animal cells.

A method for calculating the likelihood of DNA cleavage for a particular guide RNA was introduced. The obtained scores can guide experimental design in any CRISPR/Cas-based application. They employed neural networks to generate a guide RNA set for precisely changing the genes on the 22nd human chromosome. This was made feasible by the extraordinary accuracy of cleavage frequency prediction and the inclusion of a prediction uncertainty assessment feature that none of the prior approaches provided.

The discoveries may be utilized for several CRISPR/Cas-based technological applications, such as genetic disorder therapy, farming technologies, and fundamental research trials. The team’s time- and resource-saving strategy made it easier to pick the proper guide RNA for high-precision DNA editing, which might aid in the development of novel treatment options for genetic disorders in the long run.