A novel neural network design to assess how successfully a guide RNA has been chosen for a gene editing procedure. This methodology will enable more efficient DNA alteration using the popular CRISPR / Cas system, which will facilitate the development of new tactics to create genetically modified creatures and find ways to cure serious inherited diseases. The study, funded by the Russian Science Foundation, was published in Nucleic Acid Research.
Genomic editing, in particular the CRISPR / Cas technique, is widely used in experimental biology, agriculture and biotechnology. CRISPR / Cas is one of the many weapons bacteria use to resist viruses. When the pathogen’s DNA enters the cell, the Cas proteins detect it as foreign hereditary material and break it down because its sequences differ from those of the bacteria. To react to the virus more quickly, the bacteria records pieces of the pathogen’s DNA, much like a computer antivirus keeps a collection of viral signatures, and passes them on to subsequent generations so that its case can prevent future attacks. .
Teams from different laboratories have independently adapted the CRISPR / Cas system to introduce arbitrary changes in the DNA sequences of human and animal cells. This has made genomic editing much easier and more efficient. Critical components of the mechanism are guide RNA, which “marks the site”, and the Cas9 protein, which cleaves DNA there. Subsequently, the cell “heals the wound”, but the genetic code has already been altered.
The problem is that guide RNA targeting is not always precise, leading to misinterpretation of Cas9. It is essential to transform CRISPR / Cas technology into a useful high precision tool, especially for medical treatments.
Deep learning, Gaussian processes, and other approaches have been used by Skoltech researchers to improve the accuracy of identifying suitable guide RNAs. The researchers created a collection of neural networks, trainable mathematical models represented as a sequential multiplication of matrices, which are huge arrays of numbers with complicated underlying structures. A neural network can learn because it contains “memory” in numbers that update in some way each time the system performs the computation in learning mode. The models were trained on datasets comprising tens of thousands of experimentally confirmed guide RNAs that have demonstrated high accuracy in human and animal cells.
A method of calculating the probability of DNA cleavage for a particular guide RNA has been introduced. The scores obtained can guide the experimental design in any CRISPR / Cas-based application. They used neural networks to generate a set of guide RNAs to precisely modify genes on the 22nd human chromosome. This was made possible by the extraordinary accuracy of the cleavage frequency prediction and the inclusion of a prediction uncertainty evaluation function that none of the previous approaches provided.
The findings can be used for several CRISPR / Cas-based technology applications, such as genetic disorder therapy, agricultural technologies, and basic research assays. The team’s time and resource saving strategy facilitated the selection of the appropriate guide RNA for high precision DNA editing, which could aid in the development of new treatment options for genetic disorders. long-term.