AI improved editor of genome, predicting the results


With the editor of the genome of the CRISPR-Cas9 is pretty well learned to deal. However, with a fairly good knowledge of technology and how it works, the results obtained in the course of editing, not always predictable. This is due to the peculiarities of the editing process and that the outcome is learned to predict artificial intelligence, further improving the performance of the method for genome editing.

To start I would like to explain what is the main problem. Oddly enough, the editor of the genome there is almost nothing to do with it. The protein Cas9 in this case performs the function of the blade and cuts a required component of the genome. But next to “empty space” you need to insert another component. It uses matrix DNA, which, if explained in plain language, acts as a “donor” material. This process is called DNA repair. There are other systems that work on a similar principle in the absence of matrix DNA, but in them lies the problem: after the replacement of the genetic material may remain deletions (roughly speaking, missing plots). The outcome of these interventions is informed and can be computed only empirically.

As the editors of the journal Nature, a team of scientists from the Massachusetts Institute of Technology (MIT) have developed a program based on artificial intelligence that calculates the result of changes in the genome. AI is able to tell what kind of sequence is formed in place of the intervention after editing, as well as at least 50% of the cases, according to whether after the operation to remain deletions. In addition, if you use AI to predict all stages, the prediction accuracy can be increased to 5-11%, which is a very good result.

In order to teach the AI to predict the outcome of the intervention in the genome, the scientists created a library of 2000 the guide RNA for Cas9 (namely RNA specifies protein where to cut the molecule). Next of them was selected as 14, which could after reparations cause the appearance of excess nucleotide, and then AI chose the ones that most often led to errors reparations. Based on these data was built “model behavior”, which has already been used for other operations. Scientists have managed effectively and without any side effects edit of mutation in cells with the syndrome of the German-Pudlak and disease Menekse.

This and other news you can discuss in our chat in Telegram.