Deep Learning for Gene Scissors
Yonsei University Medical School professor Kim Hyung-bum and Seoul National University engineering professor Yoon Sung-ro announced on January 30 that their joint research team developed an AI program choosing the most effective gene scissors for cutting different target parts.
Gene scissors can be defined as artificial enzymes cutting certain DNA parts by coupling with animal and plant genes. Each of the enzymes is divided into a cleavage enzyme for DNA cutting and a guide RNA functioning as a cleavage enzyme carrier.
One of the most important parts for effective gene editing is attachment of a selected enzyme to a target DNA sequence. These days, researchers around the world are studying which guide RNAs can be most effective for different target DNA sequences. Although there have been some computer simulation programs for predicting the effects of gene scissors, the predictions have not been accurate due to the shortage of data.
The team used deep learning as their solution to the limitation. Earlier, professor Kim Hyung-bum came up with the gene editing effects of CRISPR-Cpf1 with 15,000 guide RNAs based on an analysis method for measuring the activities of genetic scissors. Professor Yoon Sung-ro combined the data with deep leaning so that the most effective gene scissors can be presented in different conditions.
The newly developed AI program has a correlation coefficient of 0.87 whereas existing simulation programs’ range from 0.5 to 0.6. Details of the research are available in the Nature Biotechnology journal.