New deep learning tool for prediction of CRISPR activity
Scientists in a collaboration of Seoul (South Korea) institutions trained a deep learning model on ~13 thousand examples of CRISPR-Cas9 activity. This new tool, given a sequence to target in the human genome, provides a list of proposed guide RNA sequences, ranked by a prediction score.
The model was based on convolutional neural network with five layers: one convolutional layer, one pooling layer, and three fully connected layers. Its hyperparameters were optimized in 324 attempts. Final tests involved multiple independent datasets.
DeepSpCas9 comes after many other tools for CRISPR gRNA design, including other machine learning-based, such as DeepCRISPR. The main improvement lies in large training dataset generated in an experimental approach (lentiviral library used in HEK 293T cells).
Publication: Hui Kwon Kim, Younggwang Kim, Sungtae Lee, Seonwoo Min, Jung Yoon Bae, Jae Woo Choi, Jinman Park, Dongmin Jung, Sungroh Yoon, Hyongbum Henry Kim (2019). SpCas9 activity prediction by DeepSpCas9, a deep learning–based model with high generalization performance. DOI:10.1126/sciadv.aax9249