Item Details

title

Predicting Single Guide RNA Targets For Genome Editing Using Deep Learning

author

Mannion, Joshua

abstract

Clustered Regularly-Interspaced Short Palindromic Repeats (CRISPR) gene editing has become the primary tool to use for genome editing. One challenge of CRISPR-based gene editing is to predict target sites for existing single guide RNA (sgRNA) libraries. In this work, ten deep learning architectures for the prediction of target sites were designed and validated using nine high quality benchmarks. The gated recurrent unit followed by a convolutional neural network (GRUCNN) was identified as the top performing architecture in five-fold cross validation and hold-out validation experiments. Further experimentation showed mild sensitivity of GRUCNN models to the size of the training data. Parameter exploration was performed, and the best parameter values were selected for training of the final GRUCNN model. The final model was used to predict sgRNA target sites of size 23 base-pairs within the human genome. The GRUCNN model predictions were then compared with an existing state-of-the-art tool, DeepCRISPR. The GRUCNN model trained on the large benchmark with 69,848 positive and 69,848 negative target sites from the human genome had excellent binary classification performance with a class separability of 95.4% and sensitivity of 87.8%.

subject

CNN

CRISPR

Deep Learning

GRU

sgRNA

contributor

Khuri, Natalia (committee chair)

Turkett, William (committee member)

Chen, Minghan (committee member)

date

2021-06-03T08:36:05Z (accessioned)

2022-06-02T08:30:11Z (available)

2021 (issued)

degree

Computer Science (discipline)

embargo

2022-06-02 (terms)

identifier

http://hdl.handle.net/10339/98798 (uri)

language

en (iso)

publisher

Wake Forest University

type

Thesis