CRISPRnet: Deep learning and data-driven CRISPR design for network-based multiplexed targeting

The ongoing CRISPR development is revolutionizing the gene editing technology in basic and applied research. A major challenge in doing CRISPR is how to make the desired edits (on-target) while avoiding unintended ones elsewhere (off-target), as these can lead to damaging effects, e.g. in therapeutics. For this computational design is needed, as the first analysis in any CRISPR gene editing experiment. Here, we will employ data- and model-driven approaches to improve on- and off-target predictions. The modeling will be addressed by a combination deep learning and graph neural networks. However, even if on- and off-target predictions are optimal, the current paradigm is limited by testing individual on-target candidates independently and do furthermore completely ignore that some edits might be compensated for by other molecular functions. For example, studying gene function from the effects of silencing a single gene ignores that genes work together in networks with built-in redundancy. We aim to change this paradigm by addressing the redundancy through construction of networks that connects the intended edits to these other molecular functions. The methodologies developed in this project will be widely applicable in biology, biotechnology and biomedical research.