The BioHIT group of SKEL | The AI Lab organises an online talk on Monday 6 June at 13:00 titled Improving CRISPR experimental design through measuring predictive uncertainty.
Talk login details:
Time: Jun 6, 2022, 13:00 EET (Athens)
Meeting ID: 846 9255 2167
Since the emergence of CRISPR gene editing as a field, there was always a need for accurate estimation for cleavage efficiency of gRNA-Cas complex, in both on-target and off-target cases. In on-target cleavage case, a researcher would benefit from maximizing the probability of the cleavage, but in off-target case, the reverse is true – a researcher would benefit from avoiding cleavage. The methods for such estimation employ various forms of machine learning, starting from simple rule-based systems, followed by more complicated classical Machine Learning and Deep Learning solutions. Evolution of on-target and off-target cleavage efficiency prediction methods does not stop at Explainable Deep Learning methods. In this talk, we will discuss the applications of uncertainty quantification for prediction of cleavage efficiency – from characterization of the model error to the implications of an additional uncertainty axis for gene editing experiment design, including the existence of heterogeneity among potential off-target sites.
Bogdan Kirillov is a bioinformatician from Center for Precision Genome Editing and Genetic Technologies for Biomedicine (Institute of Gene Biology, Russian Academy of Sciences). He graduated from ITMO University with B.Sc. in Applied Physics and now studies for a PhD in Life Sciences at Skolkovo Institute of Science and Technology (Skoltech) after graduating from Skoltech with M.Sc. in Biotechnology. Bogdan also worked as a Machine Learning developer in Russian tech industry. His scientific interests include applications of ML in Bioinformatics, CRISPR-Cas systems and unconventional computing.