AI-Based CRISPR Guide-RNA Efficiency Prediction

A stacked model of gradient-boosted trees and a 1-D convolutional network predicts, from the 30-mer context of a single guide RNA, how efficiently Cas9 will cut - and then designs the guides of a whole gene in the style of CHOPCHOP: every NGG site on both strands, the standard design filters, an exhaustive off-target search against 5,890,572 unique NGG sites of the Ensembl human coding sequence with CFD and MIT specificity scores, and an optimised panel of k guides chosen by greedy, genetic and exhaustive solvers. Validation is strictly gene-disjoint and the model is compared with nine published scores on three datasets it never saw. A matching MATLAB version covers the regression and the optimisation.

Category: Bioinformatics - Genomics and Sequence Analysis (Gradient Boosting, Neural Networks, Combinatorial Optimisation, Explainable AI, MATLAB, Python)

Bioinformatics Project B07 from ProTechSC Academy, with graphical abstract, video demonstration and project guidance.