A statistical model for the dynamic regulation of alternative splicing in D. melanogaster

A version of the Ising model is used to study time-varying interactions between RNA-binding proteins and cassette exons in the biological development of D. melanogaster. Interaction parameters are estimated from data via the Tesla procedure, which is closely related to l1-regularized logistic regression. Using an efficient convex optimization program based on forward-backward splitting, the inference procedure can be scaled to thousands of variables. Simulated data is used for benchmarking the procedure, and the estimated parameters are interpreted with clustered heatmaps.