POLARIS
POLARIS (Prediction Of Localized Affinity for RBPs In Sequence) is a new method for achieving fast, site-specific binding affinity predictions of RNA-binding proteins to genome. It combines a convolutional neural network (CNN) model, trained on RBP eCLIP binding regions, with a gradient-weighted class activation mapping implementation for efficient signal backpropogation to individual positions. With its focus on localized prediction, POLARIS is especially well suited for binding motif interpretability and prediction of functional impact of noncoding variants.
Authors: Alexander Kitaygorodsky, Emily Jin, Yufeng Shen