Aparajita Dutta

This study compares deep learning models for identifying canonical and non-canonical splice sites across three species, finding that SpliceViNCI performs best, especially with multi-species training.

This paper introduces SpliceViNCI, a model using BLSTM to identify non-canonical splice junctions and visualizes its learned features using integrated gradient and occlusion methods.

This paper evaluates visualization techniques for interpreting a neural network's learned features in splice junction prediction, finding that perturbation-based methods outperform back-propagation for canonical motifs.

SpliceVec uses distributed feature representations to improve splice junction prediction by avoiding manual feature extraction and achieving better classification performance.