Publications by authors named "Aman Dalmia"
Comput Biol Med
January 2020
Article Synopsis
- Neural models have shown excellent performance in genome sequence prediction tasks by automatically learning important features from nucleotide sequences, but interpreting these features remains difficult.
- This study evaluates various visualization techniques to extract relevant sequence information learned by a recurrent neural network (RNN) for identifying splice junctions, using genomic data at various nucleotide levels.
- Results demonstrate that different visualization methods provide comparable results for branchpoint detection, with perturbation techniques outperforming back-propagation for canonical motifs, while the opposite is true for non-canonical motifs; the tool for this visualization is available on GitHub.
