Selected article for: "input sequence and set test"

Author: Jakub M Bartoszewicz; Anja Seidel; Bernhard Y Renard
Title: Interpretable detection of novel human viruses from genome sequencing data
  • Document date: 2020_1_30
  • ID: ac00tai9_30
    Snippet: In order to visualize the learned convolutional filters, we downsample a matching test set to 125,000 reads and pass it through the network. This is modelled after the method presented by Alipanahi et al. (2015) . For each filter and each input sequence, the authors extracted a subsequence leading to the highest activation, and created sequence logos from the obtained sequence sets ("max-activation"). We used DeepLIFT (Shrikumar et al., 2019a) to.....
    Document: In order to visualize the learned convolutional filters, we downsample a matching test set to 125,000 reads and pass it through the network. This is modelled after the method presented by Alipanahi et al. (2015) . For each filter and each input sequence, the authors extracted a subsequence leading to the highest activation, and created sequence logos from the obtained sequence sets ("max-activation"). We used DeepLIFT (Shrikumar et al., 2019a) to extract scoreweighted subsequences with the highest contribution score ("max-contrib") or all subsequences with non-zero contributions ("all-contrib"). Computing the latter was costly and did not yield better quality logos.

    Search related documents:
    Co phrase search for related documents
    • contribution score and high contribution score: 1