Dilated Convolutions for Modeling Long-Distance Genomic Dependencies
By Ankit Gupta, Alexander M. Rush. 2017
62 citations recorded by the source corpus.
We consider the task of detecting regulatory elements in the human genome directly from raw DNA. Past work has focused on small snippets of DNA, making it difficult to model long-distance dependencies that arise from DNA’s 3-dimensional conformation. In order to study long-distance dependencies, we develop and release a novel dataset for a larger-context modeling task. Using this new data set we model long-distance interactions using dilated convolutional neural networks, and compare them to standard convolutions and recurrent neural networks. We show that dilated convolutions are effective at modeling the locations of regulatory markers in the human genome, such as transcription factor binding sites, histone modifications, and DNAse hypersensitivity sites.
Subfields: Molecular Biology
How researchers see this paper
The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.
In Molecular Biology
- Debora Marks
- Christopher Sander
- Adam J. Riesselman
- Jung-Eun Shin
- A. Kruse
- Conor McMahon
- A. Manglik
- Elana Simon
Citations in the corpus
- Accelerating Protein Design Using Autoregressive Generative Models (2019, methodology): “…we parameterize this process with dilated convolutional neural networks (Figure 1b), which are feed-forward deep neural networks that aggregate long-range dependencies in sequences over an exponentially-large receptive field (van den Oord et al., 2016; Kalchbrenner et al., 2016; Gupta & Rush, 2017).”
- Transformers in Time-Series Analysis: A Tutorial (2022, methodology): “This operation is inspired by dilated convolution as proposed in [83] and [84].”
- Protein design and variant prediction using autoregressive generative models (2021, background): “1), which are feed-forward deep neural networks that aggregate long-range dependencies in sequences over an exponentially large receptive fi eld 65 – 67 (see “ Methods ” ).”
- Informer: Beyond Efficient Transformer for Long Sequence Time-Series Forecasting (2020, methodology): “s the input’s time dimension sharply, seeing the n-heads weights matrix (overlapping red squares) of Attention blocks in Fig.(3). Inspired by the dilated convolution (Yu, Koltun, and Funkhouser 2017; Gupta and Rush 2017), our “distilling” procedure forwards from j-th layer into (j+ 1)-th layer as Xt”