Expectation pooling

Expectation pooling is a global pooling method for predicting DNA-protein binding using convolutional neural networks (CNNs). The expectation-maximization algorithm inspires the method and offers benefits that can be interpreted statistically and through deep learning theory.

1. Expectation pooling improves the prediction performance of DNA-protein binding compared to conventional methods.

2. The method combines probabilistic ideas with global pooling by taking the expectations of inputs without increasing the number of parameters, making it interpretable.

3. The authors analyze the hyperparameters in the method and propose optional structures to help fit different datasets.

4. The study demonstrates that combining statistical methods with deep learning is highly beneficial for predicting DNA-protein binding.

Topic

Machine learning;Statistics and probability;ChIP-seq

Detail

  • Operation: Standardisation and normalisation;Modelling and simulation

  • Software interface: Command-line user interface

  • Language: Python

  • License: Not stated

  • Cost: Free of charge

  • Version name: -

  • Credit: National Key Research and Development Program of China, National Key Basic Research Project of China, National Natural Science Foundation of China, National Key R&D Program of China, China 863 Program, Beijing Advanced Innovation Center for Genomics, and State Key Laboratory of Protein and Plant Gene Research, Peking University.

  • Input: -

  • Output: -

  • Contact: Xinming Tu xinmingtu@pku.edu.cn ,Minghua Deng dengmh@pku.edu.cn ,Ge Gao gaog@mail.cbi.pku.edu.cn

  • Collection: -

  • Maturity: -

Publications

  • Expectation pooling: an effective and interpretable pooling method for predicting DNA-protein binding.
  • Luo X, et al. Expectation pooling: an effective and interpretable pooling method for predicting DNA-protein binding. Expectation pooling: an effective and interpretable pooling method for predicting DNA-protein binding. 2020; 36:1405-1412. doi: 10.1093/bioinformatics/btz768
  • https://doi.org/10.1093/BIOINFORMATICS/BTZ768
  • PMID: 31598637
  • PMC: PMC7703793

Download and documentation

< Back to DB search