Deep-Channel

This repository contains code for the paper, [Deep-Channel: A Deep Convolution and Recurrent Neural Network for Detection of Single Molecule Events] (https://www.biorxiv.org/content/10.1101/767418v3 ), by Numan Celik, Fiona O'Brien, Yalin Zheng, Frans Conens, and Richard Barrett-Jolley. Accepted for Publication in Nature's Communications Biology Dec 2019 (https://www.nature.com/articles/s42003-019-0729-3)

This code contains the implementation of a deep learning method to automatically identify transition events of raw time-series ion channel data files. This deep learning method for the analysis of patch-clamp electrophysiological data, relies on convolutional neural networks (CNN) and long short-term memory (LSTM) architecture. This network automatically idealises complex single molecule activity more accurately and faster than traditional threshold crossing or segmental K-means (SKM).

Dependencies

Steps (How to run)

  1. Download the dataset:

    • Download via Github: The raw ion channel data files can be downloaded from this repository in the folder "training dataset"
    • Download via Kaggle: The raw ion channel data files (.csv) with idealized records can be downloaded from kaggle.com/
  2. Activate conda environment:

    • Run conda env create -f environment.yml && conda activate DeepChannel to install dependencies and activate virtual environment.
  3. Run:

    • Run the file deepchannel_train.py to train the Deep-channel model based on LSTM-convolution layers using the training set and evaluates the model on a different test set.

    • Once training procedure has been completed, run the file predictor.py to predict automatically given a different dataset.

    • Check the trained model with other benchmark modalities. To do this run the file predict_deepchannel_QuB and this will compare the Deep-channel model performances with other benchmark (QuB, thresold) modalities.

    • A pre-trained model has been saved in model folder called nmn_oversampled_deepchanel2_5.h5 and this file can be run through predictor.py file to test the idalized record through Deep-channel model without training.

Synthetic data creation

A large collections of synthetic one channel and/or multi-channel raw ion channel data files have been created through a Signal software script called synthetic_data.sgs. These large collections of data files are used for training the Deep-channel model.

A sample of synthetic ion channel data created through Signal script:

One channel opening and closing events on synthetic ion channel data file

Citation / DOI

Please cite this work through this DOI link: https://doi.org/10.1101/767418