https://github.com/lkytal/pepnet
The state of the art Deep CNN neural network for de novo sequencing of tandem mass spectra
https://github.com/lkytal/pepnet
convolutional-neural-networks deep-learning mass-spectrometry
Last synced: about 1 year ago
JSON representation
The state of the art Deep CNN neural network for de novo sequencing of tandem mass spectra
- Host: GitHub
- URL: https://github.com/lkytal/pepnet
- Owner: lkytal
- Created: 2021-12-23T08:51:22.000Z (over 4 years ago)
- Default Branch: master
- Last Pushed: 2024-10-30T21:22:09.000Z (over 1 year ago)
- Last Synced: 2025-04-06T13:02:54.764Z (over 1 year ago)
- Topics: convolutional-neural-networks, deep-learning, mass-spectrometry
- Language: Python
- Homepage: https://denovo.predfull.com/
- Size: 3.94 MB
- Stars: 35
- Watchers: 4
- Forks: 11
- Open Issues: 7
-
Metadata Files:
- Readme: readme.md
Awesome Lists containing this project
README
# PepNet
## Code for "Accurate __De Novo__ Peptide Sequencing Using Fully Convolutional Neural Networks"
Link to publication: [Accurate __De Novo__ Peptide Sequencing Using Fully Convolutional Neural Networks](https://link.springer.com/article/10.1038/s41467-023-43010-x)
The state of the art Deep CNN neural network for *de novo* sequencing of tandem mass spectra, currently works on unmodified HCD spectra of charges 1+ to 4+.
Free for academic uses. Licensed under LGPL.
__Visit [https://denovo.predfull.com/](https://denovo.predfull.com/) to try online prediction__
## Update History
* 2023.04.27: 2nd Revised version.
* 2022.11.28: Revised version.
* 2021.12.28: First version.
## Method
Based on the structure of the residual convolutional networks. Current precision (bin size): 0.1 Th.
## How to use
__After clone this project, you should download the pre-trained model (`model.h5`) from [zenodo.org](https://zenodo.org/record/7869847) and place it into PepNet's folder.__
### Important Notes
* Will only output unmodification sequences.
* This model assumes a __FIXED__ carbamidomethyl on C
* The length of output peptides are limited to =< 30
### Required Packages
Recommend to install dependency via [Anaconda](https://www.anaconda.com/distribution/)
* Python >= 3.7
* Tensorflow >= 2.5.0
* Pandas >= 0.20
* pyteomics
* numba
Packages Required for traning:
* Tensorflow-addons
### Output format
Sample output looks like:
TITLE | DENOVO | Score | PPM Difference | Positional Score
------- | ------ | ---- | ------- | ------
spectra 1 | LALYCHQLNLCSK | 1.0000 | -3.8919184 | [1.0, 0.9999956, 1.0, 1.0, 1.0, 1.0, 0.99999976, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0]
spectra 2 | HEELMLGDPCLK | 1.0000 | 4.207922 | [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.99999976, 1.0]
spectra 3 | AGLVGPEFHEK | 1.0000 | 0.54602236 | [1.0, 1.0, 1.0, 1.0, 1.0, 0.99999917, 1.0, 1.0, 1.0, 1.0, 1.0]
### Usage
Simply run:
`python denovo.py --input example.mgf --model model.h5 --output example_prediction.tsv`
The output file is in MGF format
* --input: the input mgf file
* --output: the output file path
* --model: the pretrained model
Typical running speed: sequencing 10,000 spectra in ~59 seconds on a NVIDIA A6000 GPU.
## Prediction Examples
We provide sample data on [](https://doi.org/10.5281/zenodo.7869847) for you to evaluate the sequencing performance. The `example.mgf` file contains ground truth spectra (randomly sampled from [NIST Human Synthetic Peptide Spectral Library](https://chemdata.nist.gov/dokuwiki/doku.php?id=peptidew:lib:kustersynselected20170530)), while the `example.tsv` file contains pre-run predictions.
Also, you can run `python evaluation.py --novorst example_prediction.tsv` to generate figures presenting the de novo performance.
## Train this model
See `train.py` for sample training codes
## Note on testing wtih DIA Data
As we demonstrated in the manuscript, we follow the DeepNovo-DIA's method to generate a pseudo-spectrum of each precursor, so we can perform De novo like it's a DDA spectrum. These steps are describe in DeepNovo-DIA's Method section 'Precursor feature detection' and 'In-house database searching'. We actually reused the pseudo-spectrum MGF generated by DeepNovo-DIA.
## Related works
__Also, Visit [https://www.predfull.com/](https://www.predfull.com/) to check our previous project on full spectrum prediction__