Source code for SpecEncoder: Deep Metric Learning for Accurate Peptide Identification in Proteomics

Free for academic uses. Licensed under LGPL.

Visit https://predfull.com/ to check related works

• 2024.03.07: Second version.
• 2023.10.28: First version.

Based on the structure of the residual convolutional networks.

Different workflows:

Recommend to install dependency via Anaconda

• Python >= 3.7
• Tensorflow >= 2.5.0
• Pandas >= 0.20
• pyteomics
• numba
• Tensorflow-addons

After clone this project, you should download the pre-trained model (encoder.h5) from zenodo.org and place it into SpecEncoder's folder.

Frist we convert query and database into vectors:

python encode.py --query query.mgf --model model.h5 --output query.npy

python encode.py --query database.mgf --model model.h5 --output database.npy

Then we can perform searching:

python search.py --query query.npy --db database.npy --output result.tsv

Typical running speed: convert around 700 spectra in 1 second on a NVIDIA A6000 GPU.

See train.py for sample training codes