MS2-Noise-Reduction
Implementing neural networks to reduce noise in MS2 spectra.
Dir: MaSS-Simulator Description: software to generate MS2 spectrum data. Source: https://github.com/pcdslab/MaSS-Simulator
Dir: small_data_1000 Description: Folder containing input, validation, and test MS2 data for noise_reduction_ms2.py. This data was generated with the MaSS-Simulator software. no_noise: Files without the added noise peaks binned: Files output from protein.py that have the generated MS2 data binned. The m/z values are binned to integer values instead of float values.
Dir: small_data_5000 Description: Folder containing the same information as small_data_1000, but the size of the data was started at 5000 peptides. This data was then split randomly into train, validation, and test data sets.
General Outline:
- Start with a fasta protein database
- Call gen_peptide.py to read the fasta file, output proteins in a protein id, protein sequence format. This python script will then read the protein file and generate all the peptide sequences that can be generated for the protein file. Our implementation limits sequences to lengths of 6 - 50. (data written to "protein" directory)
- Call split_sets.py to read the peptide data and split the peptide sequences between train, test, and validation files.
- Call MaSS-Simulator/SimSpec.java to generate the theoretical MS2 spectra
for the 3 peptide files (train, test, and validation).
Compilation ---------------------------------------------------------------------------- There is a compile_java.txt in the src directory of MaSS-Simulator with instructions on compiling the software
cd "MaSS-Simulator/src"
javac -cp "corba-api-5.0.1.jar" SimSpec.java Ion.java MyPair.java
Input
----------------------------------------------------------------------------
Input files are included in the src/data folder
peptides_5000.txt: List of peptide sequences
The included file is ~5000 peptide sequences from yeast.fasta
params.txt: default parameters
mods.ptm: modifications to amino acids (none are default)
Run
----------------------------------------------------------------------------
java SimSpec 1 1 ./data/test.txt test ./data/params.txt ./data/mods.ptm
Output
----------------------------------------------------------------------------
This java file will output two files (no_noise.ms2 and noise.ms2).
noise.ms2 is the input data that contains the b-ion, y-ion, and noisy peaks of the input peptide sequences.
no_noise.ms2 contains only the b-ion and y-ion data. This is used as label data for noise.ms2.
*** The output files must be saved to the expected folder in noise_reduction_ms2.py ***
This folder is defaulted to "MS2-Noise-Reduction/big_data" and named appropriately
Train files:
train_no_noise.ms2
train_noise.ms2
Validation files:
valid_no_noise.ms2
valid_noise.ms2
Test files:
test_noise.ms2
test_no_noise.ms2
- Call noise_reduction_ms2.py A. Read the 6 MS2 files B. Bin the m/z float values to nearest integer and write to binned files C. build a neural network to learn how to reduce the noise in the MS2 data D. test
noise_reduction_keras.py is an implementation of noise_reduction_ms2 with
the Keras API.