This repository contains the code for the analyses presented in our manuscript, "A machine learning strategy that leverages large datasets to boost statistical power in small-scale experiments".
The code provided can fully reproduce the figures and analyses presented in the manuscript, provided that the necessary software are installed and data are present.
The analysis scripts are written for Python 3.7+ and also require a recent version of the Crux mass spectrometry toolkit.
Specifically, we recommend using the Anaconda Python distribution and running the following commands to verify the prerequisite software are installed:
conda install numpy pandas matplotlib seaborn
conda install -c conda-forge tqdmAdditionally, you'll need the Python packages that I use internally to provide common utility functions. These can be installed from GitHub using pip:
pip install git+git://github.com/wfondrie/wispy
pip install git+git://github.com/wfondrie/targetdecoyFinally, you will also need a recent version of Crux tookit. Navigate to the "Download" page, select an appropriate version and click "I agree to the licensing terms, download the most recent build of Crux". You will then need to verify Crux has been added to your path.
The directory structure is expected to look like this:
README.md (this file)
results (analysis scripts and where results are written)
data
|- 2019specht-high
| `- mzML
| `- (mzML.gz files for the SCoPE-MS data)
|- 2019basilicata-denovo
| `- mzML
| `- (mzML.gz files for the gel bands)
|- fasta
| `- human.fasta (the fasta protein database)
`- pin
`- kim.pin.gz (PSMs from the Kim et al. dataset)
The mass spectrometry data files for the Basilicata et al. gel band experiments and the Specht et al. single-cell proteomics experiments are required for the analysis scripts. These must be downloaded an converted to gzipped mzML (*.mzML.gz) format for analysis. Make sure to put these in the appropriate directory after conversion. All of the Thermo raw files can be downloaded with the following (assuming wget is available):
mkdir -p data/2019basilicata-denovo/raw
wget -N -P data/2019basilicata-denovo/raw ftp://ftp.pride.ebi.ac.uk/pride/data/archive/2018/09/PXD009317/*.raw
mkdir -p data/2019specht-high/raw
wget -N -P data/2019specht-high/raw ftp://massive.ucsd.edu/MSV000083945/raw/scope2_raw/*.rawGiven the size of the Kim et al. draft map of the human proteome dataset, we've provided the Percolator input file (pin) for the entire dataset via figshare. Run the following commands to download it into the correct location:
mkdir -p ./data/pin
wget -N -O ./data/pin/kim.pin.gz https://ndownloader.figshare.com/files/19068101Finally, you need the FASTA protein database to search against. For these analyses, you need the Uniprot/Swiss-Prot canonical human proteome. This can be obtained from the following address: https://www.uniprot.org/uniprot/?query=proteome:UP000005640%20reviewed:yes
If all of the data are downloaded in the correct locations, running the scripts in easy:
To run the downsampling experiments:
cd results/repeated-downsampling && python runall.py
cd results/seed-downsampling && python runall.py
To run the analysis of histone gel bands:
cd results/gel-bands && python runall.py
To run the analysis of SCoPE-MS experiments:
cd results/scope-ms && python runall.py
All of the result and intermediate files are written to the results directory in
appropriate subdirectories alongside the runall.py script that generated them.
If you have problems or questions, feel free to ask Will Fondrie (wfondrie@uw.edu).