Code for work in "Cerebrospinal fluid reference proteins increase accuracy and interpretability of biomarkers for brain diseases", created by Linda Karlsson 2023.
This repository aims to provide details of how the results in "Adjusting for CSF Reference Proteins Improves Biomarker Accuracy" were created. Due to data sharing policies, most scripts and functions have been reduced to algorithmic descriptiveness and should not be run without firstly specifying input data and data variable names. A simulated csv file exists to try out the code and functions.
Code was written in Python version 3.9 and R version 4.2. Required packages:
Python:
- sklearn 0.0
- pandas 1.4.4
- numpy 1.23.3
- matplotlib 3.5.3
- pingouin 0.5.3
- tqdm 4.64.1
- statsmodels 0.13.2
R:
- tidyverse 1.3.2
- pROC 1.18.0
- EWCE 1.4.0
Installation is typical fast and can be done within a few minutes.
All created python functions were included in Functions.py, and examples of how they were used are given in the notebooks. In TSNE-clustering.ipynb, dimensionality reduction, K-means clustering and cluster evaluation is included. In Evaluation.ipynb, logistic regression evaluations and significance test on bootstrapped models can be found. Additionally, computation of linear regression and partial correlations are provided. In Robustness_Analysis, a methodology of evaluating reference's robustness to added noise is given (which was not included in the manuscript).
The R-script Evaluation_testdata.R evaluates performance and significance on unseen testdata, comparing ROC curves. Bootstrap_enrichment_test.R compares cell type expression of a cluster of proteins against a background set of proteins.
The running time is very dependent on the size of the dataframe - with the simulated data all runs should maximum take a few seconds. With larger dataframes, expect times increasing to one or several minutes.