Repository containing the machine learning analyses based on the nucleosome positioning of cell-free DNA samples around binding sites of regulatory elements. These analyses were performed as part of a master thesis.
Example feature data sets from own analyses is given in this folder. These features were extracted from scores calculated in the https://github.com/leaburkard/cfDNA_midpoint_coverage repository. As samples the data of Christiano et al. (Nature, 2019) was used and as regions the transcripion factor binding sites of the GTRD (version 19.10).
The features are compared to the feature set of the Griffin analyses published by Doebley et al. (Nature Communications, 2022). Their data is provided at https://github.com/adoebley/Griffin_analyses/tree/main/delfi_data_cancer_detection/number_of_sites_analysis/merged_data and can be downloaded from there.
Execute the cfDNA_midpont_coverage repository with own cfDNA samples and regions of interest. Then use the feature_extraction.py script.
-r: run ID given in the config/samples.tsv file in the cfDNA_midpont_coverage repository
-s: which of the three scores (MIDPOINT, WPS or COV) was calculated
-a: which method to calculate the amplitude (FFT or spec_pgram)
-p: path to the cfDNA_midpont_coverage repository
Example:
scripts/feature_extraction -r <ID> -s MIDPOINT -a FFT -p ../cfDNA_midpont_coverage
More information is given using the -h help option.
The notebooks contain different machine learning analyses conducted with the described features.