Methods for estimating FDR in metabolomics. The methods presented are based on a target-decoy search strategy.
The first two methods are based on generating decoys via a naive and spectrum-based method, as discussed here. The third method is based on generating "knockoffs" by first fitting a spec2ve model on our target spectra, then creating knockoffs from the embedding space via a Gaussian mixture model (as discussed here).
Assessment of the FDR approaches is done by producing QQ-plots of estimated q-values against true q-values.
conda create --name FDR_Metabolomics python=3.7
conda activate FDR_Metabolomics
conda install --channel nlesc --channel bioconda --channel conda-forge matchms
conda install --channel nlesc --channel bioconda --channel conda-forge spec2vec
pip install sklearn
pip install jupyter