Alignment executables¶
FeatureAlignment executable¶
The Feature Alignment executable can be run as
python feature_alignment.py
and the for help please use
python feature_alignment.py --help
Some of the most used options are the following
fdr_cutoff¶
This is the seeding score cutoff, if a precursor has an identification in one run with at least this score, it will be included for alignment.
max_fdr_quality¶
This is the extension score cutoff. During each step of the algorithm, a peakgroup from a new run is added to the initial seed (see above). Only if the additional peakgroup in the new run has a score better than max_fdr_quality will it be included in the final result.
target_fdr¶
Experimental option for dynamic parameter estimation of the fdr_cutoff parameter. If you want to use this, please turn off fdr_cutoff (but max_fdr_quality still needs to be set).
method¶
Defines the method to use for the clustering. Available options are
- best_overall
- best_cluster_score
- global_best_cluster_score
- global_best_overall
- LocalMST
- LocalMSTAllCluster
Note that the MST options will perform a local, MST guided alignment while the other options will use a reference-guided alignment. The global option will also move peaks which are below the selected FDR threshold (while the best_overall and best_cluster_score will not touch any peak that is below fdr_cutoff).
realign_method¶
Method to use to re-align retention times between pairs of runs. The following options are available:
- None: use the raw RT from the file (not recommended)
- diRT: use only deltaiRT from the input file
- linear: perform a linear regression using best peakgroups
- splineR: perform a spline fit using R (this feature relies on the rpy2 package)
- splineR_external: perform a spline fit using R (start an R process using the command line, not tested under Windows)
- splinePy: use Python native spline from scikits.datasmooth (not recommended, very slow)
- nonCVSpline, CVSpline: splines with and without cross-validation from scipy.interpolate
- lowess: use Robust locally weighted regression (lowess smoother)
- earth : use Multivariate Adaptive Regression Splines using py-earth
- WeightedNearestNeighbour: the weighted RT of the nearest neighbours is used
- SmoothLLDMedian: a local kernel of linear differences is computed
Recommended options are CVSpline and splineR and splineR (if you have R). Both WeightedNearestNeighbour and SmoothLLDMedian gave acceptable results.