Facelift is a stand-alone program for performing multidimensional basepoint correction upon any n-dimensional, real, frequency-domain matrix created by the commercial NMR software program FELIX (Biosym Software). Facelift is intended to remove baseline artifacts one-dimension at a time from any NMR spectrum. The width and intensity of the signals which are removed are determined from the input parameters.
Facelift can be executed from the UNIX prompt by typing "facelift" in the directory containing the program. Alternatively, the user can execute the program form a different directory if the program name is preceded by the path name of the directory containing facelift.
The user must have write permission in the directory containing the input and output FELIX matrixes. Facelift create a new FELIX matrix and the program will display an error message if it is unable to create a new file in the directory containing the input matrix.
Upon execution of Facelift, the following prompts will appear on your terminal. For this documentation, the convention is that the prompts will appear in bold and the parameter values will appear in italics. The following prompts would appear if the input matrix was a 3D matrix.
Enter name of input FELIX matrix
filenm1
Enter name of output matrix to be created
filenm2
Enter dimension for basepoint correction
n
Enter filter width along dim #n
fw
Enter number of standard deviations along dim #n
sd
Enter number of points to smooth along dim #1
smo1
Enter number of points to smooth along dim #2
smo2
Enter number of points to smooth along dim #3
smo3
filenm1: Name of the matrix which is to be basepoint corrected. This matrix is not altered by Facelift. The person executing facelift, however, should have write permission on this matrix. The name of the matrix should include the file extenstion (e.g. ".mat").
filenm2: Name of the matrix which will contain the output basepoint corrected data. This matrix is created by Facelift. Because this matrix is created by Facelift, it must be give a UNIQUE file name. If filenm2 is identical to filenm1 then the original matrix (filenm1) will be destroyed!!!
n: Dimension number along which basepoint correction is to be made. Facelift performs basepoint correction one dimension at a time. If basepoint distortions exist along multiple dimensions, these artifacts can be remove one dimension at a time by separate executions of facelift (See Facelift reference for more details.)
fw: Half-width of the smoothing data window over which data points are sampled (this is parameter "m" in the Facelift reference). The value of this parameter determines the line width of artifacts that wil be suppressed (removed from the spectrum). Lower values for this parameter will result in more signals being removed from the spectrum. Conversely, higher values for this parameter allow only broader signals to be removed from the spectrum.
A reasonable range of values for this parameter is 16 to 128 data points. A recommended range is 32-64 data points (powers of 2 are not necessary). The execution time of the program is linearily proportional to the value of this parameter.
sd: The number of standard deviations of "Guassian" noise above which any peak is considered to be a "signal" as opposed to a "noise" peak. A lower value for this parameter means that weaker signals will be retained by Facelift. Conversely, a higher value of sd will result in the suppression of stronger signals.
A reasonable range of values for this parameter is 2.0-4.0. A recommended range is 2.5 to 3.0.
smo1,smo2,smo3...: The half-width of the smoothing data window which is used to smooth the base point correction matrix along each dimension. Facelift works by constructing a series of line segments for each baseline of each vector along dimension n. The resulting matrix of baseline vectors is then smoothed along dimensions 1,2,3 according to the parameters smo1,smo2, smo3... .
The smo parameter along dimension n (the dimension along which basepoint correction is being performed) should usually be set to the value of fw. The smo parameter along each of the dimensions orthogonal to dimension n should be set to a low value (e.g. 2-4). The purpose of smoothing along each of the orthogonal dimensions is to make the separate baseline vectors continuous along all dimensions of the matrix. The failure to do this during basepoint correction might introduce high frequency noise along each of the orthogonal dimensions.
Important to the performance of the algorithm is that appropriate values for the values fd and sw be chosen. These parameters are chosen according to the linewidth characteristics of the signal and the signal to noise ration of relevant signals in the spectrum. The program well work well, however, for a wide range of these values. For most spectra, a value of sd = 2.5-3.0 is optimal. If the chosen spectrum has a low signal to noise ratio, however, lower values of sd may improve the performance of the algorithm. The hallmark of too high a value of sd is that the resulting spectrum appears to be "overcorrected" i.e. peaks which look like signals are suppressed in the ouput spectrum. Conversely, the hallmark of too low a value of sd is that the resulting spectrum is "undercorrected" i.e. unwanted baseline artifacts are still present in the spectrum. Changing the sd parameter thus affects the intensity of peaks which are suppressed. The fw parameter, however, affects the linewidths of the peaks that are either retained or suppressed. fw should be lowered if the output spectrum contains too many broad peaks which should have been recognized as baseline artifacts. A good starting value of fw is 32 for most spectra.
The best way to obtain optimal performance when dealing with untried NMR spectra is to experiment with a number of different fw and sd combinations. To decrease the time required to do this, small submatrixes should be made from the original matrix in a region of interest. For example, in a 3D spectrum of dimensions 512x256x128, baseline correction along the first dimension could be carried out on a small matrix of dimensions 512x16x16. The 16x16 region along dimensions 2 and 3 can be copied from the original matrix in an "interesting" area containing many baseline artifacts. After obtaining optimal parameters for basepoint correction of these submatrixes, the values of fw and sd can be applied to the basepoint correction of the original large matrix.
If the execution time of Facelift becomes prohibitively long on large matrixes, lowering the value of fw will decrease the execution time of the program. Since a large range of fw values often works on a given spectrum, there is sometimes a tendency to use a value of fw larger than what is needed to produce sufficient basepoint correction.