Processing

Smooth

The smooth function is public and applies on MSscanor MSscans objects, with an optional method argument set to MSj.SG(5, 9, 0). Smoothing is performed on the int field using the Savinsky and Golay. The first argument is the order (5 by default), the second is the number of points (default 9) and the last, is the derivative level (0). The function returns an MScontainer type identical to the input.

Base line correction

Base line correction is performed using the baseline_correction function. This function as two methods and operates either on MScontainer or on Array of MSscan such as obtained after importing data.

baseline_correction(scans)
baseline_correction(scans, method = MSj.IPSA(51, 100))

The method argument allows choosing the algorithm.

Top Hat

This filter is based Top Hat transform used in image processing (wikipedia, Sauve et al. (2004). The region onto which the operation is performed is set using the regionfield of the MSj.TopHat. This filter removes every structure from the input which are smaller in size than the structuring element. Usually a region of 100 points is enough.This filter is fast and works quite well on large and complex backgrounds.

Iterative polynomial smoothing algorithm (IPSA)

The default algorithm is the IPSA for iterative polynomial smoothing algorithm (Wang et al. (2017). This iterative algorithm use a zero ordre Savinsly and Golay smoothing to estimate a background. Then a new input, constructed by taking the minimum of either the original spectrum or the background, is smooth again. The process is repeated until the maximum iteration is reached or when the background does not change much. The termination criteria has been changed from the original paper.

Locally weighted error sum of squares regression (LOESS)

The LOESS family of algorithm is based on non-parametric linear local regression where the regression is weighted to reduced the influence of more distant data. We use here the iterative robust estimation procedure where the weights are updated with a bisquare function of the median of the residuals. This algorithm takes the number of iteration to be performed. Usually 3 iteration is enough. This algorithm is slow and is not recommended. The implementation will be improved in future versions.

Peak picking

Pick-picking is performed using the public centroid function. It operates on MSscanor MSscanstype of data and return a similar type. It takes a method argument, set by default to the Signal to Noise Analysis method: MSj.SNRA.

centroid(scan)

Signal to Noise Ratio Analysis (SNRA)

Signal to noise ratio analysis is a very general approach, which relies on the definition of noise. Here, we use TopHat filter to define the noise. Then the signal to noise ratio is calculated. Peaks are found by searching for a local maximum for which the signal to noise ratio is above the threshold. By defaults the MSj.SNRA uses a threshold = 1.0 and a structuring element of 100 points.

centroid(scan, method = MSj.SNRA(1., 100)

Threshold base peak detection algorithm (TBPD)

The TBPD method identifies features based on their similarity (as described by the Pearson correlation coefficient) with a template peak. By default the MSj.TBPD method type uses a Gaussian function, with 1000 mass resolving power and a threshold level set to 0.2% as :

centroid(scan, method = MSj.TBPD(:gauss, 1000, 0.2)

Two other shape functions are available:

• :loretz which uses a Cauchy-Lorentz function and
• :voigt which implements a pseudo-voigt profile (Ida et al., J. Appl. Cryst. (2000), Wikipedia)

The :lorentz profile fits better Fourrier Transform mass spectra. The :voigt shape is the result of the convolution of gaussi and Cauchy-Lorentz shape.

centroid(scan, method = MSj.TBPD(:lorentz, 1000., 0.1)
centroid(scan, method = MSj.TBPD(:voight,  1000., 0.1)