Review of HRMS: focus on the m/z values estimated by the profile-to-centroid calculation by Boulet et al.

This is a very simple paper about how to calculate the accurate centroid of a peak in mass spectrometry, but as they are trying a different approach from the typical ones, I think it is marginally acceptable for publication.

Specific comments:
1. The authors are only looking at orbitrap data, and their improvement in peak centroiding accuracy is mostly occurring because of the way that Thermo reports the data to the users. Thermo uses two modes “Centroid data” using their own secret algorithm and “profile data” which also records a set of data points around the highest point on each peak in order to give some information about the peakshape. The authors state (p3 L18) that “In the profile mode, all signals are recorded.” This is incorrect as Thermo made the decision a long time ago that storing all of the datapoints (most of which are noise) is a waste of storage space and so settled on these two modes. This decision has led to peakshape problems (like the one solved in this paper) as well as peaks missed because they are below a threshold. I encourage the authors to expand this discussion a bit in the introduction to explain the problem with profile mode peaks (which are well shown in figure 1).
2. P3, l43. The authors state that peak widths are caused by the width of the ion beam. That was true in sectors, and is still somewhat true in TOF instruments, but quadrupoles, ion traps, orbitraps, and FTICR peak widths are controlled by base pressure and field accuracy and homogeneity. Suggest rephrasing.
3. When introducing the SG method, it might be worth mentioning that it’s actually a cubic spline that does a piecewise polynomial fit to the data. It does smooth the data and widens the peaks slightly, but it does give polynomials which are easy to calculate a derivative which is very useful in this context.
4. Figure 1a and 1b, and even 1d, the axis numbers are too small to read – even on my big computer screen. Please expand them and make the figure more readable.
5. I calculate that the resolving power of the peak in figure 1 is about 50k at m/z 231. The Orbi is supposed to do 500k at m/z 200. What’s wrong?
6. Normally, the best method to find the peak center is to have the appropriate peakshape model, which for an orbitrap should be Lorentzian if the transient is heavily damped or sinc (sinc^2 in magnitude mode) if not. Usually apodization functions are used to smooth these peakshapes, at the expense of broadening the peak 50-100%. Do you know what apodization function is used here? I recommend reading Yulin Qi’s review article in Mass Spec Reviews on data processing of Fourier transform mass spectrometry data from a few years ago to understand these terms. Alan Marshall’s classic book on Fourier Transforms in NMR, IR, and MS is also well worth a read.
7. I’m fascinated that the barycenter, weighted average approach worked so poorly for the peaks shown, but looking at the peak, I’m not surprised. For Figure 1a, it’s clear that the orbi data system chose to cut off peaks to the right which makes the peakshape quite asymmetric which will badly distort a weighted average. If this is what peaks look like in profile mode from the orby datasystem, then it’s easy to see why the SG smoothed/derivative approach works. I’m intrigued to know if the SG method would work with better data, and if it would work better than the weighted average or peakshape fitting approaches with data which is not so artificially distorted. Can you control the data system to record more data points for each peak in profile mode?