tutorial

DIFRATE detector optimization tutorial

Note that included with the DIFRATE software is a file ‘tutorial.m’ that helps set up detector analysis of the Ubiquitin data as shown in the JCP paper. This can be used in conjuction with the tutorial guidelines given here.

Now that one sees the functions of the detector_opt program, we give a short tutorial for usage. We start by generating a set of experiments and running the ‘detector_opt’ program using the following commands:

tc=logspace(-14,-3,200); %Generate correlation time vector
info.Nuc=’15N’; %Nucleus measured in all experiments
info.R1.v0=[400 600 850]; %B0 Field of R1 measurements (1H, MHz)
info.R1p.v0=[850 850]; %B0 Field of R1p measurements (1H, MHz)
info.R1p.vr=[60 60]; %MAS frequency for R1p (kHz)
info.R1p.v1=[10 50]; %B1 field strength for R1p (kHz)
rates=calc_rates(tc,info); %Calculate rate constants
detector_opt(rates) %Run the detector_opt program

This will generate Figure 1 and Figure 2 in MATLAB (or in the DIFRATE compiled software). One then takes the following steps:
1) Initial placement. Select the ‘lock to tc’ option in the spaces window. Place detectors at all sharp corners in the projections. This amounts to 4 detectors in this case (click ‘Add r’ after placing a detection vector to add a new vector). The result is shown in Figure 3. Note the detector positions shown with colored dots, and the black lines showing how well the detectors surround the space.

Figure 3. DIFRATE spaces window after initial vector placement.

2) Optimize the placement. Typically, one should unselect ‘Lock to tc’ and check ‘Separ. Trans. & Long’. Here, we adjust the red and green detection vectors to better surround the full space. The resulting sensitivity window is shown in Figure 4. Note that some sensitivities are negative in places, and the sensitivity of the red and green detectors (2 detectors at short correlation times) are not very well separated.

Figure 4. DIFRATE sensitivity window after initial detector placement and manual optimization.

3) Computerized optimization. We now click the ‘Optimize’ button in the sensitivity window, yielding better separated and non-negative sensitivities (see Figure 5, bottom). Note that one may output the current set of detection vectors in the ‘output’ field, and if computerized optimization makes the detectors worse, one may then re-input the initial set of detection vectors.

Figure 5. DIFRATE sensitivity window after computer optimization, inclusion of S², and selection of normalization.

4) Select normalization, include S². One should finally choose the normalization mode, and if S² has been measured, then one should select the “Include S2” checkbox to include this in the eventual fit (see Figure 5). Note that these steps can be done at any point.

5) Output detection vectors. To get the detection vectors from the detector_opt program, one types a variable name into the ‘output’ field of the sensitivity window, which will then write a structure variable to the workspace (with the given name). This may be later used in the fit_data function. Note that one may output different sets of detection vectors, stored under different variable names, and try these to see fit performance (for example, one may try different numbers of detection vectors, to evaluate fit quality).

6) Further notes: although detector_opt is designed to simplify detector optimization, some experimentation may be required. One should be able to obtain good fits of the experimental data in most cases, with well-separated and non-negative detectors. One typically should use two detectors for two or more R₁ measurements (in some cases, 3 if more than 1/2 an order of magnitude is spanned by the B₀ field strength), 2-3 detectors for several R_1ρ measurements (3 if |ω_r–ω₁| spans an order of magnitude, 2 if only ~1/2 an order of magnitude), one detector for one or more R₂ measurements. With data sets that have many of the same type of measurement (but under different conditions), it can be difficult to optimally place all detectors to yield both separated, non-negative detectors and good fits. It can be helpful to get detector positions as near as possible to where the experiments are most sensitive, which is indicated by the colors of the correlation time axis on each plot.