TABLE 1 |.
Step | Problem | Possible Reason | Solution |
---|---|---|---|
4 | NITPIC cuts off injections before return to baseline | Slow reaction kinetics when saturation is approached | In “Injection & Baseline Parameters”, toggle radio-button “User” in the Minimum Injection Time field, and set to a visually estimated injection length (%-value relative to total time between injections) |
4 | NITPIC includes too much of each injection in the integrations | Low signal-to-noise ratio | In “Injection & Baseline Parameters” increase “Target” and “Max” values (usually tenfold) in the entry boxes for “Cut-off differentials for inj. end” |
4 | Calculated thermogram baseline is too smooth or warning message that NITPIC uses too many shape components | Failure in automated shape truncation | In the “SVD Parameters” reduce the maximum # of SVD components |
14 | The displayed lines for the model as initialized are far away from the data points | Poor parameter initialization | There is no general mathematical solution for parameter initialization. Revise estimates manually and re-run. Reasonable values for dHAB_app may be obtained by entering the heats of the initial injections. A broad search for log10(Ka_app) by trial and error is often successful. If the inflection points do not match the data, inspect the concentration correction factor. |
15 | The resulting parameter values are physically inappropriate | The fit has not converged, too many parameters are optimized simultaneously, some of the parameters might be correlated, or the binding model is inappropriate. | When using incompetent fractions and concentration correction factors, ensure that both are not simultaneously refining. Try a multi-stage approach, first fixing a subset of parameters (for example, fix the concentration correction at 0) while other parameters optimize, then start optimizing them all in a second stage. Use a different model (see below). |
16 | The resulting fit does not resemble the data | An incorrect model was used. | Use a different model while incorporating the knowledge obtained from other experiments performed with orthogonal methods if possible. It can be useful to establish whether a data subset can be fit by de-activating individual experiments from consideration (using the small ‘i’ button in the upper right corner of the experiment plot). |
16 | An individual experiment is not fit well | Either an incorrect model was used, or an experimental error occurred. | A single aberrant fit can be caused by an erroneous experiment or incorrectly entered experimental parameters. Double-check parameters or repeat the experiment. However, the data may be correct and the conditions may make this particular experiment more informative regarding the binding model, highlighting shortcomings of the model. Considering the unique experimental conditions may lead to the identification of processes that occur in the binding reaction but might have gone unnoticed and are not accounted for. |
17 | Confidence interval is too broad or one of the intervals was not determined | Information content of the data is limited, possibly due to shallow or partial transition or low signal/noise ratio, or the model has too many or correlated parameters. | Examine the entire projection of the error surface using the “Generate 1-dim Error Surface Projection” function, and parameter correlations in the “Generate 2-dim Error Surface Projection” function of the Statistics menu. Add more informative data. Use the “Generate” function to simulate experimental data conditions in a search for better experimental conditions |
18 | GUSSI does not start up | Insufficient wait, un-answered disclaimer prompt, or location of gussi.exe | Make sure the location of gussi.exe is in a GUSSI subfolder relative to the path of sedphat.exe or itcsy.exe. The disclaimer may be hidden behind other open windows. On some computers, Python libraries will take several seconds to load. |
19 | Graph is too wide in GUSSI | Occurs stochastically on some computer systems | Press “Update”. |