3 Replies Latest reply on Feb 27, 2018 3:28 PM by ryoboyle

    Can one use different regression types for different compounds but across samples? MassHunter

    kaustin

      When using MassHunter software for LC MS/MS does it bias/impact the experiment if different regression types (linear, quadratic, power) are used for different analytes?  In other words, is the experiment negatively impacted if the calibration curve is calculated using a different equation for different compounds across samples?

       

      Example: Using a quadratic curve to calculate sample concentrations for THC and using linear to calculate ******** concentrations.  Do you need to stick to all linear equations or all quadratic equations or can you mix and match to get the best R^2 value and cals/controls in spec?

       

      If mixing and matching is detrimental to the integrity of the experiment, why?

       

      Thanks 

        • Re: Can one use different regression types for different compounds but across samples? MassHunter
          ryoboyle

          Ideally you would be able to get a linear response for all of your compounds, but sometimes things like ion suppression can occur which might cause a non-linear response for some compounds in your samples. If you choose to use different curve types for different compounds in your samples the important thing is that you need to be able to justify that selection (i.e. reference peer reviewed papers, perform a study to validate the curve fit, etc...).

           

          A useful tool in MassHunter Quantitative Analysis is the Curve Fit Assistant which can help you identify which curve type will give you the best R^2 value. You can access it by right-clicking on the Calibration Curve window and selecting Curve Fit Assistant.

           

          The Curve Fit Assistant defaults to being able to disable 3 points on your calibration curve, so you would want to sort for options which have 0 disabled points.

           

          You can then select from the results whichever curve gives you the best fit. Again, you want to makes sure that you are able to justify your curve fit selection. Usually you want to stick to either a linear or quadratic curve type.

           

          Something else to take into consideration is the Weighting option for the curve. When a wide range of calibrator concentrations are used, response variance tends to increase with concentration in the calibration curve. If the default Curve Fit Weight (None) is used, the highest standards will have the greatest influence on the cure fit. In some cases, it may be useful to consider 1/x, or another inverse weighting option.

          2 people found this helpful