next up previous index
Next: Assuming Pseudo-Steady-State Up: How to create a Previous: Model Approximations and Assumptions   Index

Click for printer friendely version of this HowTo

Implicit Models

When I run an experiment, I have two types of models running around in the back of my head. One is a model of the underlying process that I am studying, mechanisms of ion channel blockade. This model is a piece of cake. I make measurements, then fit the measurements to the values predicted by the underlying physical model of ion channel blockade.

But also an implicit model is running around in the back of my head. An implicit model that reflect data manipulations I do in order to salvage results from an experiment, that for whatever reason, is not stable. For example, it is well known that in whole cell voltage clamp studies, the preparation "runs down" over the course of the study. By this I mean that if you do nothing except make a measurement (peak Na current), every minute for 20 minutes, the results will show a gradual reduction of peak Na current - perhaps as much as 10 - 20 percent, sometimes even larger.

When an experimentalist have rundown in a preparation, the traditional analysis strategy is to "normalize" the data - i.e. to make a measurement, apply the intervention (superfusing the cell with a drug) and divide or subtract (depending on the situation) the first "control" result from each of the measurements made during the intervention. Then the next intervention would be to wash out the drug by superfusing with a drug-free solution. If there has been little rundown, then the peak Na current after washout will be similar to that before the drug was applied. This, however is rarely the case.

With normalization, we are imposing a model on the data. We are assuming that the changes in our measurements are due only to rundown or the intervention and nothing else. Often this is the case, but there are situations where this is not the case.

Consider the study of lidocaine block of cardiac Na channels. It is well known that the fraction of blocked channels changes with the transmembrane potential of the cell. Hyperpolarized cells experience little block whereas depolarized cells experience significant block. Now we study the voltage dependence of lidocaine block. We make a control measurment, apply the drug, make measure the fraction of blocked channels associated with each pulse of a train of depolarizing pulses. Then we depolarize the holding potential, and repeat this protocol: control pulse, train of pulses.

Now, since we know that lidocaine blockade is dependent on the transmembrane potential, (holding potential), if we divide each current measured during the train of pulses by the control pulse, we correct not only for rundown, but also we abolish the known voltage dependence of lidocaine blockade. Superposition of an explicit and implicit model occurs often, simply because the experimenter is unaware that normalization of data actually carries with it, an alteration of data that is equivalent to adding a component to the model. Unfortunately, this hidden addition to the model goes unnoticed and analyses can lead to incorrect conclusions due to confounding of effects described by the explicit and implicit model.



Subsections
next up previous index
Next: Assuming Pseudo-Steady-State Up: How to create a Previous: Model Approximations and Assumptions   Index

Click for printer friendely version of this HowTo

Frank Starmer 2004-05-19
>