Modeling:MIP
From QuB
| Prev: Modeling:MPL | Outline | Next: Modeling:Mac |
MIP (Maximum Idealized Point-likelihood) optimizes the rate constants (with error limits) of a user-defined kinetic model according to the idealization. The kinetic model will typically have more than one state for each conductance class.
The parameters of the model may be constrained according to detailed balance, fixed rate constants, and proportionality of rate constants. Multiple files, derived from data obtained at different voltages or concentrations can be analysed by MIP simlutaneously to estimate the voltage or concentration dependence (with error estimates) of all rate constants.
Differences from Modeling:MIL:
- MIP can optimize a stimulus-dependent model using data with a time-varying stimulus (recorded on its own A/D channel).
- MIP operates like a faster, noise-free MPL. A dwell is treated as a sequence of identical data points. Unlike in MIL, there's no assumption that the mechanism stays in one class for the entire duration, only that it was observed in that class at intervals of Δt. Thus, there is an implicit correction for missed events shorter than Δt. By contrast, Modeling:MIL lets you correct for missed events shorter than tdead, a user-input threshold.
- like MIL, MIP can used fixed initial (entry) probabilities (entered in State Properties), but MIP can also use equilibrium entry probabilities (from the latest rate constants) or conditioning equilibrium: if for example you held voltage low, then flipped it high and started recording, MIP can initialize the state vector from equilibrium at the low voltage.
- minimum and negative rates (?)
The details are explained in Dr. Lorin Milescu's thesis.
Properties
| Data channel index | which A/D channel contains the idealized data, typically 0 |
|---|---|
| Quiet Output | print dramatically less info to the Report window |
| Data source | |
| Ligand, Voltage, ... | experimental conditions as needed by the model A variable can have a "Value" or take its changing value from an A/D channel. The "Conditioning" value is used to calculate conditioning equilibrium probability. |
| Add/Delete/Presets | (ignore these) add/delete experimental variables, and load/save all variables |
| Use | (column when Data source is File list) whether each file will be part of the file list |
| LL conv | stop if LL increases by less than this much |
| Grad conv | stop if all gradients are less than this much |
| Max iter | at most how many times to repeat (calculate LL and gradient, modify parameters) |
| Max step | how much to change parameters each iteration. 1.0 is the natural step size; smaller numbers may be more reliable for sensitive models, but will converge more slowly. |
| Max power | MIP runs faster than MPL by combining 2n adjacent identical data points, for the largest possible n < = Maxpower. In the algorithm, A = eQΔt, B is the diagonal observation-probability matrix of 1s and 0s, and St + 1 = St * A * B is how the forward probability advances. The max power optimization advances 2n points at once by  . If you are experiencing numerical instabilities, or wish to rule them out, set Max power to 0
|
| Search limit | keep parameters within [initial / searchlimit, initial * searchlimit] |
| Restart once | if it stops after Max iterations, restart it (works better than increasing Max iterations) |
| Min rate | |
| Allow neg rates | |
| Run mode | optimize (maximize LL) or check (compute LL with current parameters) |
| Initial probabilities | Fixed: use the "starting prob" from State Properties, normalized to sum to 1.0 Equilibrium: start from equilibrium, using the initial or constant conditions and latest rate constants Conditioning Eq: start from equilibrium, using the "conditioning" value of the stimulus. |
| Batch (segments) | Together: Runs MIP with all the data at once, summing LL and generating one final model In groups of: In groups of: |
| Max batch | |
| Identical segs | |
| MUX files | |
| Presets |
Results
A ton of relevant info is shown in the Report window. (explain please)
| Prev: Modeling:MPL | Outline | Next: Modeling:Mac |
