Currently the graph of the proportion of precursors detected against RT error (which is used for automatic optimization) is relatively flat. This new set of convergence criteria and new way of finding the optimum (based on most narrow tolerance with proximity to the maximum).
An example of the new behaviour is shown below, where the value taken forward for search is distinct from the value that maximises the number of precursors.
NB: although this example shows the behaviour of the new rule, it is perhaps not the best illustration of why this change is useful. Often the curve is much flatter and the maximum is only slightly greater than all the other points, in which case choosing the narrower value is typically preferable. This change is also justified by the curve in our ground truth optimization (as shown in the precursor identifications in the plot below):
Currently the graph of the proportion of precursors detected against RT error (which is used for automatic optimization) is relatively flat. This new set of convergence criteria and new way of finding the optimum (based on most narrow tolerance with proximity to the maximum).
An example of the new behaviour is shown below, where the value taken forward for search is distinct from the value that maximises the number of precursors.
NB: although this example shows the behaviour of the new rule, it is perhaps not the best illustration of why this change is useful. Often the curve is much flatter and the maximum is only slightly greater than all the other points, in which case choosing the narrower value is typically preferable. This change is also justified by the curve in our ground truth optimization (as shown in the precursor identifications in the plot below):