To find the nearest candidate from a spatial dataset you have to read the dataset and use the NeighborFinder transformer.
To do exactly the same using the FeatureReader would now be a two-step process (Buffer the base object and use the buffered object as Initiator in the FeatureReader, followed by a NeighborFinder.
A predicate of "Find closest object" in the FeatureReader would be a performance boost and a reduction of required trtansformers. And would reduce the amount of features read into FME Workbench.

