Lesley,
My guess is that you're misusing the word "deconvolve". The signal processing step in the brain is modeled as a convolution process, that is, the BOLD response can be theorized as the stimulus events convolved with an impulse response function. So a deconvolution of the time series at a voxel means that you're trying to find out EITHER the stimulus events based on a theoretical impulse response OR the impulse response based on the events.
If you meant to separate the BOLD response for the 4 condition in the seed time series, then you'd have to do that yourself based on the event sequence.
Gang