A meganized DAA file or an RMA file based on full alignments is required...
(However, it should be possible to extended the algorithm so that blast tab suffices:
In the extended algorithm, MEGAN would have to compute all implied protein overlaps between reads...
I will try to look into this...)
You can use gene-centric assembly in two ways: EITHER by selecting a node in viewer (such as the Kegg viewer) and have all reads assigned to that node assembled, OR you can open the alignment viewer on some node, then select a reference, and then select Layout->By Contigs and then the reads assigned to the selected reference will be assembled using that reference.
The number of genes detected is defined in the paper as the number of reference genes that get covered more than half by the longest contig that maps to it:
To assess how well gene sequences are detected for different organisms, we report the number of organisms for which the longest mapped contig covers at least half of the corresponding reference sequence.