Transcripts were clustered using CD-HIT (Cluster Database at High Identity with Tolerance) package. Used to remove the shorter redundant transcripts when they were 100% covered by other transcripts with more than 90% identity. The non-redundant clustered transcripts were then designated as unigenes.

If you want a common "baseline" for several assemblies, the only way I could think of would be annotation with a common database. For instance, if your assemblies come from closely related species, you could annotate CDSs with a common reference (e.g. a closely related model species).

For individual assemblies, if you would like to go from transcript-level to gene-level (which has advantages), you could also cluster transcripts using Corset, or Grouper based on shared read support.

Another way forward would be to combine both techniques. For each assembly, you could cluster assembled transcripts using one of the two previously mentioned programs. You could then construct SuperTranscripts using Lace. You could then try and annotate supertranscripts, and compare (likely) homologous genes with one another. Hope that helps.