I searched the uniprot database by hmmsearch using a custom alignment built from selected sequence. Now I am wondering if it will be a good idea to realign the hits using Muscle or just use the hmmsearch results for building phylogenetic tree. Also should I remove the columns with >33% gap in the MSA before building the phylogenetic tree?
What you have from your search is a multiple pairwise alignment (sequences aligned to a reference, one-vs-all relationship). Phylogenetic trees basically by definition require multiple sequence alignment (they represent all-vs-all relationship), so yes, you need to re-align the hits using MUSCLE, MAFFT or whatever else is there for calculation of MSA.