In my current project, I am working on inferring a phylogeny for a particular protein. The trouble is, the topology of the tree is often discordant with the prevailing species-tree (i.e. the 'Tree of Life'). For example, whereas nematodes are typically part of superphylum Ecdysozoa and Platyhelminthes part of superphylum Lophotrochozoa, my tree indicates the two to be sister groups. I'm quite sure that such discordances in my tree are the result of artefacts such as long branch attraction (LBA) and not due to genuine hemiplasy. Given that I am going to use this tree for ancestral sequence reconstruction at key nodes, I need to the topology to reflect the true evolutionary relationships among taxa.
What I've done is used a constraint tree in IQ-TREE that enforces the species-concordant phylogeny at the phylum level so that my tree is concordant with the 'Tree of Life'. I then performed statistical topology tests (e.g. Kishino and Hasegawa, KH) to ensure that the constrained phylogeny was not statistically significantly different from the maximum-likelihood phylogeny in terms of log-likelihood.
My question therefore is: is this appropriate? Can I enforce the species-concordant topology so long as it is not significantly different (statistically) from the ML phylogeny? I am quite certain that the species-discordant phylogeny is not reflecting genuine hemiplasy or other genuine factor such as horizontal gene transfer. Any insight would be much appreciated.