Like Josh I would express some concern at what appears to be a desire to use a program to fix your tree. Single gene phylogenies will not always reflect the expected organismal phylogenetic relationships for any number of reasons. Some of them biological, and some due to potential methodological issues. When you start looking at gene families this is particularly true. Multiple duplications of some genes likely exist, which can confuse analysis if you are not careful with your selection of orthologs and paralogs.
Improper taxon sampling can create issues (missing or rogue taxa can both create problems, and for different reasons). Differential gene loss, gene replacement, laterla transfer... all of these can create legitimate biological confusions. Long branch attraction or rejection can cause sequences to cluster together when they shouldn't. Composition effects can cause the same thing through convergent evolution.
And, keep in mind that your underlying annotations may possibly be incorrect as well. I've seen it happen in many datasets.
My suggestions are to evaluate first your method of reconstructing the ohylogeny. For instance you should be using a full maximum-likelihood based method (or bayesian) instead of simple neighbor-joining methods. If you are using an ML or Bayesian method then evaluate the model you are using (LG versus WAG or JTT for instance with proteins). Using a program like ModelTest may take time but it will give you some insight if you are misspecifying your model.
Go through and check your ortholog/paralog selections carefully. Add taxa if you are undersampling diversity (this is usually one of my main criticisms of many papers). If some taxa look problematic do some tests to see why they are causing problems. If you do decide to remove any taxa or sequences you probably need to explain why in the publication and have a good justification for it and if you do remove sequences/taxa you need to actually redo the phylogeny as it may effect other branches and branch lengths.