Bayesian Partial Least Squares and Model Comparison

Quite a bit of my work is in the chemometrics and cheminformatics fields, where Partial Least Squares (PLS) models overwhelmingly dominate. I’d be interested to compare this model to others approaches such as GPs or even neural nets, perhaps using Bayes Factors. From my understanding, though PLS involves an iterative algorithm. Can it be implemented in pymc to allow for model comparison? Is there a way to compare a pymc model with another trained via different library?