Reviews: Bayesian multi-domain learning for cancer subtype discovery from next-generation sequencing count data

I am not familiar enough with the domain to really assess the novelty of the contribution. From my reading this paper is a "model - prior - Gibbs sampler" paper which seems to improve the classification scores but does not provide breakthrough to the learning community. The novelty essentially seems to come from the choice of the prior which allows but does not require factors to be shared across domains. Moreover, the authors state that details on their Gibbs sampler are provided in the supplementary materials but I can only trust them as there seems to have been some mistake on uploading the supplementary materials (the actual manuscript was submitted instead). I would have liked to have some intuition on how to chose the parameter K, and how does its value affect the results, both in terms of subtyping and in terms of complexity. In general, how does the approach scale with competitors in terms of run-time? In the case study section, how many subtypes of lung cancer were considered? Have the authors tried their approach with more than two domains?