Comparison of computational approaches for maximum likelihood estimation on compositional data

Compositional data can arise in many ways and with omics-technologies high dimensional compositional data can be easily produced. The number of available parametric distributions to analyse these data is not huge and for some of them computational methods are needed to find the maximum likelihood estimates. Good starting values strategies and reliable computational methods are essential to provide convergence. In this work we compare the computational stability and efficiency of different approaches on a specific distribution suggested to analyse these data. Their performances will be evaluated on real and simulated data