MixMPLN

MixMPLN is an R software package that enables the inference of multiple microbial interaction networks from a single sample-taxa count matrix. It is based on the assumption that the sample-taxa matrix can be associated with more than one network, which is reasonable given that various environmental or host factors influence microbial community compositions and interactions.

Key features of MixMPLN:

1. Mixture model framework: MixMPLN models the count matrix using a mixture of K Multivariate Poisson Log-Normal distributions, allowing for the inference of K microbial networks.

2. Parameter estimation: The software uses a maximum likelihood framework for parameter estimation, employing a combination of the minorization-maximization principle, gradient ascent, and block updates.

3. Handling different data types: MixMPLN can handle absolute count, compositional, and normalized data.

4. Sparse network recovery: The package also addresses the recovery of sparse networks using an l1-penalty model.

5. Performance assessment: Synthetic datasets were used to evaluate the performance of MixMPLN on various data types.