MrIML

MrIML implements an interpretable machine learning framework in R for multilocus landscape genetics to model and interpret nonlinear and interactive relationships between genomic loci and environmental variables.

Key Features:

• Interpretable Framework: Implements interpretable machine learning methods and model-interpretation functions for inference from fitted models.
• Multilocus Genomic Analysis: Analyzes thousands of loci jointly to assess multilocus genetic variation and its associations with environmental factors.
• Nonlinear and Interactive Modeling: Captures nonlinear and interactive relationships between genetic variation and environmental variables using flexible model forms.
• Broad Analytical Framework: Supports a range of modeling techniques, including linear regression and extreme gradient boosting, within a common analytical framework.
• Extensibility Beyond Genetics: Applies the same multilocus, interpretable machine learning methods to other data domains such as microbiome studies and coinfection dynamics.

Scientific Applications:

• North American Balsam Poplar (Populus balsamifera): Modeled genetic variation across environmental gradients to elucidate genotype–environment relationships.
• Feline Immunodeficiency Virus in Lynx rufus: Identified landscape and host factors associated with viral genetic variation in bobcat (Lynx rufus) populations.
• Microbiome and Coinfection Studies: Adapted the multilocus interpretable framework to interrogate complex host–microbe and coinfection dynamics.

Methodology:

Uses simulations for method validation, enables construction, fitting, and interpretation of multilocus models, supports model comparison and inference, and implements modeling approaches from linear regression to extreme gradient boosting to capture nonlinear and interactive effects.