ikernInt

The software tool "ikernInt" introduces a kernel framework to bridge the gap between supervised and unsupervised analyses in microbiome research. It addresses the common challenges posed by the compositional nature of next-generation sequencing data and the integration of spatial and temporal dimensions. This comprehensive approach offers a unified platform for microbiome analysis, leveraging the strengths of supervised learning, which focuses on predicting a phenotype of interest from taxonomic abundances, and classical unsupervised analyses, which are typically based on computing ecological dissimilarities for visualization or clustering purposes.

Key Features of ikernInt:

- Compositional Kernels: ikernInt introduces two compositional kernels, the Aitchison-RBF and compositional linear, explicitly designed for handling the compositional nature of microbiome data. These kernels allow for the practical analysis of taxonomic abundances, maintaining the integrity of the data's compositional structure.

- Transformation of Beta-dissimilarity Measures: The tool discusses methodologies for transforming non-compositional beta-dissimilarity measures into kernels, facilitating a broader application of various dissimilarity measures within the kernel framework and enhancing the analytical flexibility.

- Integration of Spatial Data: ikernInt employs multiple kernel learning to integrate spatial data into the analysis, enabling the exploration of spatial patterns and relationships within microbiome communities. This feature is useful for studies investigating microbial taxa's spatial distribution and environmental interactions.

- Evaluation of Longitudinal Data: The framework includes specific kernels for the evaluation of longitudinal data, allowing for the analysis of temporal variations within microbiome communities. It is crucial for studies that aim to understand how microbiome communities evolve over time or in response to external factors.

- Microbial Signatures Retrieval: A significant advantage of ikernInt is its ability to retrieve microbial signatures, or taxa importances, providing insights into the taxa that are most influential or indicative of certain conditions or phenotypes.