Table 4.

Characteristics of different omics data integration strategies.

Integration Strategy	Description	Processing	Main Drawback
Concatenation-based	Combines different datasets into a large matrix, increasing variables but not samples.	Applies feature selection or dimensionality reduction.	Makes the data matrix complex and noisy.
Transformation-based	Transforms each dataset into a less dimensional and noisy graph or kernel matrix.	Transformation into graph, kernel matrix, or deep learning architectures.	Can lead to information loss and distorted data.
Model-based	Applies separate ML models to each dataset and combines predictions.	Supervised: Bagging or voting to combine models. Unsupervised: Aggregates clustering results based on optimization criteria.	Prevents learning features shared across different omics data, limiting understanding of underlying mechanisms.