Table 1 Summary, strength, and limitation of each method of machine learning and systems genomics (MLSG) software frameworks

Model-based integration (MBI)

Multiple predictive models are generated by using various multi-omics data types; then a final predictive model is generated by using the multiple models.

Predictive models can be consolidated from various multi-omics data types, and each data type can be gathered from a various set of patients with same phenotype.

It may be challenging to avoid overfitting.

Concatenation-based integration (CBI)

Multiple data matrices of different multi-omics data types are incorporated into a large input matrix; then a predictive model is generated by using the large input matrix.

It is fairly easy to leverage various machine learning methods for analyzing continuous or categorical data once a large input matrix is formed.

It may be challenging to combine a large input matrix.

Transformation-based integration (TBI)

Datasets for various multi-omics data types are first converted into intermediate forms, which are united into a large input matrix; then a predictive model is generated by using the large input matrix.

Unique variables such as patient identifiers can be used to link multi-omics data types and integrate a variety of continuous or categorical data values.

It may be challenging to transform into intermediate forms.