Skip to main content

Table 12 Characteristics of the multi-omics integration methods

From: A fair experimental comparison of neural network architectures for latent representations of multi-omics for drug response prediction

Architecture

Training

Triplet loss

Integration type

Encoding

Early integration

End-to-end

–

Early

Supervised encoder

MOLI

End-to-end

+

Intermediate

Supervised encoder

Super.FELT

Encoding and classifying

+

Intermediate

Supervised encoder

Omics stacking

End-to-end

+

Intermediate + late

Supervised encoder

MOMA

End-to-end

–

Intermediate + late

Vector encoding

OmiEmbed

Three phases

–

Intermediate

Variational supervised autoencoder

PCA

PCA and classifier

–

Intermediate

Principal components