MOCSS: Multi-omics data clustering and cancer subtyping via shared and specific representation learning

. 2023 Jul 13;26(8):107378. doi: 10.1016/j.isci.2023.107378

Algorithm 1: Shared and Specific Representation Learning
Input: Multi-omics dataset $X = {X^{1}, X^{2}, \dots, X^{V}}$ , number of clusters $K$ , batch size $M$ , temperature parameter $τ$ 1: Normalization 2: Initialize the parameters of autoencoders and projection head $f (\cdot)$ 3: While not reaching the maximum epoch $T$ do 4: randomly select M samples from $X^{v}$ 5: generate shared and specific representation from eachomic using $E q . (1)$ - $(4)$ 6: compute reconstruction loss $L_{r e c}$ by $E q . (5)$ 7: compute contrastive loss $L_{c o n}$ by $E q . (6)$ - $(10)$ 8: compute orthogonality loss $L_{o r t}$ by $E q . (11)$ 9: compute overall loss $L$ and updata entire network by $E q . (12)$ 10: generate the shared information matrix $C^{v}$ and specific information matrix $S^{v}$ for all samples 11: End while Output: Shared information matrix $C^{v}$ and specific information matrix $S^{v}$

Algorithm 1: Shared and Specific Representation Learning

Input: Multi-omics dataset $X = {X^{1}, X^{2}, \dots, X^{V}}$ , number of clusters $K$ , batch size $M$ , temperature parameter $τ$
1:
Normalization
2:
Initialize the parameters of autoencoders and projection head $f (\cdot)$
3:
While not reaching the maximum epoch $T$ do
4:
randomly select M samples from $X^{v}$
5:
generate shared and specific representation from eachomic using $E q . (1)$ - $(4)$
6:
compute reconstruction loss $L_{r e c}$ by $E q . (5)$
7:
compute contrastive loss $L_{c o n}$ by $E q . (6)$ - $(10)$
8:
compute orthogonality loss $L_{o r t}$ by $E q . (11)$
9:
compute overall loss $L$ and updata entire network by $E q . (12)$
10:
generate the shared information matrix $C^{v}$ and specific information matrix $S^{v}$ for all samples
11:
End while
Output: Shared information matrix $C^{v}$ and specific information matrix $S^{v}$