Algorithm 1: Algorithm of our proposed method

Inputs: Known associations matrix $\mathrm{A}\in {ℝ}^{n_m\times n_d}$, microbe similarity matrix $K_s^m\in {ℝ}^{N_m\times N_m}$, disease similarity matrix $K_s^d\in {ℝ}^{n_d\times n_d}$; Output: The completed training of the generative network model Step 1: Constructing the heterogeneous network $Y\in {ℝ}^{\left(n_m+n_d\right)\times \left(n_d+n_m\right)}$ according to Formula (16); Step 2: Input the feature matrix into the generative network, initializing Optimizer Parameter Information; Step 3: for $i=1\to N\text{do}$ (N is the number of training rounds of the generative adversarial network) for $l=1\to L\text{do}$ (L is the depth of the graph convolution model) Compute the feature embedding of the L layer and output the generated prediction results end for Input the generated results and sample data into the decision network Update optimizer parameter information

end for Step 4: Save the model of the generative network