Algorithm 1:The ssPCA algorithm

Given the cluster labels $\mathbf{y}$, scRNA-seq data X, similarity (kernel) matrix $K_X$ and hyperparameter $\lambda$:   1. Recode $\mathbf{y}$ into a binary matrix Y, calculate $K_Y=Y{Y}^T$, and compute       $K_X$ from X if $K_X$ is not available.   2. Find W, the eigenvectors of $(1-\lambda )K_{X}H{K}_X+\lambda K_{X}H{K}_{Y}H{K}_X$ corresponding to       the k largest eigenvalues.   3. Project to low-dimension with $Z=K_{X}W$ for cluster visualization.