Algorithm 1 The extended RDA model

1: Input: Observed genotype matrix $X\in R^{n\times m}$; Index set of observed entries $\Omega$ 2: Initialize $L_D\in R^{n\times m}$, $S\in R^{n\times m}$ to be zero matrices, $LS\leftarrow X$, and an autoencoder $D\left(E\left(·\right)\right)$ with randomly initialized parameters. 3: while True do: 4:   $L_D=X-S$ 5:   $\theta =\underset{\theta }{argmin}\Vert L_D-D_{\theta }{\left(E_{\theta }\left(L_D\right)\right)\Vert }_2$ 6:   $L_D=D_{\theta }\left(E_{\theta }\left(L_D\right)\right)$ 7:   $S=X-L_D$ 8:   ${\mathcal{P}}_{\Omega }\left(S\right)=prox\left({\mathcal{P}}_{\Omega }\left(S\right)\right)$ 9:   $c_1=\Vert X-L_D-{S\Vert }_2/{\Vert X\Vert }_2$ # check the convergence condition 10:  $c_2=\Vert LS-L_D-{S\Vert }_2/{\Vert X\Vert }_2$ # check the convergence condition 11:  if $c_1<\epsilon$ or $c_2<\epsilon$: break 12:  $LS=L_D+S$ # update $LS$ for convergence checking in the next iteration 13: return $L_D$ and $S$