Algorithm 1: Autoencoder Algorithm. The Pseudocode of the proposed Autoencoder algorithm for 2D molecular fingerprints.

1: Mols = 2 D figerprints dataset descriptor

2: M = number of database molecules.// 102516

3: N = number of hidden layers.

4: α = learning rate value.

5: Epoch = 0;

6: For k = 1:M   // for all dataset molecules

7:  Input = Mlos(k)    // input data

8:  x = mols(k); // initial the encoder input layer with molecule k.

9:  AE(x)   // Autoencoder function

10:    For i = 1 until N   //start the encoder phase

11:        If epoch = 0 do // first time training

12:        wi = random (0,1) // initial the weigh matrix for the first time training

13:        bi = random (0,1) // initial the bias vector for the first time training

14:      Else

15:        ${\mathrm{w}}_{\mathrm{i}}={\mathrm{w}}_{\mathrm{i}}+ℒ$ // update the weigh matrix based on the error value

16:        ${\mathrm{b}}_{\mathrm{i}}={\mathrm{b}}_{\mathrm{i}}+ℒ$ // update the bias vector based on the error value

17:        ${\mathrm{h}}_{\mathrm{i}}=\raisebox{1ex}{$1$}\!\left/ \!\raisebox{-1ex}{$1+e^{-({\mathrm{w}}_{\mathrm{i}}\mathrm{x}+{\mathrm{b}}_{\mathrm{i}})}$}\right.$ //calculate the hidden layers values based on Equation (1)

18:      x = hi   // make the hidden layer values to be an input to the next hidden layer.

End   // end encoder phase

19:   Encoded date = x     //keep the last encoder layer which the new represented molecule.

20:   h = x          //keep last encoder layer be an input encoder layer.

21:   n = N

22:   For j = 1 until N    //start decoder phase

23:      ${\hat{\mathrm{w}}}_{\mathrm{j}}$ = wnT   //male the weight matrix of the decoder layer j be the transpose of n encoder weight matrix layer

24:      ${\hat{\mathrm{b}}}_{\mathrm{j}}={{\hat{\mathrm{b}}}_{\mathrm{n}}}^{\mathrm{T}}$  // make the value of the bias vector of the decoder layer j be the transpose of bias vector of n encoder layer.

25:      ${\mathrm{z}}_{\mathrm{j}}=\raisebox{1ex}{$1$}\!\left/ \!\raisebox{-1ex}{$1+e^{-({\hat{\mathrm{w}}}_{\mathrm{j}}\mathrm{h}+{\hat{\mathrm{b}}}_{\mathrm{j}})}$}\right.$  // calculate the zj reconstructed decoder layer values

26:      h = zj; // keep the hidden decoder layer values to be an input to the next hidden layer.

27:      n = n−1

End       // end decoder phase

28:  output = h     // Reconstructed data

29:  $ℒ=\Vert \mathrm{input}-{\mathrm{output}\Vert }^2$   // calculate the error value based on Equation (3)

30:  If ($ℒ>\alpha$)         // if the error value is greater than the learning rate.

31:   Epoch = epoch+1     // need more training to reduce the error value

32:   Go to 9       // call the AE function again for new training—fine tune.

Else

33:   New_Rep_mols(k) = Encoded date;//