Sequential feature selection and autoencoders for classification

I have been reading the documentation about sequential feature selection reported HERE. What I would to do at first is to run the code example reported there, but instead of using the classification method shown there I would replace it with a novel network made up of an autoencoder and a softmax output layer (as described HERE) so that I could use it in place of

fun = @(XT,yT,Xt,yt)...
      (sum(~strcmp(yt,classify(Xt,XT,yT,'quadratic'))));

In principle, function classify could be used for autoencoders as well, being derived objects in a OOP perspective. How could this issue be addressed?