How do I "predict" output of a lasso derived regression model?

I have a set of ~50 predictors and ~100 observations. I am using lasso to find the smallest set of predictors that can reliably predict a certain observation (observations are categorial yes or no responses). Everything works find and I find my set of minimal set of predictors given by the smallest lambda value. MSE is around 0.2.

My problem is, the "predict" function in matlab doesn't seem to have an option for lasso and even if it did, the "lasso" function in matlab doesn't seem to output any kind of model to use as input for "predict".

Eventually I would like to generate an ROC curve to evaluate the performance of the model. For that I need to run a test set and calculate true and false positive rates.

Any ideas? Am I missing something? Thanks in advance guys.