This is clearly doable and we intend to add the feature in the near future. You will need to register in order to save and reuse the biomarker models next time, as uploading a model (executable code) to our server is too risky.
Please note if your data is not too big, when your future samples are ready, you can combine all samples for prediction. This could reduce potential batch effects which negatively impact most biomarker models performance.
When you have sufficient data, the best practice to have clean separation of training and testing datasets (including the pre-processing stage). Please refer to this excellent paper on common pitfalls using machine learning methods for prediction.