# What algorithm(s) should I use for a regression training and a classification prediction?

I am trying to work on a project on MALDI-TOF MS dataset.

The dataset contains mass-spectrometry data of pure samples (1 bacterium species) and mixed samples (mixtures in known proportions of 2 bacteria species).

The topic I'd like to explore is to use mixture observations to train a classifier for predicting the pure ones. Say we have altogether 5 species (actually we have 20, just for an example). The labels of the training set are like $$(\frac{2}{3},0,\frac{1}{3},0,0),(0,0,0,0,1),$$ or $$(0,0,0,\frac{1}{2},\frac{1}{2})$$, which may contain one or two positive elements that sum up to $$1$$. However, we have the prior knowledge that the labels of the tests can only be like $$(1,0,0,0,0),(0,0,0,1,0),$$ or $$(0,0,1,0,0)$$, which stand for pure samples.

I am looking for appropriate algorithms to deal with it. Any suggestions would be highly appreciated. Or, maybe my research topic is not well-formulated?