Could machine learning be used for exploring the space of organic crystals composed of a set of different molecules, known as co-crystals? Could we achieve reliable predictions by overcoming the lack of negative data, when we do not know which pairs of molecules cannot form co-crystals? The aim of this work is to explain how different one class classification algorithms can be trained to learn the boundaries of the area in which the known molecular coformer combinations exist. The acquired ‘knowledge’ is then applied to detect novel sets of molecules that are expected to form stable structures with desirable properties. The structural characterization as well as the energetic stability assessment of some of the highest scored structures showcases that the synergy between ML workflows and high-throughput computations can accelerate the materials discovery and drive to creative and innovative solutions.