In concept, molecules share similar conformational space also share similar biological activities, allowing us to predict the molecular biological activities by comparing molecular similarity. In our model, Ouroboros encodes the dynamic conformational space of molecules into vector representations, thereby enabling its application in similarity-based virtual screening.
Here, you only need to provide reference molecular structures and group them, and you can then screen for novel chemical structures similar to multiple groups of molecules. This serves purposes in drug discovery such as scaffold hopping, multi-target drug discovery, fragment fusion, and optimization of drug-like properties. When providing groupings, you can also assign a weight to each group, which helps Ouroboros distinguish between strong and weak reference molecules, and even to avoid off-target or poorly drug-like structures by assigning negative weights.
In addition to molecular similarity, you can also specify some molecular property predictors during virtual screening to take these molecular properties into account. In the returned files, you will see the predicted values for all the specified molecular properties. In the publicly available property predictors we provide, all prediction values are normalized to a range from 0 to 1, where 0 corresponds to the absence of the property and 1 to its presence. For example, in the case of the water solubility predictor, a value of 0 indicates poor water solubility, while a value of 1 indicates good water solubility.