Pairwise Similarity of Polymer Ensembles

In contrast to small molecules having well-defined molecular structures, polymers are rarely a single, well-defined structure. Instead, they are ensembles composed of polymer chains with various sequences, topologies, and molar mass distributions. While there are numerous approaches to measuring the pairwise similarity of small molecules, accurately calculating the pairwise similarity between polymer ensembles is still challenging. A common approach is to generate a single embedding vector for each ensemble and subsequently compute the distance between these vectors to yield a similarity score. For example, the embedding of a random copolymer can be defined by a weighted average of embedding vectors of the repeat units. However, these common average approaches neglect significant features. Here, we explicitly consider all the entities of the ensembles in the calculation of the similarity score and show that our method captures differences that are neglected using the average method, enabling us to resolve subtle differences between molecular distributions. Our method presents a critical step to quantitatively calculate polymer ensemble similarity and enable nearest neighbor search in polymer database.