Cocrystals in the Cambridge Structural Database: a network approach

Abstract

To obtain a better understanding of which coformers to combine for the successful formation of a cocrystal, techniques from data mining and network science are used to analyze the data contained in the Cambridge Structural Database (CSD). A network of coformers is constructed based on cocrystal entries present in the CSD and its properties are analyzed. From this network, clusters of coformers with a similar tendency to form cocrystals are extracted. The popularity of the coformers in the CSD is unevenly distributed: a small group of coformers is responsible for most of the cocrystals, hence resulting in an inherently biased data set. The coformers in the network are found to behave primarily in a bipartite manner, demonstrating the importance of combining complementary coformers for successful cocrystallization. Based on our analysis, it is demonstrated that the CSD coformer network is a promising source of information for knowledge-based cocrystal prediction. research papers Acta Cryst. (2019). B75, 371-383 Devogelaer et al. Cocrystals in the CSD: a network approach 383

Cocrystal design by network-based link prediction

“…Since the AUC of RA is larger than AA's, RA was selected as the prediction method of choice. ¶ Due to several improvements made to our classifier algorithm (including a better neutrality check) described in the study of Devogelaer et al, 34 this number is slightly different from our earlier work.…”

“…In a recent publication, 34 we have shown how the cocrystals in the CSD can be used to build a network G(N,E), formed by a set of nodes N (in this case coformers) and a set of edges or links E between these nodes (representing the cocrystals). A network is commonly represented as a (symmetrical) adjacency matrix A ∈  N×N , for which the indices of the rows and columns correspond to the nodes, and for which the elements are labeled as 1 for known node combinations (and as 0 otherwise).…”

“…1, it is clear that both scoring methods express the score using a measure for the density of the intermediate network between two coformers: only relationships between common neighbors contribute, and neighbors with a more diverse cocrystallization profile are penalized (via the degree k in the denominator). The reason that these methods perform so well can be attributed to the structure of the network itself: a comprehensive analysis 34 has shown that existing cocrystals are characterized by the presence of local bipartite communities between them. Hence, methods capable of condensing this structural phenomenon effectively into their score value are expected to perform better.…”

“…An example of a bipartite network is the network formed by salts: the network consists of a set of cations and anions, and salts can only be formed by combining opposite ions. We have recently analyzed the type of the coformer network 34 and have found that it implicitly behaves in a bipartite way. Therefore, the link-prediction methods were selected or adapted to this network type, which requires the formulation of two additional bipartite properties.…”

“…In this paper, we introduce a new knowledge-based cocrystal prediction method based on a network of coformers and linkprediction algorithms. In our previous work, 34 we have demonstrated how cocrystals in the Cambridge Structural Database 35 (CSD) can be transformed into a network of coformers and shown how clusters, a quantification of the socalled popularity bias and the type of aggregation behavior, can be extracted from this network. Here, we combine several techniques from network science and classification to predict new cocrystals based on the information contained in the coformer network.…”