Background: As the amount of publicly available biomedical data increases, discovering hidden knowledge from biomedical data (i.e., Undiscovered Public Knowledge (UPK) proposed by Swanson) became an important research topic in the biological literature mining field. Drug indication inference, or drug repositioning, is one of famous UPK tasks, which infers alternative indications for approved drugs. Many previous studies tried to find novel candidate indications of existing drugs, but these works have following limitations: 1) models are not fully automated which required manual modulations to desired tasks, 2) are not able to cover various biomedical entities, and 3) have inference limitations that those works could infer only pre-defined cases using limited patterns. To overcome these problems, we suggest a new drug indication inference model.
Methods: In this paper, we adopted the Typed Network Motif Comparison Algorithm (TNMCA) to infer novel drug indications using topology of given network. Typed Network Motifs (TNM) are network motifs, which store types of data, instead of values of data. TNMCA is a powerful inference algorithm for multi-level biomedical interaction data as TNMs depend on the different types of entities and relations. We utilized a new normalized scoring function as well as network exclusion to improve the inference results. To validate our method, we applied TNMCA to a public database, Comparative Toxicogenomics Database (CTD).
Results: The results show that enhanced TNMCA was able to infer meaningful indications with high performance (AUC = 0.801, 0.829) compared to the ABC model (AUC = 0.7050) and previous TNMCA model (AUC = 0.5679, 0.7469). The literature analysis also shows that TNMCA inferred meaningful results.
Conclusions: We proposed and enhanced a novel drug indication inference model by incorporating topological patterns of given network. By utilizing inference models from the topological patterns, we were able to improve inference power in drug indication inferences.