AbstractMotivationMuch effort has been invested in the identification of protein-protein interactions using text mining and machine learning methods. The extraction of functional relationships between chemical compounds and proteins from literature has received much less attention, and no ready-to-use open-source software is so far available for this task.MethodWe created a new benchmark dataset of 2,753 sentences from abstracts containing annotations of proteins, small molecules, and their relationships. Two kernel methods were applied to classify these relationships as functional or non-functional, named shallow linguistic and all-paths graph kernel. Furthermore, the benefit of interaction verbs in sentences was evaluated.ResultsThe cross-validation of the all-paths graph kernel (AUC value: 84.2%, F1 score: 81.8%) shows slightly better results than the shallow linguistic kernel (AUC value: 81.6%, F1 score: 79.7%) on our benchmark dataset. Both models achieve state-of-the-art performance in the research area of relation extraction. Furthermore, the combination of shallow linguistic and all-paths graph kernel could further increase the overall performance. We used each of the two kernels to identify functional relationships in all PubMed abstracts (28 million) and provide the results, including recorded processing time.AvailabilityThe software for the tested kernels, the benchmark, the processed 28 million PubMed abstracts, all evaluation scripts, as well as the scripts for processing the complete PubMed database are freely available at https://github.com/KerstenDoering/CPI-Pipeline.Author summaryText mining aims at organizing large sets of unstructured text data to provide efficient information extraction. Particularly in the area of drug discovery, the knowledge about small molecules and their interactions with proteins is of crucial importance to understand the drug effects on cells, tissues, and organisms. This data is normally hidden in written articles, which are published in journals with a focus on life sciences. In this publication, we show how text mining methods can be used to extract data about functional interactions between small molecules and proteins from texts. We created a new dataset with annotated sentences of scientific abstracts for the purpose of training two diverse machine learning methods (kernels), and successfully classified compound-protein pairs as functional and non-functional relations, i.e. no interactions. Our newly developed benchmark dataset and the pipeline for information extraction are freely available for download. Furthermore, we show that the software can be easily up-scaled to process large datasets by applying the approach to 28 million abstracts.
Benchmark dataset for mid-price forecasting of limit order book data with machine learning methods
