Experience: Automated Prediction of Experimental Metadata from Scientific Publications

Abstract

While there exists an abundance of open biomedical data, the lack of high-quality metadata makes it challenging for others to find relevant datasets and to reuse them for another purpose. In particular, metadata are useful to understand the nature and provenance of the data. A common approach to improving the quality of metadata relies on expensive human curation, which itself is time-consuming and also prone to error. Towards improving the quality of metadata, we use scientific publications to automatically predict metadata key:value pairs. For prediction, we use a Convolutional Neural Network (CNN) and a Bidirectional Long-short term memory network (BiLSTM). We focus our attention on the NCBI Disease Corpus, which is used for training the CNN and BiLSTM. We perform two different kinds of experiments with these two architectures: (1) we predict the disease names by using their unique ID in the MeSH ontology and (2) we use the tree structures of MeSH ontology to move up in the hierarchy of these disease terms, which reduces the number of labels. We also perform various multi-label classification techniques for the above-mentioned experiments. We find that in both cases CNN achieves the best results in predicting the superclasses for disease with an accuracy of 83%.

Mixed artificial reasoning, closer to human?

“…Feature selection aims at reducing the number of variables to keep only the most relevant without changing the initial variables (Gomathi and Karlekar, 2019;Mendez et al, 2019). On the contrary, Feature extraction change the initial variables using the prior knowledge of the ontology for get relevant features (Kumar et al, 2020;Radovanovic et al, 2019;Evert et al, 2019;Agarwal et al, 2015;Radinsky et al, 2012;Greenbaum et al, 2019;Liu et al, 2021;Rinaldi et al, 2021;Castillo et al, 2008;Yilmaz, 2017;Hsieh et al, 2013;Rajput and Haider, 2011;Manuja and Garg, 2015;Ahani et al, 2021;Akila et al, 2021;Deepak et al, 2022;Messaoudi et al, 2021;Nayak et al, 2021;Pérez-Pérez et al, 2021;Zhao et al, 2021;. In semantic embedding, always in training data step, raw data are both refined by semantic knowledge and transformed into vectors to be exploited by neural networks (Chen et al, 2021;Ren et al, 2020;Qiu et al, 2019;Ali et al, 2019;Zhang et al, 2019;Makni and Hendler, 2019;Benarab et al, 2019;Moussallem et al, 2019;Gaur et al, 2019;Jang et al, 2018;Hassanzadeh et al, 2020;Ali et al, 2021;Amador-Domínguez et al, 2021;Alexandridis et al, 2021;Niu et al, 2022), SVM…”

“…Table 8 presents machine learning algorithms of each informed machine learning category. Articles were mainly published after 2017, and a great part of them concern neural networks (Hassanzadeh et al, 2020;Gaur et al, 2019;Ali et al, 2019;Jang et al, 2018;Zhang et al, 2019;Ali et al, 2021;Amador-Domínguez et al, 2021;Benarab et al, 2019;Chen et al, 2021;Wang et al, 2021bWang et al, , 2010Sabra et al, 2020;Pancerz and Lewicki, 2014;Yilmaz, 2017;Kumar et al, 2020;Rinaldi et al, 2021;Gomathi and Karlekar, 2019;Serafini et al, 2017;Kuang et al, 2021;Chung et al, 2020;Fu et al, 2015;Huang et al, 2019;Abdollahi et al, 2021;Ahani et al, 2021;Akila et al, 2021;Deepak et al, 2022;Messaoudi et al, 2021;Nayak et al, 2021;Zhao et al, 2021), especially Recurrent Neural Networks (Makni and Hendler, 2019;Ren et al, 2020;Moussallem et al, 2019;Zhang et al, 2019;Jang et al, 2018;Ali et al, 2021;Liu et al, 2021;Huang et al, 2019;Alexandridis et al, 2021;Niu...…”