Automatic Biomedical Hypothesis Generation Using T-PAIR Temporal Node-pair Embedding

PROJECT TITLE :

T-PAIR Temporal Node-pair Embedding for Automatic Biomedical Hypothesis Generation

ABSTRACT:

In this paper, we investigate a problem known as automatic hypothesis generation (HG). HG refers to the process of discovering meaningful implicit connections between scientific terms such as diseases, chemicals, drugs, and genes that are taken from databases of published biomedical research. These terms can also include other types of scientific terms. The vast majority of earlier investigations into this issue concentrated on the application of static information about terms and largely disregarded the temporal dynamics of scientific term relations. Even when the dynamics were taken into account in a few studies that were done more recently, the participants learned the representations for the scientific terms rather than concentrating on the term-pair relations. It is not sufficient to know with whom the terms are connected in order to solve the HG problem because the goal of the problem is to predict term-pair connections; rather, it is more important to know how the connections have been formed (in a dynamic process). In order to solve this HG problem, we model it as a prediction of the future connectivity of a dynamic attributed graph. The essential step here is to seize the changing dynamics of node-pair (term-pair) relations over time. T-PAIR is the name of the inductive edge (node-pair) embedding method that we have proposed. This method makes use of both the graphical structure and the node attribute in order to encode the temporal node-pair relationship. We demonstrate the usefulness of the proposed model by applying it to three real-world datasets. These datasets are graphs that were constructed using Pubmed papers that were published until 2019 in the fields of Neurology, Immunotherapy, and Virology, respectively. Evaluations were carried out in both the area of predicting future term-pair relations between millions of already seen terms (in the context of transduction), and also in the area of evaluating relations involving previously unseen terms (in the inductive setting). The effectiveness of the proposed model has been demonstrated through both experimental findings and case study analyses.

Did you like this research project?

To get this research project Guidelines, Training and Code... Click Here