Figure 1.
An overview of HypothesisFinder development approach.
The workflow for the development of HypothesisFinder shows how the model was trained, optimized and on what data sets its performance was evaluated.
Figure 2.
Classification of speculative patterns.
Figure presents examples of strong, moderate and weak speculative patterns along with their estimated ‘percent efficacy’ or ability of pattern to cast a sentence as speculative.
Figure 3.
Example showing usage of HypothesisFinder integrated in SCAIView for extracting hypotheses related to Alzheimer's disease.
Figure shows how HypothesisFinder is used within SCAIView in conjugation with other pre-indexed terminologies and ontologies to retrieve Alzheimer-specific hypotheses. Presented example shows how a hypothesis positioning Tau and Amyloid-beta as potential biomarker candidates in relation to AD is identified by HypothesisFinder in scientific abstracts.
Table 1.
Performance of HypothesisFinder on the HYPO–TEST corpora.
Table 2.
Performance of HypothesisFinder over Bioscope and Remote corpus.
Figure 4.
Comparison of information densities: HypothesisFinder vs.
AlzSWAN. A- The statistical comparison between the numbers of hypotheses related to AD captured by HypothesisFinder within SCAIView (stage-specific retrieval) and the hypotheses with extended annotation derived from citations mentioned in the AlzSWAN database. B- A comparison between biological entity retrieval using SCAIView and relevant entries in AlzSWAN.
Figure 5.
Chronological order of hypotheses proposed in Mild (A), Moderate (B) and Severe (C) AD.
Figure shows a schematic representation of how AD stage specific hypothesis related to top five genes that are high-frequently investigated in the literature has evolved in number over time Abbreviations mentioned stands for Amyloid beta (A4) precursor protein (APP), Apolipoprotein E (APOE), Microtubule- associated protein tau (MAPT), Choline O-acetyltransferase (CHAT), Brain-derived neurotrophic factor (BDNF), Beta-site APP-cleaving enzyme 1(BACE 1), Galanin prepropedtide (GAL).
Figure 6.
Figure presents protein interaction networks for Mild(A), Moderate(B), Severe(C) stage of Alzheimer's disease. These stage specific networks have been generated by using BioNetBuilder plugin in Cytoscape, which was given genes and proteins, associated to stage-wise hypotheses as input.