Developing a Deeper Understanding of Autism: Connecting Knowledge through Literature Mining

Abstract

In the field of autism, an enormous increase in available information makes it very difficult to connect fragments of knowledge into a more coherent picture. We present a literature mining method, RaJoLink, to search for matched themes in unrelated literature that may contribute to a better understanding of complex pathological conditions, such as autism. 214 full text articles on autism, published in PubMed, served as a source of data. Using ontology construction, we identified the main concepts of what is already known about autism. Then, the RaJoLink method, based on Swanson's ABC model, was used to reveal potentially interesting, but not yet investigated, connections between different concepts in research. Among the more interesting concepts identified with RaJoLink in our study were calcineurin and NF-kappaB. Both terms can be linked to neuro-immune abnormalities in the brain of patients with autism. Further research is needed to provide stronger evidence about calcineurin and NF-kappaB involvement in autism. However, the analysis presented confirms that this method could support experts on their way towards discovering hidden relationships and towards a better understanding of the disorder.

Bisociative Literature Mining by Ensemble Heuristics

“…Tf-idf based (7) : sum of document tf-idf weights of a term in a dataset, (8) : average of document tf-idf weight of a term in a dataset, (9) : product of domain centroid tf-idf weights of a term, (10) : sum of domain centroid tf-idf weights of a term. (11) : similarity of a term to an average dataset document -the distance of a term to the dataset centroid, (12) : product of similarities of a term to domain centroids, (13) : min of similarities of a term to domain centroids.…”

“…appearInAllDomn 93,31% (20) outFreqRelSVM 95,06% (10) tfidfDomnSum 93,85% (12) simDomnProd 93,27% (18) outFreqRelCS 94,96% (6) freqDomnProdRel 93,71% (1) freqTerm 93,20% (17) outFreqSum 94,96% (13) simDomnRatioMin 93,58% (2) freqDoc 93,19% (8) tfidfAvg 94,87%…”

“…─ Representatives of BoW heuristics based on frequency and tf-idf were chosen in a way which tries to avoid overfitting the migraine-magnesium dataset. We chose (3) freqRatio and (10) tfidfDomnSum with the reasoning that they are not among the best performing on the training dataset (but we expect them to perform better on other datasets) and will therefore act as a counterweight to prevent overfitting. ─ We completely discarded all the heuristics of the type similarity, as their performance is in the range of the baseline heuristic (23) appearInAllDomn.…”

“…1 shows no surprises among the base heuristics, since the results are equal to those presented in our previous work (see [7]), however, when focusing on the ensemble heuristic, we notice that it is better in both, higher AUC value and lower AUC interval compared to all the other heuristics. We constructed the ensemble using also two not so well performing heuristics ( (10) tfidfDomnSum and (3) freqRatio) in order to avoid overfitting on the training domain. This could have a negative effect to the ensemble performance, however, the ensemble performance was not seriously affected which signals an evidence on the right decisions when designing the ensemble.…”

An ontology for Autism Spectrum Disorder (ASD) to infer ASD phenotypes from Autism Diagnostic Interview-Revised data

“…Additional ontologies related to ASD have been developed to support other use-cases, such as automated ontology construction by performing text-mining [32, 33] or ontology-based information retrieval [34, 35]. The aforementioned ontologies differ from the work described in this paper as they were created in an automatic process to support their ultimate goal of text-mining or information retrieval.…”