Find Alternative Biomarker via Word Embedding
Kyeong-Min Nam, Hey-Jung Song, Jong-Dae Kim, Can-Young Park, Yu-Seop Kim · Advanced science and technology letters · 2015
We use word embedding to find an alternative biomarker for the early diagnosis of ovarian cancer from the biomedical corpus. First, we derived a low dimensional representation of each biomarker embedding induced from CCA. Second, we found a similar pair of biomarkers in the literature by using cosine similarity of biomarker embedding. In order to determine the clinical similarity between the pair of biomarkers, we used the AUC of the combination of 2 biomarkers used previously. In the experiment, we confirmed that correlation between the high similarity biomarker pair, was highly correlated as the average 0.698 of the actual AUC correlation of the top 10 of the pair.