Correction: Predictive modeling for odor character of a chemical using machine learning combined with natural language processing (PLoS ONE (2018) 13, 6 (e0198475) DOI: 10.1371/journal.pone.0198475)

Abstract

There are errors in S2 Table. The values in the “Applicable descriptor Number (from S1 Table)” column are incorrect and should be 1 value lower. Additionally, the sensory dataset which was used in the computer simulation in S2 Table, is not perfectly equivalent to the original data source, Sigma-Aldrich’s “Flavors and Fragrances” [2]. Therefore, S2 Table data are different from the original source. Some of the descriptors may be ignored for samples described by more than 6 descriptors. As descriptors are listed ascending in alphabet, ignored descriptors are mainly: “sweet”, “vanilla” and “wine-like”. Those descriptors are used when the number of descriptors is not more than six. Approximately 11% of samples in the dataset affected. Please see the corrected S2 Table caption and file below. Supporting information S2 Table. Odor character profile of chemicals. S2 data set is different from original source. (CSV)