Figure 1: Conceptualization of the (pseudo) two-dimensional mapping
using the Clustering NMR method proposed in this work. A pair of
(T1, T2) relaxation time for each
objects (e.g., edible oils, blood) were measured using micro NMR
relaxometry system. A (pseudo) two-dimensional map is constructed with
(T2, T1) relaxation time with a (X, Y)
scatter plot (Fig. 2c), where the object clustering became obvious in
comparison to its´ one-dimensional counterparts (i.e, T1relaxation or T2 relaxation). The efficacies of
Clustering NMR method were validated using both the supervised and
unsupervised learning methods. The relationship between each objects is
established using (unsupervised) clustering analysis methods (e.g., tree
classfication, hierarchical clustering) and its´ quantitative linkage
(e.g., inter/intra cluster similarity) of each objects which is depicted
on a dendogram with a heat map (details in Supp. Figs. 2-3). Supervised
learning techniques (e.g., kNN, random forest, logistic regression) were
used to train the classification of objects and the best trained model
is subsequently chosen to predict the object classification. (e.g., oils
content, infection/non-infection).
Methods
NMR measurement and detection. The relaxometry measurements
(T1 relaxation, T2 relaxation) were
carried out on four group edible oils (i.e., peanut, olive, sunflower,
corn) labelled as (A, B, C, D), respectively (Figs. 2a-b). In order to
avoid bias, more than one different manufacturers were used for the same
oil (with the exception of corn oil) and the detail on fat compositions
were presented in Supp. Fig. 1. (A, A´, B, B´, C, C´, C´´, D) were the
variants of the same oil from various manufacturers. The manufacturer
labelling indicated 100% of oil contents (no mixture of oils). The
edible oils were cooking oils bought locally in Braga, Portugal. No
further alteration was made before the NMR measurements.