Table 1 Source discrimination results of SiO₂ NPs of known sources into five classes by the machine learning model^a

Sample	Total	Source identified^b					Accuracy
Sample	Total	EP	EF	ES	NQ	ND	Accuracy
SiO₂ NPs	90	Number of correct: 84^c					93.3%
└ Engineered NPs	50	49^d			1^e		98.0%
└ EP	28	27	0	0	1	0	96.4%
└ EF	15	3	12	0	0	0	80.0%
└ ES	7	5	0	2	0	0	28.6%
└ Natural NPs	40	5^d	35^e				87.5%
└ NQ	20	0	0	0	20	0	100%
└ ND	20	5	0	0	0	15	75.0%

^a Engineered NPs were collected from 14 manufacturers located in 6 different regions. Natural NPs included both real and pseudo-samples (see Supplementary Section 1.4). For real samples, NQ samples were collected from 9 manufactures and ND samples were from 3 manufactures. More details about samples are given in Supplementary Table 1 and 2. The source discrimination results into three and four classes are given in Supplementary Table 6 and 7.
^bThe machine learning model could give a probability value for each candidate source (see Supplementary Table 3), and the statistics in this table was based on the most probable source.
^c The “number of correct” means the number of samples with correct discrimination result between engineered and natural SiO₂ NP.
^d The total number of engineered SiO₂ NPs identified (EP + EF + ES).
^e The total number of natural SiO₂ NPs identified (NQ + ND).

Quick links

Search