Profiling of cocaine using ratios of GC-MS peaks

TY - JOUR

T1 - Profiling of cocaine using ratios of GC-MS peaks

AU - Villesen, Palle

AU - Stride Nielsen, Louise

PY - 2017/12/1

Y1 - 2017/12/1

N2 - Illicit cocaine seizures are often compared to each other by using gas chromatography-mass spectrometry (GC-MS) data from cocaine alkaloid compounds to determine whether two specimens originate from the same production batch or not. This can provide intelligence or investigative information at the early stages of an investigation or evidence in court. Traditional classification methods assume high stability of all alkaloids, use all of them to calculate the correlation between two profiles and use a threshold to classify samples. Unstable alkaloids will have a strong influence on the performance. We show that comparing each alkaloid target compound individually improves the classification. Unfortunately, it requires normalization and is also sensitive to the stability. Instead we suggest to use ratios of all possible pairwise combinations of the GC-MS peaks. These ratios are scale free and directly comparable between samples. The peaks can be given different weights in the comparison of profiles using appropriate classification methods and we show that randomForest classification using these ratios have a high and reproducible performance in comparison with other methods. The performance of this method is not affected by noise, transformation or normalization and should be considered for future comparison of chromatographic profiles in general.

AB - Illicit cocaine seizures are often compared to each other by using gas chromatography-mass spectrometry (GC-MS) data from cocaine alkaloid compounds to determine whether two specimens originate from the same production batch or not. This can provide intelligence or investigative information at the early stages of an investigation or evidence in court. Traditional classification methods assume high stability of all alkaloids, use all of them to calculate the correlation between two profiles and use a threshold to classify samples. Unstable alkaloids will have a strong influence on the performance. We show that comparing each alkaloid target compound individually improves the classification. Unfortunately, it requires normalization and is also sensitive to the stability. Instead we suggest to use ratios of all possible pairwise combinations of the GC-MS peaks. These ratios are scale free and directly comparable between samples. The peaks can be given different weights in the comparison of profiles using appropriate classification methods and we show that randomForest classification using these ratios have a high and reproducible performance in comparison with other methods. The performance of this method is not affected by noise, transformation or normalization and should be considered for future comparison of chromatographic profiles in general.

KW - Journal Article

UR - http://www.scopus.com/inward/record.url?scp=85029511565&partnerID=8YFLogxK

U2 - 10.1038/s41598-017-12042-x

DO - 10.1038/s41598-017-12042-x

M3 - Journal article

C2 - 28912606

SN - 2045-2322

VL - 7

SP - 11646

JO - Scientific Reports

JF - Scientific Reports

IS - 1

M1 - 11646

ER -