Professional Documents
Culture Documents
Rapid screening of complex matrices: Utilizing Kendrick mass defect to enhance knowledge-based
group type evaluation of GC×GC-HR(EI)ToFMS data.
Table of Contents
EXPERIMENTAL SECTION
114 compounds standard mixture S-2-5
Initial data processing parameters S-6
RESULTS AND DISCUSSION
Criteria for developed and evaluated classes S-6-7
Suggestions for additional classes S-7
FIGURES
Mass spectral similarity: Naphtenes and alkenes S-8
TPR vs FPR S-9
ROC discrimination Diesel Kerosene S-9
ROC weathered Diesel S-10
PCA weathered Kerosene S-10
ROC weathered Kerosene S-11
Approach incl. RT information S-12
S-1
EXPERIMENTAL SECTION
114 compounds standard mixture:
1-Ethyl-2-
N-Heptane 142-82-5 9% 611-14-3
methylbenzene
1-Ethyl-3-
N-Hexane 110-54-3 9% 620-14-4
methylbenzene
1-Ethyl-4-
N-Nonane 111-84-2 9% 622-96-8
methylbenzene
100µL for mix
Isopropylbenzene
N-Octane 111-65-9 9% 98-82-8
(cumene)
2,000 μg/mL
N-Pentane 109-66-0 9% Toluene
1,2,3-
N-Tetradecane 629-59-4 9% 526-73-8
Trimethylbenzene
1,2,4-
N-Tridecane 629-50-5 9% 95-63-6
Trimethylbenzene
1,3,5-
N-Undecane 1120-21-4 9% 108-67-8
Trimethylbenzene
m-Xylene 108-38-3
DHA
C6:0 Methyl caproate 106-70-7 4%
NAPHTHENES
S-2
C10:0 Methyl
110-42-9 4% Cyclohexane 110-82-7 4.7%
decanoate
C12:0 Methyl
111-82-0 4% Cyclopentane, ethyl- 1640-89-7 7.8%
dodecanoate
1,1,4-
C15:1 (cis-10) Methyl
90176-52-6 2% Trimethylcyclohexan 2234-75-5 2.9%
pentadecenoate
e
Cyclohexane,
C18:0 Methyl stearate 112-61-8 4% 1678-98-4 3.7%
isobutyl-
C18:2 (all-trans-9,12)
2566-97-4 2% DHA Olefines*
Methyl linolelaidate
C18:2 (all-cis-9,12)
112-63-0 2% Name CAS w/w% Used
Methyl linoleate
C18:3 (all-cis-6,9,12)
16326-32-2 2% 1-Heptene 592-76-7 6.79%
Methyl linolenate
100µL for mix
C18:3 (all-cis-
9,12,15) Methyl 301-00-8 2% (Z)-3-Heptene 7642-10-6 1.02%
linolenate
C20:0 Methyl
1120-28-1 4% 1-Octene 111-66-0 11.34%
arachidate
S-3
C20:1 (cis-11) Methyl
9/2/2390 2% 2-Octene, (E)- 13389-42-9 2.32%
eicosenoate
C20:2 (all-cis-11,14,)
2463-02-7 2% 2-Octene, (Z)- 7642-04-8 4.37%
Methyl eicosadienoate
C20:3 (all-cis-
8,11,14) Methyl 21061-10-9 2% 1-Nonene 124-11-8 6.91%
eicosatrienoate
C20:3 (all-cis-
11,14,17) Methyl 55682-88-7 2% 3-Nonene 20063-92-7 3.03%
eicosatrienoate
C20:4 (all-cis-
5,8,11,14) Methyl 2566-89-4 2% 2-Nonene 6434-78-2 2.28%
arachidonate
C20:5 (all-cis-
5,8,11,14,17) Methyl 2734-47-6 2% 2-Methyl-1-nonene 2980-71-4 3.41%
eicosapentaenoate
C21:0 Methyl
6064-90-0 2% 1-Decene 872-05-9 2.27%
heneicosanoate
C22:0 Methyl
929-77-1 4%
behenate
C22:2 (all-cis-13,16)
Methyl 61012-47-3 2% Name CAS Conc. Used
docosadienoate
C22:6 (all-cis-
4,7,10,13,16,19)
2566-90-7 2% Naphthalene 91-20-3
Methyl
docosahexaenoate
C23:0 Methyl
2433-97-8 2% Acenaphthylene 209-96-8
tricosanoate
C24:0 Methyl
2442-49-1 4% Acenaphthene 83-32-9
lignocerate
Phenanthrene 85-01-8
S-4
1-Heptanol 111-70-6 Benzo[b]fluoranthene 205-99-2
1-Nonanol 143-08-8
1-Decanol 112-30-1
1-Dodecanol 112-53-8
Aldehydes
Pentanal 110-62-3
Hexanal 66-25-1
Heptanal 111-71-7
100µL for mix
Octanal 124-13-0
2,000 μg/mL
Nonanal 124-19-6
Decanal 112-31-2
Undecanal 112-44-7
Dodecanal 50-28-2
Table S 1: Compounds used in the" 114 compounds standard mixture". Each compound class was either purchased as
prepared mix directly from the manufacturer or prepared individually. 100µL of each mix was used for the total mix.
Compound classes marked with * contained more compounds, which could not be detected with the presented GC×GC-
HRTOFMS method.
S-5
Initial data processing parameters
occurrence of 3 or
more of this KMD
will be excluded
S-6
Naphthenes Ratio 55.05 and
83.06 > 1
CH+ 0.0072 2 or more O+ (alcohol) 0.0168
CH2+ 0.0005 1 or more CHO2 (ester) 0.0530
R1 = 56.06 or 69.06
or 84.09
Table S 3: Kendrick Mass Defects and selection criteria for the developed and evaluated classifiers. The Fragments where
selected based on the fragmentation patterns for the respective classes as described in McLafferty, Interpretation of Mass
Spectra
S-7
MASS SPECTRA SIMILARITY NAPHTHENES AND ALKENES
Figure S1: Illustration of Spectral Similarity between Naphthenes and Alkenes. Spectra reprinted with permission from
https://sdbs.db.aist.go.jp (National Institute of Advanced Industrial Science and Technology, 06/03/2019)
S-8
TPR vs FPR
Figure S2: Performance evaluation of the developed classifiers in the receiver ROC space. The dotted red line represents a
random guess of the chemical compound class. The upper left corner (X/Y=0/1) represents a perfect classification
Figure S3: ROC Curves for the predictions of Diesel and Kerosene Samples based on the summed area values generated by
the GC×GC-HRTOFMS classification approach. AUC values of 1 indicate perfect distinction in-between the two different
fuel types. K-Nearest Neighbor clustering with 5fold cross validation and previous PCA that explains 85% variance.
S-9
Prediction: unweathered Prediction: 25% weathered
1 1
0.8 0.8
True positive rate
0.4 0.4
0.2 0.2
0 0
0 0.5 1 0 0.5 1
False positive rate False positive rate
0.8 0.8
True positive rate
0.6 0.6
AUC = 0.77778 AUC = 1
0.4 0.4
0.2 0.2
0 0
0 0.5 1 0 0.5 1
False positive rate False positive rate
Figure S4: ROC Curves for the predictions of the different weathering grades of Diesel based on the summed Area values
for the chemical compound classes generated by the GC×GC-HRTOFMS classification approach. Unweathered and Diesel
weathered 75% can be easily differentiated from the bulk as indicated by an AUC value of 1. The differentiation of 25% and
50% weathering grade is more difficult and less accurate (AUC values of 0.78).
1
unweathered
0.4
0.2
PC2 (29.5%)
-0.2
-0.4
-0.6
-0.8
-1
Figure S5: Results from PCA directly applied to the area per peak values for the weathered Kerosene experiments.
S-10
Prediction: unweathered Prediction: 25% weathered
1 1
0.8 0.8
True positive rate
0.4 0.4
0.2 0.2
0 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
False positive rate False positive rate
0.8 0.8
True positive rate
0.4 0.4
0.2 0.2
0 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
False positive rate False positive rate
Figure S6: ROC Curves for the predictions of the different weathering grades of Kerosene based on the summed Area values
for the chemical compound classes generated by the GC×GC-HRTOFMS classification approach. Unweathered and
Kerosene weathered 75% can be easily differentiated from the bulk as indicated by an AUC value of 1. The differentiation of
inweathered and 25% weathered kersosene however is not as distinguished as compared to the diesel weathereing studies.
S-11
Pseudochromatographic Representation
6
5
Second Dimension Retention Time [sec]
Peaks
1 linear alkanes
Branched alkanes
elution profile curve linear alkanes (fit)
0
0 1000 2000 3000 4000 5000 6000 7000
First Dimension Retention Time [sec]
Figure S7: Utilizing post-retention information for further discrimination of the classifier’s performance. Here, the
retetntion pattern of the linear alkanes (red) is predicted (blue). Everything compound eluting in the second dimension with
a retention time below the blue prediction is assigned to the “branched” alkane class. Since the prediction of the linear
alkanes is based on empirical measurements (such as e.g. standards etc.) this particular classifier provides good results
expect for very volatile compounds (in the beginning of the run) due to similar elution times in the second dimension.
S-12