LC-MS ESI Parameter Optimization with Bayesian Optimization for … · 2020-06-01 · LC-MS ESI...

1
Based on the response surfaces of several compounds, we changed default parameters in LCMS-8060NX. We demonstrated that new default parameters improved intensity about 2.6 times on average with various fields of compounds (Fig. 5). 4. Conclusion We’ve developed efficient ESI parameter optimization method. In the process of this study, we found an ESI parameter set which enable high signal intensity with various fields of compounds. Based on this study, We set this new ESI parameter set as default parameters in LCMS-8060NX. By LCMS-8060NX, users can obtain good measurement result without making complicated analytic condition search, and decrease their workload. References [1] Snoek et al., Practical Bayesian Optimization of Machine Learning Algorithms, Proc. NIPS, Vol. 2, 2012 [2] Tagawa et al., LC-MS Interface Parameter Optimization for High Sensitivity Measurement, Shimadzu Review, Vol. 75, 2018 Prominence is a trademark of Shimadzu Corporation. LC-MS ESI Parameter Optimization with Bayesian Optimization for High Sensitivity Measurement Yusuke Tagawa 1 , Yuki Ishikawa 1 , Levi Mikael 1 , Wataru Fukui 1 , Jun Watanabe 1 , Hisanori Morita 1 1 Shimadzu Corporation, Kyoto, Japan WP 279 1. Introduction Electrospray Ionization (ESI) is one of the methods to ionize compounds in mass spectrometry (MS) (Fig.1). To realize high sensitivity, ESI parameters are optimized by measuring actual samples a few dozen times. It is important to reduce the number of measurements during optimization to save the time and the sample. In this study, we have analyzed sensitivity behavior for ESI parameters, and developed efficient optimization method. 2. Methods Optimization Procedure We have developed two step optimization procedure (Fig. 2). At the 1st step, three temperature parameters are combined to one parameter set, and the temperature parameter sets are optimized. At the 2nd step, other four parameters are optimized by Bayesian optimization (BO)[1]. To apply BO to ESI parameter optimization, we have developed noise-robust BO method in previous study[2]. This two step optimization procedure makes the latency time for temperature stabilization short, and enables to find the high sensitivity condition at a small number of measurement. Figure 1 ESI parameters of LCMS-8060NX Interface(I/F) voltage Nebulizer gas flow Heating gas flow Drying gas flow ESI heater(ESIH) temperature Block Heater(BH) temperature Desolvation Line(DL) temperature Hydrophillic compounds 1.0 1.5 2.0 2.5 3.0 3.5 4.0 min 0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 (x10 7 ) 1.5 2.0 2.5 3.0 3.5 4.0 4.5 min 0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 (x10 5 ) Proline Uric acid Negative m/z 167.10>123.95 S:×2.4 S:×2.8 AEA EPA 17.0 17.5 18.0 18.5 19.0 19.5 min 0.0 0.5 1.0 1.5 2.0 (x10 6 ) 19.0 19.5 20.0 20.5 min 0.0 0.5 1.0 1.5 2.0 2.5 (x10 4 ) Lipids Positive m/z 348.20>62.10 S:×2.6 Negative m/z 301.20>257.20 S:×2.7 Pesticides 3.0 3.5 4.0 min 0.0 1.0 2.0 3.0 4.0 5.0 6.0 7.0 8.0 (x10 5 ) 3.5 4.0 4.5 min 0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 (x10 4 ) Positive m/z 255.00>208.90 Negative m/z 199.25>127.10 Fluroxypyr Cloprop S:×3.2 S:×2.2 Parameters Experimental Data Collection To understand the true response surface for ESI parameters, we collected LC-MS data with 2,088 combinatorial conditions. Because full factorial design of 7 parameters is difficult to conduct, we conducted two patterns of combinatorial experiment (Table 1). Total 4 compounds (Angiotensin II, Atrazine, Dicamba and Flurbiprofen) are measured. In this experiment, Shimadzu HPLC (Prominence™) and LC-MS (LCMS-8060, of which ESI structure is very similar to LCMS-8060NX) are used. 3. Results Evaluation of Optimization Performance We evaluated two optimization performances, our method and one factor at a time(OFAT) method. To evenly compare these two optimization performances, we used same experimental data collected as mentioned before as a ground truth. As a result, our method could search for the high sensitivity condition about half number of the measurement compared to OFAT method (Fig. 3). In 4 compounds, our method was 49% efficient on average. Investigation of Appropriate Default Parameter We built response surface models for ESI parameters from the collected experimental data (Fig. 4). Condition 1 Condition 2 I/F vol. [kV] 5 levels: 0.2,1.5,3.0, 4.0, 5.0 Fixed to optimal of condition 1 gas flow [L/min] Neb. 4 levels: 0.5,1.5,2.5, 3.0 Heat. 5 levels: 3,5,7,10,15 <20 Dry. 5 levels: 3,5,7,10,15 tempe rature [] ESIH 5 sets: (IF, DL, BH) (100,100,100), (170,150,200), (240,200,300), (300,250,400), (400,300,500) 5 levels100,170,240,300,400 DL 5 levels100,150,200,250,300 BH 5 levels100,200,300,400,500 Table 1 Data collection conditions Figure 3 Evaluation of Optimization Performance (Flurbiprofen) 1. ESIH, DL, HB temperature set Optimization default parameters Optimized temperatures 2. Heating Gas 3. Nebulizer Gas 4. Drying Gas 5. I/F Voltage Bayesian Opt. Opt high Intensity Optimized parameters SingleBays e_Atrazine_ No1_iter19 Opt x 2 x 3 x 4 x 5 Figure 2 Optimization Procedure Low Inteinsity x 1 (temperature set) 0 0.2 0.4 0.6 0.8 1 0 5 10 15 20 25 30 Relative Intensity number of measurement Proposed method OFAT Three temperatures Voltage Three gas flows Temperature set Voltage and three gas flows Same Intensity at 14 steps Figure 4 Examples of constructed response surfaces (Angiotensin) (ESIH,DL,BH) = (100,100,100) (240,200,300) (400,300,500) I/F Vol.(1~5) Neb. Gas(1~3) Dry. Gas(3~17) Heat. Gas(3~17) high Low Inteinsity Figure 5 Analyses of various samples with new default ESI parameters New Conventional Positive m/z 116.10>70.15

Transcript of LC-MS ESI Parameter Optimization with Bayesian Optimization for … · 2020-06-01 · LC-MS ESI...

Page 1: LC-MS ESI Parameter Optimization with Bayesian Optimization for … · 2020-06-01 · LC-MS ESI Parameter Optimization with Bayesian Optimization for High Sensitivity Measurement

Based on the response surfaces of several compounds, we

changed default parameters in LCMS-8060NX. We

demonstrated that new default parameters improved

intensity about 2.6 times on average with various fields of

compounds (Fig. 5).

4. ConclusionWe’ve developed efficient ESI parameter optimization

method. In the process of this study, we found an ESI

parameter set which enable high signal intensity with various

fields of compounds.

Based on this study, We set this new ESI parameter set as

default parameters in LCMS-8060NX. By LCMS-8060NX,

users can obtain good measurement result without making

complicated analytic condition search, and decrease their

workload.

References

[1] Snoek et al., Practical Bayesian Optimization of Machine Learning

Algorithms, Proc. NIPS, Vol. 2, 2012

[2] Tagawa et al., LC-MS Interface Parameter Optimization for High

Sensitivity Measurement, Shimadzu Review, Vol. 75, 2018

Prominence is a trademark of Shimadzu Corporation.

LC-MS ESI Parameter Optimization with Bayesian Optimization for High Sensitivity MeasurementYusuke Tagawa1, Yuki Ishikawa1, Levi Mikael1, Wataru Fukui1, Jun Watanabe1, Hisanori Morita1

1 Shimadzu Corporation, Kyoto, Japan

WP 279

1. IntroductionElectrospray Ionization (ESI) is one of the methods to

ionize compounds in mass spectrometry (MS) (Fig.1). To

realize high sensitivity, ESI parameters are optimized by

measuring actual samples a few dozen times. It is important

to reduce the number of measurements during optimization

to save the time and the sample. In this study, we have

analyzed sensitivity behavior for ESI parameters, and

developed efficient optimization method.

2. Methods〇 Optimization Procedure

We have developed two step optimization procedure (Fig.

2). At the 1st step, three temperature parameters are

combined to one parameter set, and the temperature

parameter sets are optimized. At the 2nd step, other four

parameters are optimized by Bayesian optimization (BO)[1].

To apply BO to ESI parameter optimization, we have

developed noise-robust BO method in previous study[2].

This two step optimization procedure makes the latency time

for temperature stabilization short, and enables to find the

high sensitivity condition at a small number of measurement.

Figure 1 ESI parameters of LCMS-8060NX

Interface(I/F) voltage

Nebulizer gas flow

Heating gas flow

Drying gas flow

ESI heater(ESIH)

temperature

Block Heater(BH)

temperature

Desolvation Line(DL)

temperature

Hydrophillic compounds

1.0 1.5 2.0 2.5 3.0 3.5 4.0 min

0.0

0.2

0.4

0.6

0.8

1.0

1.2

1.4

(x107)

1.5 2.0 2.5 3.0 3.5 4.0 4.5 min

0.0

0.5

1.0

1.5

2.0

2.5

3.0

3.5

(x105)

Proline

Uric acid

Negativem/z 167.10>123.95

S:×2.4

S:×2.8

AEA

EPA

17.0 17.5 18.0 18.5 19.0 19.5 min

0.0

0.5

1.0

1.5

2.0(x106)

19.0 19.5 20.0 20.5 min

0.0

0.5

1.0

1.5

2.0

2.5

(x104)

Lipids

Positivem/z 348.20>62.10

S:×2.6

Negativem/z 301.20>257.20

S:×2.7

Pesticides

3.0 3.5 4.0 min

0.0

1.0

2.0

3.0

4.0

5.0

6.0

7.0

8.0

(x105)

3.5 4.0 4.5 min

0.0

0.5

1.0

1.5

2.0

2.5

3.0

3.5

4.0(x104)

Positivem/z 255.00>208.90

Negativem/z 199.25>127.10

Fluroxypyr

Cloprop

S:×3.2

S:×2.2

Parameters

〇 Experimental Data Collection

To understand the true response surface for ESI

parameters, we collected LC-MS data with 2,088

combinatorial conditions. Because full factorial design of 7

parameters is difficult to conduct, we conducted two patterns

of combinatorial experiment (Table 1). Total 4 compounds

(Angiotensin II, Atrazine, Dicamba and Flurbiprofen) are

measured. In this experiment, Shimadzu HPLC

(Prominence™) and LC-MS (LCMS-8060, of which ESI

structure is very similar to LCMS-8060NX) are used.

3. Results〇 Evaluation of Optimization Performance

We evaluated two optimization performances, our method

and one factor at a time(OFAT) method. To evenly compare

these two optimization performances, we used same

experimental data collected as mentioned before as a

ground truth. As a result, our method could search for the

high sensitivity condition about half number of the

measurement compared to OFAT method (Fig. 3). In 4

compounds, our method was 49% efficient on average.

〇 Investigation of Appropriate Default Parameter

We built response surface models for ESI parameters from

the collected experimental data (Fig. 4).

Condition 1 Condition 2

I/F vol. [kV]5 levels:

0.2,1.5,3.0, 4.0, 5.0

Fixed to

optimal of

condition 1

gas

flow

[L/min]

Neb.4 levels:

0.5,1.5,2.5, 3.0

Heat.5 levels:

3,5,7,10,15<20

Dry.5 levels:

3,5,7,10,15

tempe

rature

[℃]

ESIH 5 sets: (IF, DL, BH)

(100,100,100),

(170,150,200),

(240,200,300),

(300,250,400),

(400,300,500)

5 levels:100,170,240,300,400

DL5 levels:100,150,200,250,300

BH5 levels:100,200,300,400,500

Table 1 Data collection conditionsFigure 3 Evaluation of Optimization Performance (Flurbiprofen)

1. ESIH, DL, HB

temperature set

Optimization

default

parametersOptimized

temperatures2. Heating Gas

3. Nebulizer Gas

4. Drying Gas

5. I/F Voltage

Bayesian Opt.

Opthigh

Inte

nsity

Optimized

parameters

SingleBays

e_Atrazine_

No1_iter19

Opt

x2

x3

x4

x5

Figure 2 Optimization Procedure

Low

Inteinsity

x1(temperature set)

0

0.2

0.4

0.6

0.8

1

0 5 10 15 20 25 30

Rel

ati

ve In

ten

sity

number of measurement

Proposed methodOFAT

Three temperatures

Voltage

Three gas flows

Temperature set Voltage and three gas flows

Same Intensity at 14 steps

Figure 4 Examples of constructed response surfaces (Angiotensin)

(ESIH,DL,BH) = (100,100,100) (240,200,300) (400,300,500)

I/F Vol.(1~5)N

eb. G

as(1

~3)

Dry. Gas(3~17)

Hea

t. G

as(3

~17

)

high

Low

Inteinsity

Figure 5 Analyses of various samples with new default ESI parameters

NewConventional

Positivem/z 116.10>70.15