Chromatographic Data Segmentation Method: a Hybrid Analytical

Subscriber access provided by Gothenburg University Library

Article

Chromatographic Data Segmentation Method: a Hybrid Analytical Approach for Investigation of Antiviral Substances in Medicinal Plant Extracts Tomas Drevinskas, Audrius Maruska, Laimutis Telksnys, Stellan Hjerten, Manatas Stankevi#ius, Raimundas Lelešius, R#ta Mickien#, Agneta Karpovaite, Algirdas Salomskas, Nicola Tiso, and Ona Ragazinskiene Anal. Chem., Just Accepted Manuscript • DOI: 10.1021/acs.analchem.8b04595 • Publication Date (Web): 29 Nov 2018 Downloaded from http://pubs.acs.org on December 1, 2018

Just Accepted “Just Accepted” manuscripts have been peer-reviewed and accepted for publication. They are posted online prior to technical editing, formatting for publication and author proofing. The American Chemical Society provides “Just Accepted” as a service to the research community to expedite the dissemination of scientific material as soon as possible after acceptance. “Just Accepted” manuscripts appear in full in PDF format accompanied by an HTML abstract. “Just Accepted” manuscripts have been fully peer reviewed, but should not be considered the official version of record. They are citable by the Digital Object Identifier (DOI®). “Just Accepted” is an optional service offered to authors. Therefore, the “Just Accepted” Web site may not include all articles that will be published in the journal. After a manuscript is technically edited and formatted, it will be removed from the “Just Accepted” Web site and published as an ASAP article. Note that technical editing may introduce minor changes to the manuscript text and/or graphics which could affect content, and all legal disclaimers and ethical guidelines that apply to the journal pertain. ACS cannot be held responsible for errors or consequences arising from the use of information contained in these “Just Accepted” manuscripts.

is published by the American Chemical Society. 1155 Sixteenth Street N.W., Washington, DC 20036 Published by American Chemical Society. Copyright © American Chemical Society. However, no copyright claim is made to original U.S. Government works, or works produced by employees of any Commonwealth realm Crown government in the course of their duties.

Page 1 of 19 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Analytical Chemistry

Chromatographic Data Segmentation Method: a Hybrid Analytical Approach for Investigation of Antiviral Substances in Medicinal Plant Extracts

Tomas Drevinskas1,2, Audrius Maruška1*, Laimutis Telksnys2,3, Stellan Hjerten4, Mantas Stankevičius1, Raimundas Lelešius5,6, Rūta Mickienė1, Agneta Karpovaitė5, Algirdas Šalomskas5, Nicola Tiso1, Ona Ragažinskienė7 1Instrumental

Analysis Open Access Centre, Faculty of Natural Sciences, Vytautas Magnus University, Vileikos str. 8, LT - 44404, Kaunas, Lithuania 2Department

of Systems’ Analysis, Faculty of Informatics, Vytautas Magnus University, Vileikos 8, LT44404 Kaunas, Lithuania 3Institute

of Data Science and Digital Technologies, Vilnius University, Goštauto 12, LT-01108 Vilnius, Lithuania

4Department

of Chemistry - BMC, Biochemistry, Uppsala University, Husargatan 3, 752 37 Uppsala, Sweden

5Department

of Veterinary Pathobiology, Veterinary Academy of Lithuanian University of Health Science Tilžės str.18, LT - 47181, Kaunas, Lithuania 6Institute

of Microbiology and Virology, Veterinary Academy of Lithuanian University of Health Science Tilžės str.18, LT - 47181, Kaunas, Lithuania 7Sector

of Medicinal Plants, Kaunas Botanical Garden of Vytautas Magnus University, Z. E. Žilibero str. 6, LT 46324 Kaunas, Lithuania

Keywords: Gas chromatography-mass spectrometry, Essential oils, Segmentation tree, Chemometrics, Antivirals

Correspondence to: Prof. Audrius Maruška [email protected]

ACS Paragon Plus Environment

Analytical Chemistry 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 2 of 19

Abstract

The methodology described in this paper will significantly reduce the time required for understanding the relations between chromatographic data and bio-activity assays. The methodology is a hybrid of hypothesis based and datadriven scientific approaches. In this work, a novel chromatographic data segmentation method is proposed, which demonstrates the capability of finding what volatile substances are responsible for antiviral and cytotoxic effects in the medicinal plant extracts. Up to now the full potential of the separation methods has not been exploited in the life sciences. This was due to the lack of data ordering methods capable to adequately preparing the chromatographic information. Furthermore, the data analysis methods suffer of multidimensionality requiring a large number of investigated data points. A new method is described for processing any chromatographic information into a vector. The obtained vectors of highly complex and different origin samples can be compared mathematically. The proposed method, efficient with relatively small sized datasets, does not suffer of multidimensionality. In this novel analytical approach, the samples did not need fractionation and purification, what is typically used in hypothesis based scientific research. All investigations were performed using crude extracts possessing hundreds of phyto-substances. The antiviral properties of medicinal plant extracts were investigated using gas chromatography-mass spectrometry, antiviral tests and proposed data analysis method. The findings suggested that: (i) beta-cis-caryophyllene, linalool, eucalyptol possess antiviral activity, while (ii) thujone does not, and (iii) alpha-, beta-thujones, cis-p-menthan-3-one, estragole show cytotoxic effects.


Page 3 of 19 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


Introduction Analytical research is performed using hypothesis-based (holistic), or data-driven (reductionistic) approaches.1 In the era of big-data and high performance computing, data-driven scientific approach has become extremely important, yet methodologies providing highly multidimensional data with a small sized dataset are only used with hypothesis-based scientific approaches.2 In this paper a methodology that uses a hybrid scientific approach (a combination of hypothesis-based and data-driven) is described and it will help speeding up whole analytical process. Current data analysis methods are not capable to clarify the required attributes even if sophisticated separation and molecular biology methods exist.3 Separation methods are undervalued as they are only used for the identification and quantification of the compounds of interest, except for the applications, where they are used together with sophisticated data analysis methods.4,5 The separation methods should be treated as highly multidimensional information providing analytical techniques.5 They not only separate compounds, provide quantificational data, but also contain information on physico-chemical features: (i) volatility, (ii) polarity, (iii) charge, (iv) molecular weight, etc.6,7 Several groups of analytical separation methods exist and mainly electrophoretic and chromatographic separations are performed in life-sciences investigations. Gas chromatography (GC) in combination with mass spectrometry (MS) is a powerful technique capable of separating hundreds of substances in a complex mixture. Even though the GC-MS is advantageous over other separation methods, it still suffers main separation method issue – the shifting of the retention time makes the identification more difficult. Another group of analytical methods is bioactivity assays. Bioactivity assays are outstanding in hypothesis-based and data-driven scientific investigations, yet they provide simple 1-dimensional information and these methods should be treated as the reference revealing if the sample is positive in respect of the investigated attribute. On the contrary, the bioactivity assays do not reveal any detailed information on which features of the sample make it bioactive. The input of bioactivity assays investigating antivirals is undoubted. The most effective antivirals have the highest ratio (selectivity index (SI)) between half maximum effective concentration (EC50) and half-maximum cytotoxic concentration (CC50). Consequently, the most effective antivirals possess low cytotoxicity, and high antiviral activity. At the beginning, the cytotoxicity tests are performed to determine CC50 of potential antivirals. Later CC50 and lower concentrations are used for the treatment of virus under different conditions to calculate EC50. Some preparations can act as virucides, inactivating the viruses, and some can act as antivirals, inhibiting the replication of the virus. Medicinal plant extracts can exhibit antiviral activity during different stages of virus replication by different mechanisms.8,9 Therefore, different methods are used for the determination of inhibition of virus replication and efficacy of antivirals. The most important index is the absence of virus or significant decrease in its quantity (titre) after the treatment. For this purpose, various methods of virus quantification can be chosen. The quantity of virus can be measured after titration by means of the laboratory methods for detection of virus (virus titration, plaque assay), virus antigen (ELISA), or virus nucleic acid detection (real-time PCR). The generated data using separation methods and bioactivity assays require more sophisticated data analysis than current tools can provide. Mainly statistical, machine learning and artificial intelligence methods are used. In some cases, even sophisticated data analysis methods are not capable of solving real tasks. Real samples can hardly be compared mathematically if dimensions do not perfectly match. For instance, in voice recognition theory, dynamic programming methods are used that process the signals of uneven lengths, containing different positions of markers and sampling frequency.10 In separation science similar problems exist: the peaks and retention time can be shifted due to fluctuating conditions. Dynamic programming approach allows adjustment of important markers, so that information of different length signals can be compared mathematically10. For signal improvement also ACS Paragon Plus Environment


Page 4 of 19

adaptive algorithms can be applied.11,12 Furthermore, current data analysis methods suffer of multidimensionality and should be applied carefully for the separation technique derived data.5,13 Currently, artificial intelligence methods are applied for bio-investigations, drug discovery, etc.14–16 Such methods are mainly unsupervised, learning by itself. Unfortunately, unsupervised methods require very large sized data sets, what is not possible using separation methods.17 Being classified as a hybrid scientific approach, this work is a junction between separation science, bioactivity assays and data analysis methods that explain the bioactivities of interest and associate them with the peaks and substances in the complex mixtures. This work describes chromatographic data ordering and processing method, which produces vectors from chromatographic information. The vectors are of even lengths and easily comparable between different samples. The method is applicable for bio-activity feature investigation. The aim of this work was to develop a novel chromatographic data segmentation method applicable for the clarification of factors that are responsible for certain bio-activities in highly complex mixtures.

Experimental Section Research algorithm The research comprised of 6 main parts: (i) preparation of the extracts, (ii) chemical analysis (hypothesis-based approach), (iii) reference generation – bioactivity assays (hypothesis-based approach), (iv) generation of segmentation frame (data-driven approach), (v) generation of peak trees (data-driven approach) and (vi) calculation of the correlation between the reference data and the peaks in the segments (hybrid approach) (Supplementary Figure S-1). In the sample preparation part, the extracts were prepared for further investigations. In the chemical analysis part, the gas chromatographic-mass spectrometric analysis of plant extracts was performed, chromatograms were integrated. During this stage chromatographic dataset – tables possessing peak retention time, area and identified substances were generated and used in further steps. In the bioactivity assays part, three different methods were used: (i) Assay for the determination of the antiviral effect, (ii) cytotoxicity and (iii) Viral yield inhibition. There tests provided numerical values (reference data) that corresponded to a certain extract. In the data analysis part the chromatographic dataset was used for the generation of (i) segmentation frame, (ii) segment trees and (iii) calculation of correlations between reference data and the generated segments. Finally, the compounds corresponding to the bioactivity tests were identified.

Extraction and sample preparation Investigating bioactivity, 16 plants rich in volatile compounds based on previous investigations were selected: Satureja montana L. (S. montana), Chamaemelum nobile L. (Ch. nobile), Perilla frutescens L. Britton. (P. frutescens), Agastache foeniculum (Pursh) Kuntze (A. foeniculum), Origanum vulgare L. (O. vulgare), Mentha piperita L. (M. piperita), Geranium macrorrhizum L. (G. macrorrhizum), Melissa officinalis L. (M. officinalis), Angelica archangelica L. (aerial part) (A. archangelica aerial part), Angelica archangelica L. (roots) (A. archangelica roots), Thymus vulgaris L. (T. vulgaris), Hyssopus officinalis L. (H. officinalis), Nepeta cataria L. (N. cataria), Echinacea purpurea L. Moench. (E. purpurea), Salvia officinalis L. (S. officinalis) and Desmodium canadense L. DC. (D. canadense).18,19 The plant material was ground and 0.5 g was extracted using 20 ml 40 %


Page 5 of 19 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


(v/v) ethanol (ethanol (95.0 % vol.) from (Sigma Aldrich)/ water (bidistilled using Fistreem Cyclon bidistillator (UK)) mixture.20 The extraction was performed in an orbital shaker (SIA Biosan, Latvia) for 24 hours at room temperature (ca. 20 °C).

Analytical Procedures Gas chromatograph-mass spectrometer (GC2010 and GCMS-QP2010) with an auto-injector (AOC-5000) Shimadzu (Japan) was used. A low polarity stationary phase capillary column RTX-5MS, length 30 m, ID 0.25mm, film thickness 0.25 µm (Restek, USA) was used. One microliter of 40 % (vol.) ethanol (EtOH) extract was injected into the system. The flow rate of helium gas in the column was 1.2 ml/min. The temperature gradient was programed: initial temperature was set to 60 oC and maintained for 3 minutes, raised to 150 oC, at 5 oC/min velocity; from 150 oC temperature was raised to 180 oC at 20 oC/min velocity and maintained for 3 minutes. Injection port temperature was set to 240oC, column interface and ion source temperature was set to 260 oC. Substances were ionized using the electron ionization method, ionization energy was 70eV. Mass range was scanned from 30 m/z to 400 m/z.20,21

Bioactivity assay procedures Cytotoxicity tests. CC50 of the extracts was determined on Vero cells (ATCC CCL-81, provided by Dr. I. Jacevičienė, National Food and Veterinary Risk Assesment Institute, Lithuania) using MTT assay.22 Assay for the determination of antiviral effect. For determination of antiviral properties, one-day-age Vero cells in a 96-well plate were used. Avian infectious bronchitis virus (IBV) (Beaudette strain, provided Dr. M. H. Verheije, Utrecht University, The Netherlands) was used at a multiplicity of infection of 0.05. Extracts were assessed for the ability to inhibit IBV replication using four mechanisms. Every sample was tested in quadruplicates. Controls of cells, virus, and extracts were included. In the first method (I method), the virus was treated with extract and then poured onto the cells. In the second method (II method), mixtures of virus and the extract were poured onto the cells. In the third method (III method), the cells were infected with the virus and then treated with extract. In the fourth method, the cells were treated with extract and then infected with the virus. After 72 h of incubation, the plates were microscopically examined using an inverted microscope (Leica, Germany) to detect the cytopathic effect (CPE). Based on the results of antiviral effect assay the plant extracts were chosen for the determination of EC50 and SI using I method. Extracts were titrated from 1 to 1:128 CC50 and used for virus (MOI 0.05) treatment. After 72 hours the MTT assay was done as was stated above. EC50 were calculated from the plot of percentages of cell viability against extract concentrations. Viral yield inhibition. Plaque reduction assay was done according to reference.23 Extracts (concentrations equivalent from 1 CC50 to 0.125 CC50) and 25.000 plaque forming units (PFU) of IBV were mixed and incubated at room temperature for one hour. Then confluent monolayer of Vero cells in 6-well plates was inoculated with mixtures prepared from 1 ml of IBV (MOI 0.05) and plant extracts. After one hour at 37 ºC in 5 % CO2 the mixtures were discarded and washed twice with PBS. The agarose 0.4 % in maintenance medium was added to cells, and the plates were stored at room temperature for 15 minutes and incubated at 37 ºC and 5 % CO2. After 72 hours the plates were microscopically examined for detection of CPE and then 200 μl MTT (5 mg/ml) was used for staining. Plaques were counted after incubation at 37 ºC (5 % CO2) for four hours. The number of plaques was expressed as log10 and the reduction rate was calculated.23 The virus yield reduction was evaluated by means of virus titration ACS Paragon Plus Environment


Page 6 of 19

and real time PCR. The cells were inoculated as described above in the plaque reduction assay section. Virus titre was measured by detection of CPE in 96-well plates. TCID50 of control virus and treated ones was evaluated after three days. Virus titres were calculated using Kärber method.24

Data analysis Mathematical data analysis was performed using Rstudio software.25 After chromatographic analysis the tables with peak retention times – RTp, peak areas – ap and identified compounds were generated for each medicinal plant extract (Supplementary Table S-13). This data was used for further investigation. The essence of the segmentation method is that full peak set of the sample is divided into smaller segments (Figure 1).

Figure 1. Explanation of the data segmentation method. (a) Representation of data segmentation procedure. (b) Example of segmented chromatographic data – dendrogram of S. officinalis extract

The first node (mean) that was used for division of a segment was the mean retention time of the whole dataset (RTμ). Then the first segment was divided into 2 smaller segments: the beginning of the first segment was the minimum retention time value min (RTp) and the end of the segment was the mean value of retention time of the peaks (RTμ). The beginning of the second segment was previous RTμ and the end of the second segment was maximum value of retention time values max(RTp). Each new segment was divided using the same procedure, where the mean, minimum and maximum of the segments were calculated individually (Figure 1 a). In this method, the first division into segments was considered as the first level, second division of segments was considered as the second level etc. (Figure 1 b). The procedure can be described using 3 algorithms (Figure 2). In the 1st algorithm, the segmentation frame is generated (Figure 2 a). The sequence of procedures is following: (i) a merged table of 16 sample tables is read and the number of segmentation levels (in this work 9) is set; (ii) The number of nodes (means) in a level is calculated according to following equation (1):

𝑛(𝑙) = 2𝑙 (1) ACS Paragon Plus Environment

Page 7 of 19 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


Where n(l) – is the number of nodes in a certain level – l. Each node corresponds to the mean of the segment – μs. (iii) knowing that each segment is divided into 2 smaller segments the number of the segments in a level is found following equation (2):

𝑠(𝑙) = 2 × 2𝑙 (2)

Where s(l) is the segment number and l – is the level. (iv) The mean (min) of a segment is calculated; (v) the minimum (min) of a segment is found; (vi) the maximum (min) of a segment is found; (vii) the segment is divided into 2 smaller segments. (viii) It is checked, if all segments were divided inside a level: if not then the segment is incremented and (iv) – (vii) procedures repeated; if all segments were divided then it is proceeded to next procedure; (ix) it is checked if all levels were generated: if not, the level is incremented and steps (ii) – (ix) are repeated; if yes the sequence is finished and the product is a generated frame.

Figure 2. The algorithms that describe chromatographic data segmentation procedure. (a) Frame generation algorithm, (b) chromatographic segmentation tree generation algorithm and (c) the algorithm of calculation of correlations between reference data and the segments

In the 2nd algorithm, the segmentation tree is generated (Figure 2 b). The sequence is following: (i) the individual sample peak set table is read and the number of samples must be provided; (ii) needed information related to the current segment is obtained from the generated frame; (iii) the peaks are assigned to the segment according to the frame information; (iv) the sum of assigned peak areas is calculated in a segment following equation (3):



Page 8 of 19

𝐼

𝐴𝑠(𝑠,𝑙) =

∑𝑎

𝑝𝑖

(𝑖 = 1,2,3…𝐼)(3)

𝑖=1

Where As(s,l) – sum of peak areas in a segment, s – segment number, l level number, api – peak area, I – peak i in a segment. (v) It is checked if all segments processed in a level and if not, the segment is incremented and procedures (iii) – (v) are repeated until all segments in a level are processed and then (vi) it is checked if all levels were processed and if not, the level is incremented and the procedures (ii) – (vi) are repeated until all levels were processed and then (vii) it is checked if all samples were processed and if not, the sample number is incremented and the procedures (i) – (vii) are repeated until all samples were processed and then the sequence is terminated. The result of this segmentation procedure is a multidimensional vector that describes processed chromatographic information. The method is capable of transforming any chromatographic peak table information to a vector and each vector produced from different chromatograms is comparable with each other. Each segmentation tree after processing has a defined amount of dimensions and the number is calculated following the equation (4):

𝐿

𝑑(𝑙) =

∑2 (𝑙 = 1,2,3…𝐿)(4) 𝑙

𝑙=1

Where d(l) is the number of dimensions, l – level. In the 3rd algorithm, the correlations between segments and reference data are calculated (Supplementary Table S-14) and significant peaks are identified (Figure 2 c). The sequence of procedures is following: (i) the correlation between reference data and corresponding segment of each sample is calculated; (ii) it is checked if all segments were processed and if not, the segment number is incremented and the procedures (i) – (ii) are repeated until all segments were processed. (iii) The highest correlation values and corresponding segments are obtained, it must be specified, how many segments to output. (iv) It is checked if all segments were obtained and if not, the previous segment is excluded and the procedures (iii) – (iv) are repeated until all segments of interest are obtained. (v) The peaks corresponding to the segments of interest are identified and obtained. (vi) It is checked if all segments were processed and if not, the segment is incremented and the procedures (v) – (vi) are repeated until all segments were processed, then the sequence is terminated. The R code and datasets are available at: https://[email protected]/TomasDr/cdstree.git .

Results and Discussion Chromatographic data segmentation Each compound was checked according to the retention time, including the presence/absence and the peak area in each sample and the correlations were calculated between reference data of bioactive properties and peak areas. The plant extracts possessed between 20 and 50 identifiable peaks. Proposed method avoids manual data ordering. Considering that the peak sets of the extracts were different, it was decided to generate the frame for segmentation fitting all the peak sets of the extracts. Main statistical parameters were calculated: (i) segment means, (ii) minimums and (iii) maximums (Supplementary Tables S-1 – S-3). Using the frame with generated main statistical ACS Paragon Plus Environment

Page 9 of 19 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


parameters it was checked if peak set contained data points in the required segment: in the presence of the peaks, the sum of peak areas and the number of peaks were calculated in the corresponding segment. Knowing the peak distribution in the segments, the information can be represented in 2 ways: (i) network graphs and (ii) dendrograms Figure 3.

Figure 3. Representation of chromatographic data segmentation trees. Network graphs (a, b and c), dendrograms (d, e and f). A. foeniculum (a and d), E. purpurea (b and e), D. canadense (c and f)

The clusters of peaks are distinguishable and at the end of the branch, the retention time represents the corresponding compound, which is identified using MS. The data trees were generated using statistical information of all separated extracts, therefore for the visual representation, the branches having a peak are drawn. In the data frame generated, the branches not possessing a peak are left for further data processing. Such technique allows comparison of completely different chromatographic peak sets. The method was intentionally developed, so that branching patterns, or specific branches in the dendrograms could be related to specific bioactivities using mathematical methods.

Determination of bioactivity properties in complex samples Viruscidicity, cytotoxicity and viral inhibition selectivity of the extracts were determined (Table 1 and Supplementary Table S-15).

Table 1. Bioactivity assays – reference data of the extracts Plant

CC50

EC50

SI



Page 10 of 19

S. montana 0.75 0.044 17 Ch. nobile 0.16 0.015 10.7 P. frutescens 0.77 0.07 11 A. foeniculum 0.06 0.011 5.5 O. vulgare 0.52 0.008 65 M. piperita 0.27 0.004 67.5 G. macrorrhizum 0.18 NA NA M. officinalis 0.59 0.015 39.3 A. archangelica 0.24 NA NA aerial part A. archangelica roots 0.33 NA NA T. vulgaris 0.63 0.01 63.1 H. officinalis 0.64 0.076 8.4 N. cataria 0.31 0.028 11.1 E. purpurea 0.49 0.045 10.9 S. officinalis 0.11 0.003 36.7 D. canadence 0.29 0.017 17.1 * – Antiviral test method (first, second and third method), NA – Not applicable

A. foeniculum extract possessed highest cytotoxic activity, P. frutescens and S. montana possessed lowest cytotoxic activity. S. officinalis showed highest antiviral activity, G. macrorrhizum, A. archangelica aerial part and A. archangelica roots extracts showed no antiviral activity. Highest selectivity index was observed for M. piperita extract and lowest SI was observed for A. foeniculum extract.

Segmentation data optimization and interpretation It was observed that after segmentation, some peaks fall into the same segment in certain extracts and some peaks fall into separate segments. For example, the RT of alpha-thujone in S.officinalis extract is 11.13 min, in D.canadense – 11.12 min, in S.montana 11.10 min. The shifting of retention time is observable using GC-MS. The retention time tolerance factor (RTtf), which is an absolute value (min) was added for expansion of the segment threshold (upper and lower). The segmentation frame was generated using 1st algorithm (Figure 2 a) and tested with different values of RTtf in the range of 0 min to 0.1 min (Figure 4 a). The segmentation trees were generated using 2nd algorithm (Figure 2 b) The reference data used for the method optimization was cytotoxicity information (CC50) (Table 1). The correlations were calculated following 3rd algorithm (Figure 2 c).


Page 11 of 19 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


Figure 4. Histogram of the distributions of calculated correlation coefficients for (a) cytotoxicity reference data when retention time tolerance factor is 0.0025 min and for selectivity indices reference data (b). (1) potentially cytotoxic compounds, (2) the compounds expected in non-cytotoxic extracts, (3) limonene, (4) linalool, (5) eucalyptol, (6) caryophyllene

It was observed that increasing RTtf from 0 to 0.0025 min the extreme correlation values increased. This was probably due to the fact, that lower number of significant peaks were left falling out of the segment due to RT error. Increasing the RTtf from 0.0025 to 0.1 min, it was observed that extreme correlation values became lower and were even transferred to next column in the histogram (Figure 4). It was decided to use RTtf of 0.0025 min, as it showed slightly increased extreme values of correlations: the absolute value of negative correlations increased from 0.474 to 0.488 and the absolute values of positive correlations increased from 0.503 to 0.523 and 0.497 to 0.504. Later, calculations were performed with RTtf 0.0025 min.

Identification of bioactive compounds The compounds that possess cytotoxic effect were investigated. For determination of cytotoxicity, the CC50 concentrations were used as reference data. Higher cytotoxicity extracts needed lower concentration for the CC50 concentration therefore the correlation coefficient was calculated between peak area in the segments and CC50 concentrations of the extracts. Highest negative correlations indicated segments containing mostly cytotoxic compounds (Table 2). In this case, GC-MS was used for analysis therefore each peak in the chromatograms was identified and cytotoxic compounds in the segments were also identified Table 2 (Supplementary Table S-4).

Table 2. Identification of the segments that contain cytotoxic substances Segment number Correlation coefficient Segment minimum (min)

805

989

48

211

-0.488

-0.474

-0.437

-0.423

12.524

21.132

10.884

13.525



Segment maximum (min) Plant extract

12.529 Area CC50 (*RAU×

1000)

S. montana 0.75 0 Ch. nobile 0.16 0 P. frutescens 0.77 0 A. foeniculum 0.06 1788 O. vulgare 0.52 0 M. piperita 0.27 0 G. macrorrhizum 0.18 2525 M. officinalis 0.59 0 A. archangelica 0.24 350 aerial part A. archangelica 0.33 0 root T. vulgaris 0.63 0 H. officinalis 0.64 0 N. cataria 0.31 976 E. purpurea 0.49 0 S. officinalis 0.11 0 D. canadense 0.29 0 Number of peaks of a segment in all samples *RAU - Relative area units

Page 12 of 19

21.134 Peaks in a segment 0 0 0 1 0 0 1 0

Area (RAU×

11.508 Area (RAU×

0 0 0 879 0 0 506 0

Peaks in a segment 0 0 0 1 0 0 1 0

1

0

0 0 0 1 0 0 0

13.554 Area (RAU×

6008 44292 99 4531 9856 11893 640 0


1579 0 0 121100 2392 0 0 0


0

1051

2

69938

1

0

0

0

0

0

0

0 0 0 0 0 0

0 0 0 0 0 0

1842 2810 0 530 67364 1291

1 1 0 1 2 1

3807 0 1407 6915 0 0

1 0 1 1 0 0

1000)

4

2

1000)

18

1000)

7

The proposed method has the key element of the dynamic programming – it works with the data that can shift the position (in this case – the retention time). Additionally, the segmentation tree method can work with multiple compounds at the same moment (as shown in Table 2, 48th segment). Some segments contain multiple peaks, like alpha-thujone and beta-thujone, what can be an indication for synergistic effect (up to now synergistic effects have only been investigated using hypothesis-based scientific approach), when two substances are mixed together and bio-activity is measured. A classical methodological approach can be used only with clarified factors: (i) identified and quantified analytes, (ii) determined enzymatic activity, (iii) clarified cytotoxic properties etc. This work is original in that way, that it does not require the identification of the separated compounds. As shown in the Table 2, 48th segment, the compounds showing significant correlations had different chemical structure (Supplementary Table S-4.), different class belonging substances, on the other hand, their volatility and polarity are very similar as their retention time is between 10.884 and 11.508 min. The method can be applied for the clarification of important peaks and later the peaks of importance can be identified following full analytical method validation. Such approach saves the time significantly required for the scientific process to be fulfilled and only the compounds of interest can be identified, calibrated and determined. Currently, the untargeted metabolomics investigating biomarkers require extremely high effort and time (validation of the analytical method for all investigated compounds is needed) in order to complete the research.26 Potentially cytotoxic compounds. Following compounds were identified in the extracts, they showed significant correlations with CC50 measure and all of them have been determined being cytotoxic in previous studies using ACS Paragon Plus Environment

Page 13 of 19 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


hypothesis-based scientific approach: cis-p-menthan-3-one (correlation -0.488) showed cytotoxic effect on rats,27,28 germacrene D (correlation -0.475),29 alpha- and beta-thujones (correlation -0.437) have been shown even causing seizures,30,31 extracts containing estragole (correlation -0.424) were found to be toxic,32,33 isopinocamphone and myrtenyl methyl ether (correlation -0.412) were described as neurotoxic, or even carcinogenic,34 mild cytotoxic effect was reported for limonene (correlation -0.411).35 The details of other potentially cytotoxic substances identified is represented in Supplementary Table S-4. Non-cytotoxic compounds are also important. The plant extracts that have these substances are likely to be less cytotoxic: 1-(2-furanyl)-4-methylpentanone and 2-butanoic acid, 2-methyl-,3-methylpentyl ester, (Z)- (0.610 correlation), 1-octen-3-ol and 1,7-octadien-3-ol (correlation 0.527), D-campholic acid and L-camphor (correlation 0.504), carvacrol (correlation 0.498). The details are represented in Supplementary Table S-5. Potentially antiviral compounds. Beta-cis-caryophyllene (correlation -0.458) provided significant correlation with EC50 measure in the extracts. Supplementary Table S-6. Following compounds are not antivirally active according to EC50 measure: alpha-thujone and chrysanthone (correlation 0.727), (+/-)-menthol (correlation 0.677), gamma-elemene (correlation 0.665), cis-beta-ocimene (correlation 0.656 cis-p-menthan-3-one (correlation 0.634). Supplementary Table S-7. Here, it is visible that cytotoxic compounds, such as thujones, do not have antiviral effect meaning that different substances are responsible for cytotoxic and antiviral effects. Selective antiviral compounds. The results of antiviral selectivity are represented in Figure 4 b. The following compounds showed high correlation with SI measure: beta-cis-caryophyllene, which antiviral activity was confirmed by multiple studies (correlation 0.800),36–39 linalool showed activity against adenoviruses (correlation 0.722),40 eucalyptol is a constituent of eucalyptus essential oil that was found possessing antiviral activity against herpes simplex virus type 1 (correlation 0.690).41 Supplementary Table S-8. It was found limonene showing significant negative correlation with the SI measure (correlation -0.476). This substance was identified and reported possessing mild cytotoxic activity.35 Supplementary Table S9. Antiviral substances according to the I method. beta-cis-caryophyllene was found highly correlative with the values of I method (correlation 0.672).36–39 Linalool (correlation 0.575), camphol and borneol and its derivatives were found possessing antiviral activity (correlation 0.569).38,42,43 Supplementary Table S-10. It was found that some cytotoxic substances did not have antiviral effect according to the I method. alpha-thujone and chrysanthone (correlation -0.584), p-menthan-3-one (correlation -0.533), gamma-elemene (correlation 0.524). Supplementary Table S-11. Antiviral substances according to the II method. Eucalyptol (correlation 0.684), linalool (correlation 0.636), Dcampholic acid and L-camphor (correlation 0.591). Supplementary Table S-12. No significant negative correlations have been identified between the reference data of the II method and the segments.

Discussion Chromatograms, provide highly multidimensional data.Existing modern data analysis methods require large sized datasets analyzing highly multidimensional data, what is used in data-driven scientific approach and is impossible for separation methods. Another disadvantage is that there are no data ordering methods for chromatographic information. Different samples analyzed provide different peaks and the peak-set in one sample differs from the peak-set in another sample. The peak-sets are incomparable except the cases, where different substances are preACS Paragon Plus Environment


Page 14 of 19

selected, what is typically done in untargeted metabolomics.26 The problem exist: separation methods are underused not reaching their full potential. The methodology clarifying antiviral properties in medicinal plant extracts using machine learning methods was published.20 The method identified the groups of compounds related to the antiviral activity in the plant extracts, yet the methodology was incapable of processing highly multidimensional gas-chromatographic data.20 The same methodology was applied for clarification of cytotoxic factors in medicinal plants44. More sophisticated data analysis method was needed for highly complex data of GC-MS. This work fills the gap of incapability of existing data analysis methods, processing and clarifying highly multidimensional separation data. The attempt to overcome stated problems has been performed.11,20,44 Electropherograms were segmented into 3 segments according to the migration velocity of the analytes. This technique identified relations between antiviral activity, cytotoxic activity and medium, low electro-migration velocity organic cations. The proposed technique was too crude for highly efficient GC-MS separations and no relations between volatile compounds and bioactivities were identified. The findings of this work can speed up scientific investigations significantly not requiring fractionation and purification of complex mixtures.Not all substances have to be identified and only those, which are of interest. This greatly reduces the amount of work to be done fulfilling the aim of the research. Novel data ordering and analysis method is used for clarification of multiple bio-activities in different complex medicinal plant extracts. This method can be performed on a benchtop computer and no supercomputing capabilities are needed as they are used in other sophisticated separation science and artificial intelligence combinations45. The determination of the important features was performed calculating correlation coefficients between the dimensions of the vector and reference data. The proposed method can be used with different calculation methods than correlation coefficients: (i) cosine similarity, (ii) principal component analysis, (iii) decision trees, (iv) neural networks etc. Proposed method was tested with GC-MS. It should be applicable for other separation methods such as high performance liquid chromatography, capillary electrophoresis, or other techniques coupled with different detection methods.

Conclusions To the authors knowledge this methodology is the first utilizing full potential of separation method, identifying bioactive factors in highly complex mixtures. The novel chromatographic data segmentation method is: (i) a hybrid analytical (hypothesis-based + data-driven) approach speeding up the scientific process, (ii) capable of identifying antiviral and cytotoxic substances in highly complex mixtures. This work suggests, that caryophyllene, eucalyptol, linalool and borneol are providing antiviral effect against avian infectious bronchitis virus. The approach proposed suggests, that alpha- and beta-thujones, cis-p-menthan-3-one, estragole, germacrene D are possessing cytotoxic effect in investigated medicinal plant extracts.

Supporting Information The supporting information is available free of charge on the ACS Publication website at DOI: . Supplementary Figure of the research algorithm, supplementary tables of generated data and additional statements discussing the findings of the method (Supporting Information for Publication 2018 11 26 Drevinskas et al). ACS Paragon Plus Environment

Page 15 of 19 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


Acknowledgement

The research was granted by Research Council of Lithuania, project No. MIP-065/2015.



Page 16 of 19

References (1)

Fang, F. C.; Casadevall, A. Reductionistic and Holistic Science. Infect. Immun. 2011, 79 (4), 1401–1404.

(2)

Darwish, A. Bio-Inspired Computing: Algorithms Review, Deep Analysis, and the Scope of Applications. Futur. Comput. Informatics J. 2018.

(3)

Cacciola, F.; Dugo, P.; Mondello, L. Multidimensional Liquid Chromatography in Food Analysis. TrAC Trends Anal. Chem. 2017, 96, 116–123.

(4)

Lee, H.-C.; Yokomizo, T. Applications of Mass Spectrometry-Based Targeted and Non-Targeted Lipidomics. Biochem. Biophys. Res. Commun. 2018.

(5)

Callao, M. P.; Ruisánchez, I. An Overview of Multivariate Qualitative Methods for Food Fraud Detection. Food Control 2018, 86, 283–293.

(6)

McCalley, D. V. Understanding and Manipulating the Separation in Hydrophilic Interaction Liquid Chromatography. J. Chromatogr. A 2017, 1523, 49–71.

(7)

Vercruysse, K. P.; Clark, A. M.; Bello, P. A. F.; Alhumaidi, M. Using Size Exclusion Chromatography to Monitor the Synthesis of Melanins from Catecholamines. J. Chromatogr. B 2017, 1061–1062, 11–16.

(8)

Chen, C.; Zuckerman, D. M.; Brantley, S.; Sharpe, M.; Childress, K.; Hoiczyk, E.; Pendleton, A. R. Sambucus Nigra Extracts Inhibit Infectious Bronchitis Virus at an Early Point during Replication. BMC Vet. Res. 2014, 10 (1), 24.

(9)

Schnitzler, P.; Koch, C.; Reichling, J. Susceptibility of Drug-Resistant Clinical Herpes Simplex Virus Type 1 Strains to Essential Oils of Ginger, Thyme, Hyssop, and Sandalwood. Antimicrob. Agents Chemother. 2007, 51 (5), 1859–1862.

(10)

Lileikytė, R.; Telksnys, L. Quality Estimation of Speech Recognition Features for Dynamic Time Warping Classifier. 2012, 41 (3).

(11)

Drevinskas, T.; Telksnys, L.; Maruška, A.; Gorbatsova, J.; Kaljurand, M. Capillary Electrophoresis Sensitivity Enhancement Based on Adaptive Moving Average Method. Anal. Chem. 2018, 90 (11), 6773– 6780.

(12)

Drevinskas, T.; Telksnys, L.; Maruška, A.; Gorbatsova, J.; Kaljurand, M. Compensation of the Baseline Temperature Fluctuations for Autonomous CE-C4D Instrument Working in Harsh Environments. Electrophoresis 2018.

(13)

Orphanidou, C.; Wong, D. Machine Learning Models for Multidimensional Clinical Data; Springer, Cham, 2017; pp 177–216.

(14)

Smusz, S.; Kurczab, R.; Bojarski, A. J. A Multidimensional Analysis of Machine Learning Methods Performance in the Classification of Bioactive Compounds. Chemom. Intell. Lab. Syst. 2013, 128, 89– 100.

(15)

Witek, J.; Smusz, S.; Rataj, K.; Mordalski, S.; Bojarski, A. J. An Application of Machine Learning Methods to Structural Interaction Fingerprints—a Case Study of Kinase Inhibitors. Bioorg. Med. Chem. Lett. 2014, 24 (2), 580–585.

(16)

Lavecchia, A. Machine-Learning Approaches in Drug Discovery: Methods and Applications. Drug Discov. Today 2015, 20 (3), 318–331.

(17)

Poole, S.; Schroeder, L. F.; Shah, N. An Unsupervised Learning Method to Identify Reference Intervals from a Clinical Database. J. Biomed. Inform. 2016, 59, 276–284. ACS Paragon Plus Environment

Page 17 of 19 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


(18)

Bekut, M.; Brkić, S.; Kladar, N.; Dragović, G.; Gavarić, N.; Božin, B. Potential of Selected Lamiaceae Plants in Anti(retro)viral Therapy. Pharmacol. Res. 2017.

(19)

Trivellini, A.; Lucchesini, M.; Maggini, R.; Mosadegh, H.; Villamarin, T. S. S.; Vernieri, P.; MensualiSodi, A.; Pardossi, A. Lamiaceae Phenols as Multifaceted Compounds: Bioactivity, Industrial Prospects and Role Of “positive-Stress.” Ind. Crops Prod. 2016, 83, 241–254.

(20)

Drevinskas, T.; Mickiene, R.; Maruška, A.; Stankevicius, M.; Tiso, N.; Salomskas, A.; Lelesius, R.; Karpovaite, A.; Ragazinskiene, O. Confirmation of the Antiviral Properties of Medicinal Plants via Chemical Analysis, Machine Learning Methods and Antiviral Tests: Methodological Approach. Anal. Methods 2018, 10 (16), 1875–1885.

(21)

Kaškonienė, V.; Maruška, A.; Akuņeca, I.; Stankevičius, M.; Ragažinskienė, O.; Bartkuvienė, V.; Kornyšova, O.; Briedis, V.; Ugenskienė, R. Screening of Antioxidant Activity and Volatile Compounds Composition of Chamerion Angustifolium (L.) Holub Ecotypes Grown in Lithuania. Nat. Prod. Res. 2016, 30 (12), 1373–1381.

(22)

Mosmann, T. Rapid Colorimetric Assay for Cellular Growth and Survival: Application to Proliferation and Cytotoxicity Assays. J. Immunol. Methods 1983, 65 (1–2), 55–63.

(23)

Leibowitz, J.; Kaufman, G.; Liu, P. Coronaviruses: Propagation, Quantification, Storage, and Construction of Recombinant Mouse Hepatitis Virus. In Current Protocols in Microbiology; John Wiley & Sons, Inc.: Hoboken, NJ, USA, 2011; Vol. Chapter 15, p 15E.1.1-15E.1.46.

(24)

Kärber, G. Beitrag Zur Kollektiven Behandlung Pharmakologischer Reihenversuche. Naunyn. Schmiedebergs. Arch. Exp. Pathol. Pharmakol. 1931, 162 (4), 480–483.

(25)

RStudio Team. RSudio: Integrated Development for R. RStudio, Inc.: Boston, MA 2015.

(26)

Mastrangelo, A.; Ferrarini, A.; Rey-Stolle, F.; García, A.; Barbas, C. From Sample Treatment to Biomarker Discovery: A Tutorial for Untargeted Metabolomics Based on GC-(EI)-Q-MS. Anal. Chim. Acta 2015, 900, 21–35.

(27)

Madsen, C.; Würtzen, G.; Carstensen, J. Short-Term Toxicity Study in Rats Dosed with Menthone. Toxicol. Lett. 32 (1–2), 147–152.

(28)

Olsen, P.; Thorup, I. Neurotoxicity in Rats Dosed with Peppermint Oil and Pulegone; Springer, Berlin, Heidelberg, 1984; pp 408–409.

(29)

Judzentiene, A.; Garjonyte, R.; Budiene, J. Variability, Toxicity, and Antioxidant Activity of Eupatorium Cannabinum (Hemp Agrimony) Essential Oils. Pharm. Biol. 2016, 54 (6), 945–953.

(30)

Lachenmeier, D. W.; Uebelacker, M. Risk Assessment of Thujone in Foods and Medicines Containing Sage and Wormwood – Evidence for a Need of Regulatory Changes? Regul. Toxicol. Pharmacol. 2010, 58 (3), 437–443.

(31)

Lachenmeier, D. W.; Walch, S. G. Epileptic Seizures Caused by Accidental Ingestion of Sage (Salvia Officinalis L.) Oil in Children: A Rare, Exceptional Case or a Threat to Public Health? Pediatr. Neurol. 2012, 46 (3), 201.

(32)

Okunade, A. L.; Olaifa, J. I. Estragole: An Acute Toxic Principle from the Volatile Oil of the Leaves of Clausena Anisata. J. Nat. Prod. 1987, 50 (5), 990–991.

(33)

Levorato, S.; Dominici, L.; Fatigoni, C.; Zadra, C.; Pagiotti, R.; Moretti, M.; Villarini, M. In Vitro Toxicity Evaluation of Estragole-Containing Preparations Derived from Foeniculum Vulgare Mill. (Fennel) on HepG2 Cells. Food Chem. Toxicol. 2018, 111, 616–622. ACS Paragon Plus Environment


Page 18 of 19

(34)

Tisserand, R.; Young, R. Essential Oil Safety : A Guide for Health Care Professionals.

(35)

Fagodia, S. K.; Singh, H. P.; Batish, D. R.; Kohli, R. K. Phytotoxicity and Cytotoxicity of Citrus Aurantiifolia Essential Oil and Its Major Constituents: Limonene and Citral. Ind. Crops Prod. 2017, 108, 708–715.

(36)

Astani, A.; Reichling, J.; Schnitzler, P. Screening for Antiviral Activities of Isolated Compounds from Essential Oils. Evid. Based. Complement. Alternat. Med. 2011, 2011, 253643.

(37)

Dunkić, V.; Bezić, N.; Vuko, E. Antiphytoviral Activity of Essential Oil from Endemic Species Teucrium Arduini. Nat. Prod. Commun. 2011, 6 (9), 1385–1388.

(38)

Gavanji, S.; Sayedipour, S. S.; Larki, B.; Bakhtari, A. Antiviral Activity of Some Plant Oils against Herpes Simplex Virus Type 1 in Vero Cell Culture. J. Acute Med. 2015, 5 (3), 62–68.

(39)

Venturi, C. R.; Danielli, L. J.; Klein, F.; Apel, M. A.; Montanha, J. A.; Bordignon, S. A. L.; Roehe, P. M.; Fuentefria, A. M.; Henriques, A. T. Chemical Analysis and in Vitro Antiviral and Antifungal Activities of Essential Oils from Glechon Spathulata and Glechon Marifolia. Pezzuto Pharm Biol 2015, 53 (5), 682–688.

(40)

Chiang, L.-C.; Ng, L.-T.; Cheng, P.-W.; Chiang, W.; Lin, C.-C. Antiviral Activities of Extracts and Selected Pure Constituents of Ocimum Basilicum. Clin. Exp. Pharmacol. Physiol. 2005, 32 (10), 811– 816.

(41)

Astani, A.; Reichling, J.; Schnitzler, P. Comparative Study on the Antiviral Activity of Selected Monoterpenes Derived from Essential Oils. Phyther. Res. 2009, 24 (5), n/a-n/a.

(42)

Armaka, M.; Papanikolaou, E.; Sivropoulou, A.; Arsenakis, M. Antiviral Properties of Isoborneol, a Potent Inhibitor of Herpes Simplex Virus Type 1. Antiviral Res. 1999, 43 (2), 79–92.

(43)

Sokolova, A. S.; Yarovaya, O. I.; Semenova, M. D.; Shtro, A. A.; Orshanskaya, I. R.; Zarubaev, V. V.; Salakhutdinov, N. F. Synthesis and in Vitro Study of Novel Borneol Derivatives as Potent Inhibitors of the Influenza A Virus. Medchemcomm 2017, 8 (5), 960–963.

(44)

Drevinskas, T.; Mickienė, R.; Maruška, A.; Stankevičius, M.; Tiso, N.; Šalomskas, A.; Lelešius, R.; Agneta, K.; Ona, R. Cytotoxic Attributes in Polyphenolic and Volatile Compounds Rich Antiviral Plants. Chemija 2018, 29 (2), 124–131.

(45)

Peckner, R.; Myers, S. A.; Jacome, A. S. V.; Egertson, J. D.; Abelin, J. G.; MacCoss, M. J.; Carr, S. A.; Jaffe, J. D. Specter: Linear Deconvolution for Targeted Analysis of Data-Independent Acquisition Mass Spectrometry Proteomics. Nat. Methods 2018.


Page 19 of 19 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


For TOC only


Chromatographic Data Segmentation Method: a Hybrid Analytical

Recommend Documents