Non-targeted identification of non-volatile organic compounds in a face mask and methods of characterization thereof

Abstract

The application discloses a method for non-target identification and characterization of non-volatile organic compounds in a mask, comprising the following steps: (1) sample pretreatment; (2) performing ultra-high performance liquid chromatography separation and quadrupole-electrostatic field orbitrap high-resolution mass spectrometry analysis on the sample; and (3) data processing and analysis. The application characterizes the detected compounds in different types of masks by adopting hierarchical cluster analysis and principal component analysis methods, and can provide valuable references for monitoring of new non-volatile organic chemicals and improvement of mask safety production measures.

Description

Technical Field

[0001] This invention relates to a method for identifying and characterizing chemical substances, and in particular to a method for non-targeted identification and characterization of non-volatile organic compounds in face masks. Background Technology

[0002] Face masks, as a major category of daily protective equipment, are mainly divided into medical surgical masks, disposable medical masks, N95 or KN95 masks, and cotton masks. Masks generally have a special structure and density that can effectively prevent dust, droplets, and aerosols from entering the mouth and nose, or block airborne bacteria and viruses. In March 2020, the World Health Organization called for an end to a global shortage of medical products, estimating that approximately 89 million medical masks were needed globally each month, according to its calculations.

[0003] Face masks are typically designed and constructed using materials such as woven fabrics and polypropylene, which inevitably contain certain chemical impurities, contaminants, and intermediates. During the mask sterilization process, chemicals such as ethylene oxide, hydrogen peroxide, ozone, ethanol, and formaldehyde are sometimes used. These compounds and their byproducts remaining inside the mask may pose a potential health risk to the user. For example, reports have indicated that two toxins, 4-hydroxy-4-methyl-2-pentanone and ethyl 2-hydroxyacetate, were found in N95 masks after sterilization with ethylene oxide. Studies have also shown that face masks have become a new source of exposure to phthalates and organophosphates in humans and the environment. Furthermore, it remains uncertain whether other chemicals may be introduced during the entire manufacturing, processing, packaging, and transportation process, as well as in the raw materials used to manufacture masks.

[0004] Given that masks fit snugly against the face and come into contact with the mouth and nose during use, it is necessary to conduct broad-spectrum identification and assessment of the risks posed by unidentified chemical substances contained within them. Currently, some researchers have conducted non-targeted screening of chemical substances in masks. For example, some researchers used headspace gas chromatography combined with quadrupole-electrostatic field orbital trap high-resolution mass spectrometry (HMS) to non-targeted identify unknown volatile organic compounds (VOCs) in medical masks. A total of 69 VOCs were identified in 60 masks, categorized into nine groups. Alkanes, esters, benzenes, and alcohols were the most abundant, accounting for 34.8%, 15.9%, 10.1%, and 7.2% of the total, respectively. Other researchers used gas chromatography combined with HMS to analyze potentially inhalable (semi-)volatile organic compounds in three different types of disposable polypropylene masks. A preliminary analysis of 79 compounds was conducted in all masks, including 16 benzene derivatives, 20 alkanes, 10 phenols, 11 halides, 5 naphthalenes, and 5 esters. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide a non-targeted identification and characterization method for non-volatile organic compounds in masks, which uses liquid chromatography-high resolution mass spectrometry to perform non-targeted identification of non-volatile compounds contained in different types of masks.

[0006] The present invention relates to a method for non-targeted identification and characterization of non-volatile organic compounds in face masks, comprising the following steps:

[0007] (1) Sample pretreatment;

[0008] (2) The samples were separated by ultra-high performance liquid chromatography and analyzed by quadrupole-electrostatic field orbital trap high-resolution mass spectrometry.

[0009] (3) Data processing and analysis.

[0010] The non-targeted identification and characterization method for non-volatile organic compounds in masks described in this invention, wherein the ultra-high performance liquid chromatography separation conditions in step (2) are as follows:

[0011] Chromatographic conditions: Waters ACQUITY UPLC BEH C 18 100mm × 2.1mm id, 1.7μm, column temperature: 30℃; flow rate: 0.4mL / min -1 Injection volume: 2 μL; Mobile phase: A is 5 mM ammonium formate, with pH adjusted to 3.5 by formic acid; B is methanol;

[0012] Table 1 Gradient elution program

[0013]

[0014] The gradient elution procedure is shown in Table 1.

[0015] The non-targeted identification and characterization method for non-volatile organic compounds in masks described in this invention, wherein the quadrupole-electrostatic field orbital trap high-resolution mass spectrometry analysis conditions in step (2) are as follows:

[0016] Electrospray voltage: 3.5 kV, positive ion mode and -2.8 kV negative ion mode; sheath gas pressure: 40 arb; auxiliary gas pressure: 5 arb; ion source temperature: 350℃; transmission metal capillary temperature: 320℃; S-lensRF: 60 V; scan range: m / z 100-800; first-order mass spectrometry full scan resolution: R = 70,000 FWHM, m / z 200; maximum C-trap capacity: 1 × 10⁻⁶ 6 C-trap maximum injection time: 100ms;

[0017] Data-dependent daughter ion full scan resolution: R = 17,500 FWHM; isolation window: ±2 m / z; normalized collision energy: NCE = 20, 40, 60 eV; maximum C-trap capacity: 1 × 10⁻⁶ 5 C-trap maximum injection time: 100ms; Dynamic exclusion: 6s.

[0018] The non-targeted identification and characterization method of non-volatile organic compounds in masks according to the present invention includes, in step (3), data processing and analysis including non-targeted identification of unknown non-volatile organic compounds in masks and characterization of compounds identified in masks.

[0019] The present invention relates to a method for non-targeted identification and characterization of non-volatile organic compounds in face masks, wherein the non-targeted identification of unknown non-volatile organic compounds in face masks includes the following steps:

[0020] The raw data from the collected masks and blank solvents were imported into Compound Discoverer 3.3 software, which integrates various cheminformatics tools to assist in the analysis and interpretation of data during the non-targeted identification process. A modular workflow for the non-targeted identification of non-volatile organic compounds in masks was designed, including spectrum selection, retention time alignment, compound detection, and compound grouping, as detailed below:

[0021] First, in the first "Spectrum Selection" processing module, the raw data file is imported, and the sample and blank solvent are labeled. In the second "Retention Time Alignment" processing module, the retention time deviation is set to 0.2 min. In the third "Compound Detection" processing module, the maximum number of carbon, hydrogen, oxygen, nitrogen, phosphorus, sulfur, fluorine, chlorine, bromine, sodium, and potassium elements in the compound's structural formula is set to 90, 190, 50, 50, 20, 20, 10, 20, and 2, respectively, based on the possible addition forms of the precursor ions [M+H]. +1 [M+NH4] +1 [M–H] –1 Peak extraction was performed with a mass deviation set to 5 ppm. In the fourth "compound grouping" processing module, the elemental composition of the compounds was predicted using the obtained precise mass numbers, isotope profiles, and fragment information. The databases that can be used to predict the elemental composition of compounds include the MassList database and the ChemSpide database. For accurate identification of compounds, secondary mass spectra were also used for matching, and the secondary mass spectrometry library used was the mzCloud database.

[0022] The non-targeted identification and characterization method of non-volatile organic compounds in masks described in this invention, wherein the characterization of the compounds identified in the masks in step (3) includes the following steps: classifying the compounds identified in different types of masks according to the main functional groups in their chemical structural formulas, calculating the adjusted average detection frequency of each type of compound and the molecular weight distribution of each type of compound in different types of masks; and speculating the possible commercial uses of each identified compound.

[0023] The non-targeted identification and characterization method of non-volatile organic compounds in masks according to the present invention includes the following steps in step (1): collecting disposable mask samples of different types and compositions, cutting the samples into 5mm×5mm pieces, weighing 0.5g of the cut sample into a 10mL centrifuge tube, adding 5mL of methanol and acetonitrile solution with a volume ratio of 1:1, ultrasonically extracting at 60℃ and 80kHz for 30min, centrifuging at 10000r / m for 10min, filtering the supernatant through a 0.22μm polytetrafluoroethylene membrane, and using the filtrate for instrument analysis.

[0024] The non-targeted identification and characterization method of non-volatile organic compounds in the mask of this invention differs from existing technologies in that:

[0025] This invention presents a non-targeted identification and characterization method for non-volatile organic compounds (NVCs) in face masks. It utilizes ultra-high performance liquid chromatography-quadrupole-electrostatic field orbital trap high-resolution mass spectrometry (UHPLC-QMS-QPSM) combined with a commercial secondary mass spectrometry library, open-source databases, and the mzLogic algorithm to perform broad-spectrum identification of NVCs in different types of face masks. A total of 47 NVCs were identified, classified structurally based on their functional groups, and further classified according to their potential intended uses. This research can provide valuable reference for the monitoring of novel NVCs and the improvement of safety production measures for face masks.

[0026] The non-targeted identification and characterization method of non-volatile organic compounds in the mask of the present invention will be further described below with reference to the accompanying drawings. Attached Figure Description

[0027] Figure 1 This is a schematic diagram of the non-targeted identification process of non-volatile organic compounds in a mask in the method of the present invention;

[0028] Figure 2 The detection results of different types of compounds in masks in the method of the present invention are shown, including: (a) all masks, (b) disposable medical masks, (c) medical surgical masks, (d) (K) N95 masks, (e) children's masks, and (f) masks sold online.

[0029] Figure 3The following are characterization diagrams of non-volatile organic compounds in the mask in the method of the present invention, wherein: (a) a violin plot of molecular weight distribution; (b) an adjusted average detection frequency plot. Detailed Implementation

[0030] I. Materials and Methods

[0031] 1. Instruments, materials and reagents

[0032] UltiMate 3000 ultra-high performance liquid chromatograph (Thermo Fisher Scientific, USA); Q Exactive quadrupole / electrostatic field orbit trap high resolution mass spectrometer (Q-Orbitrap HRMS, Thermo Fisher Scientific, USA); Milli-Q ultrapure water system (Millipore, USA); XS105 analytical balance (Mettler Toledo, Switzerland); CR21N high-speed refrigerated centrifuge (Hitachi, USA); 0.22μm polytetrafluoroethylene membrane (Millipore, USA); methanol, acetonitrile, and formic acid (MS grade, Fisher Scientific, USA); ammonium acetate (analytical grade, Dima Technology, USA).

[0033] 2. Sample collection and preparation

[0034] Different types and compositions of disposable face masks were purchased online or locally, including 22 batches of disposable medical masks (DMM), 15 batches of medical surgical masks (MSM), 18 batches of (K)N95 masks ((K)N95M), 12 batches of children's masks (CM), and 8 batches of online masks (OM, specifically referring to ordinary protective masks decorated with various eye-catching patterns sold online by companies or organizations). 0.5g of shredded sample (<5mm×5mm) was weighed into a 10mL centrifuge tube, and then 5mL of methanol + acetonitrile (1:1, V / V) was added. The mixture was ultrasonically extracted at 60℃ and 80kHz for 30min, followed by centrifugation at 10000r / m for 10min. The supernatant was filtered through a 0.22μm polytetrafluoroethylene membrane, and the filtrate was used for instrumental analysis.

[0035] 3. Ultra-high performance liquid chromatography separation conditions

[0036] Chromatographic conditions: Waters ACQUITY UPLC BEH C 18(100mm × 2.1mm id, 1.7μm), column temperature: 30℃; flow rate: 0.4mL min -1 Injection volume: 2 μL; Mobile phase: (A) 5 mM ammonium formate (pH adjusted to 3.5 with formic acid) and (B) methanol, gradient elution program is shown in Table 1.

[0037] Table 1 Gradient elution program

[0038]

[0039] 4. Quadrupole-Electrostatic Field Orbital Trap High-Resolution Mass Spectrometry Analysis

[0040] Electrospray voltage: 3.5kV (positive ion mode) and -2.8kV (negative ion mode); sheath gas pressure: 40arb; auxiliary gas pressure: 5arb; ion source temperature: 350℃; transport capillary temperature: 320℃; S-lens RF: 60V; scan range: m / z 100-800; full scan resolution: R = 70,000 FWHM (m / z 200); maximum C-trap capacity (AGC target): 1×10⁻⁶ 6 C-trap maximum injection time: 100ms.

[0041] Data-dependent full scan of daughter ions (dd-MS) 2 Resolution: R = 17,500 FWHM; Isolation window: ±2 m / z; Normalized collision energy: NCE = 20, 40, 60 eV; Maximum C-trap capacity (AGC target): 1 × 10⁻⁶ 5 C-trap maximum injection time: 100ms; Dynamic exclusion: 6s.

[0042] 5. Identifying unknown non-volatile organic compounds in non-targeted masks

[0043] The raw data (in .raw format) of the collected masks and blank solvent were imported into Compound Discoverer 3.3 software (Thermo Fisher Scientific, USA). This software integrates various cheminformatics tools to assist in the analysis and interpretation of data during the non-targeted identification process. Based on the software's own data analysis capabilities and the needs of this study, we designed a modular workflow for the non-targeted identification of non-volatile organic compounds in masks. For example... Figure 1 As shown.

[0044] First, in the first "Spectrum Selection" processing module, the raw data file (in .raw format) is imported, and the sample and blank solvent are labeled. In the second "Retention Time Alignment" processing module, the retention time deviation is set to 0.2 min. In the third "Compound Detection" processing module, the maximum number of carbon, hydrogen, oxygen, nitrogen, phosphorus, sulfur, fluorine, chlorine, bromine, sodium, and potassium elements in the compound's structural formula is set to 90, 190, 50, 50, 20, 20, 10, 20, and 2, respectively, based on the possible addition forms of the precursor ions [M+H]. +1 [M+NH4] +1 [M–H] –1 Peak extraction was performed with a mass deviation set to 5 ppm. In the fourth "Compound Grouping" processing module, the elemental composition of compounds was predicted using the obtained precise mass numbers, isotopic profiles, and fragment information. Databases available for predicting compound elemental composition include the MassList database (Thermo Fisher Scientific) and the ChemSpider database (http: / / www.chemspider.com / ). ChemSpider is an open-source online database containing approximately 128 million chemical structures (279 data sources), encompassing a very wide range of compound categories. Substances identified based on primary precise mass numbers can be used as a reference. For accurate compound identification, secondary mass spectra were matched. The secondary mass spectrum library used was the mzCloud database (Thermo Fisher Scientific), which currently contains secondary mass spectra of nearly 32,000 compounds, totaling over 13 million spectra.

[0045] 6. Characterization of compounds identified in face masks

[0046] The compounds identified in different types of masks were classified according to their main functional groups in their chemical structures. The rescaled average detection frequency (RADF) of each type of compound and the molecular weight distribution of each type of compound in different types of masks were calculated. The possible commercial uses of each identified compound were then speculated.

[0047] II. Results and Discussion

[0048] 1. Accurate identification of non-volatile organic compounds in face masks

[0049] In non-targeted identification, data processing tools and related algorithms may affect the accuracy of the identification results. The Compound Discoverer software assigns elemental composition to detected ions based on primary precise mass numbers and isotopic profiles. Although the mass-to-charge ratio of the ions collected in the experiment has high resolution and accuracy (deviation from the theoretical mass number is less than 5 ppm), a single elemental composition may still match thousands of compounds. Therefore, only by using secondary mass spectrometry matching can the accuracy of compound identification be improved. In this experiment, after manual verification of compounds identified by the secondary mass spectrometry library mzCloud, 46 compounds were accurately identified, and their relevant information is shown in Table 2.

[0050]

[0051]

[0052] 2. Detection of non-volatile organic compounds in different types of masks

[0053] The non-volatile organic compounds (NOCs) identified in the five types of masks (DMM, MSM, (K)N95M, CM, and OM) were classified according to the functional groups in their chemical structures: co-ols, alkanolamines, amides, esters, ketones, dye auxiliaries, phenols, and alkaloids. These can be used as antioxidants, dye auxiliaries, flame retardants, plasticizers, surfactants, UV absorbers, or antibacterial agents, as shown in Table 1. The 46 NOCs were detected a total of 503 times in all mask samples, including 22 batches of disposable medical masks, 15 batches of medical surgical masks, 18 batches of (K)N95 masks, 12 batches of children's masks, and 8 batches of masks sold online. Figure 2 a) Among them, the frequencies of compounds identified in DMM, MSM, (K)N95M, CM and OM mask samples were 149, 30, 143, 98 and 83, respectively. Figure 2 Among the three categories of compounds detected (bf), organic acids, amides, and esters had the highest detection frequency, at 27.4%, 22.1%, and 18.3%, respectively, while alkanolamines and alkaloids had the lowest detection frequency, at 1.8% and 1.2%, respectively. Alkaloids (choline and betaine) were detected in DMM, CM, and OM samples, but at low frequencies, at 0.7%, 1.0%, and 4.8%, respectively. They may be added to mask products as natural antibacterial agents. Bis(4-ethylbenzyl)sorbitol was the only alcohol compound detected, present in all mask samples, with detection frequencies of 8.7%, 6.7%, 6.3%, 6.1%, and 3.6% in DMM, MSM, (K)N95M, CM, and OM samples, respectively. Bis(4-ethylbenzyl)sorbitol can be used as a functional additive for face masks, improving breathability and moisture permeability while maintaining the mask's barrier properties.

[0054] 3. Characterization of non-volatile organic compounds in different types of masks

[0055] The mean molecular weights and SDs of the chemical substances identified in the five types of masks (DMM, MSM, (K)N95M, CM, and OM) were highly similar, at 289.9±80.2, 249.1±70.5, 284.7±82.3, 274.2±90.5, and 253.3±90.6, respectively, with no statistically significant difference in molecular weight (t-test, p>0.05). However, the violin plot showed distinct characteristics in the molecular weight distribution of the five masks. Figure 3 a). A characteristic protruding region at approximately 284 Da was observed in all samples, similar to the molecular weights of oleamide, stearamide, and stearic acid (281, 283, and 284 Da, respectively), indicating a higher detection rate of low molecular weight (approximately 260-300 Da) compounds in the masks. Furthermore, the DMM and (K)N95M samples had wider protruding regions than the MSM, CM, and OM samples, suggesting that the first two types of masks contained a higher concentration of non-volatile organic compounds (NOCs) with molecular weights of approximately 260-300 Da. Figure 3 a).

[0056] The compounds identified in each type of mask were further characterized by adjusted average detection frequency (RADF), a dimensionless measure ranging from 0 to 1 used to describe the similarity of a class of chemicals in a given type of sample. RADF is calculated as follows: RADF = (Total detection frequency of a chemical class of compounds in a sample class / (Total number of samples (batches) × Total number of compounds of that chemical class detected in samples class) - 1 / (Total number of samples (batches)) / (1 - 1 / (Total number of samples (batches))). For example, if A, B, C, and D are all four organic acid compounds detected in sample I, and sample I contains four samples (batches), namely I-1, I-2, I-3, and I-4. Now assume that A, B, C, and D are all detected in these four (batch) samples. Therefore, the total detection frequency of organic acid compounds in sample I is 4 + 4 + 4 + 4 = 16, and the total number of organic acid compounds detected in sample I is 4. According to the above formula, the RADF of organic acid compounds in sample I is [16 / (4×4)-1 / 4] / (1-1 / 4) = 1, which indicates that the situation of organic acid compounds detected in sample I is exactly the same. If the above assumption is true, but then assume that A, B, C, and D are only detected in I-1, I-2, I-3, and I-4 respectively, then the total detection frequency of organic acid compounds in sample I becomes 1 + 1 + 1 + 1 = 4. Therefore, the RADF of organic acid compounds in sample I is [4 / (4×4)-1 / 4] / (1-1 / 4) = 0, which indicates that the situation of organic acid compounds detected in sample I is completely different.

[0057] In this invention, the RADF value of alkaloids in DMM samples was 0, indicating that the alkaloids detected in each DMM sample were completely different, because the alkaloids detected in DMM samples only contained choline, and choline was detected only in one DMM sample. The same applies to the detection of alkaloids in DMM samples and CM samples. The RADF values ​​of esters in DMM samples and phenols and amides in MSM samples were relatively low (0.04-0.07), even though they accounted for the majority of the identified compounds. The low RADF values ​​and low similarity indicate that their presence varies greatly within the same type of mask. On the other hand, a high RADF value (0.52) was observed in DMM samples, which is attributed to the identification of only three phenolic compounds in DMM samples.

[0058] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made by those skilled in the art to the technical solutions of the present invention without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. A non-targeted identification and characterization method for non-volatile organic compounds in face masks, characterized in that: Includes the following steps: (1) Sample pretreatment: Collect disposable mask samples of different types and compositions, cut the samples into 5 mm × 5 mm, weigh 0.5 g of the cut sample into a 10 mL centrifuge tube, add 5 mL of methanol and acetonitrile solution with a volume ratio of 1:1, extract by ultrasonication at 60°C and 80 kHz for 30 min, centrifuge at 10000 r / m for 10 min, filter the supernatant through a 0.22 μm polytetrafluoroethylene membrane, and use the filtrate for instrument analysis; (2) The samples were separated by ultra-high performance liquid chromatography and analyzed by quadrupole-electrostatic field orbital trap high-resolution mass spectrometry; The ultra-high performance liquid chromatography separation conditions are as follows: Chromatographic conditions: Waters ACQUITY UPLC BEH C 18 100 mm × 2.1 mm id, 1.7 μm, column temperature: 30°C; flow rate: 0.4 mL / min -1 Injection volume: 2 μL; Mobile phase: A is 5 mM ammonium formate, adjusted to pH 3.5 with formic acid; B is methanol; Gradient elution program is as follows: The conditions for high-resolution mass spectrometry analysis using a quadrupole-electrostatic track trap are as follows: Electrospray voltage: 3.5 kV, positive ion mode and -2.8 kV negative ion mode; sheath gas pressure: 40 arb; auxiliary gas pressure: 5 arb; ion source temperature: 350°C o C; Temperature of the metal capillary tube: 320°C o C; S - lens RF: 60 V; Scan range: m / z 100 - 800; Level 1 mass spectrometry full scan resolution: R = 70,000 FWHM, m / z 200; C - trap maximum capacity: 1×10 6 C-trap maximum injection time: 100 ms; Data-dependent daughter ion full-scan resolution: R = 17,500 FWHM; isolation window: ± 2 m / z; normalized collision energies: NCE = 20, 40, 60 eV; maximum C-trap capacity: 1 × 10⁻⁶ 5 C-trap maximum injection time: 100ms; Dynamic exclusion: 6s; (3) Data processing and analysis, including characterization of unknown non-volatile organic compounds in non-targeted masks and compounds identified in masks; The process for identifying unknown non-volatile organic compounds in the non-targeted mask includes the following steps: The raw data from the collected masks and blank solvents were imported into Compound Discoverer 3.3 software, which integrates various cheminformatics tools to assist in the analysis and interpretation of data during the non-targeted identification process. A modular workflow for the non-targeted identification of non-volatile organic compounds in masks was designed, including spectrum selection, retention time alignment, compound detection, and compound grouping, as detailed below: First, in the first "Spectrum Selection" processing module, the raw data file is imported, and the sample and blank solvent are labeled. In the second "Retention Time Alignment" processing module, the retention time deviation is set to 0.2 min. In the third "Compound Detection" processing module, the maximum number of carbon, hydrogen, oxygen, nitrogen, phosphorus, sulfur, fluorine, chlorine, bromine, sodium, and potassium elements in the compound's structural formula is set to 90, 190, 50, 50, 20, 20, 10, 20, and 2, respectively, based on the possible addition forms of the precursor ions [M+H]. +1 [M+NH4] +1 , [M–H] –1 Peak extraction was performed with a mass deviation set to 5 ppm. In the fourth "compound grouping" processing module, the elemental composition of the compounds was predicted using the obtained precise mass numbers, isotope profiles, and fragment information. The databases that can be used to predict the elemental composition of compounds include the MassList database and the ChemSpide database. For accurate identification of compounds, secondary mass spectra were also used for matching. The secondary mass spectrometry library used was the mzCloud database. The characterization of the compounds identified in the masks includes the following steps: classifying the compounds identified in different types of masks according to the main functional groups in their chemical structures, calculating the adjusted average detection frequency of each type of compound and the molecular weight distribution of each type of compound in different types of masks; and speculating on the possible commercial uses of each identified compound. The Adjusted Average Detection Frequency (RADF) is a dimensionless metric ranging from 0 to 1, used to describe the similarity of a class of chemical substances in a given type of sample. RADF is calculated using the following formula: RADF = (Total detection frequency of a chemical compound in a certain class of samples / (Total number of samples or batches in that class × Total number of compounds of that chemical compound detected in that class of samples) - 1 / (Total number of samples or batches in that class of samples)) / (1 - 1 / (Total number of samples or batches in that class of samples)).

Images