Soft Hydrogel Actuator for Fast Machine-Learning-

Assisted Bacteria Detection

Filipp V. Lavrentev1, Igor S. Rumyantsev1, Artemii S. Ivanov2, Vladimir V. Shilovskikh1, Olga Yu.

Orlova1, Konstantin G. Nikolaev1, Daria V. Andreeva2 and Ekaterina V. Skorb*1

1
Infochemistry Scientific Center of ITMO University, Lomonosova str. 9, St. Petersburg 191002,

Russia

2
Department of Materials Science and Engineering, National University of Singapore, Singapore,

117575 Singapore

E-mail: skorb@itmo.ru

KEYWORDS: Hydrogel, interface, lactic acid bacteria, eGaIn, multilayer perceptron.

ABSTRACT: We demonstrate that our bio-electrochemical platform facilitates the reduction of

detection time from the 3-day period of existing tests to 15 minutes. Machine learning and

robotized bioanalytical platforms require new principles such as hydrogel-based actuators for fast

and easy analysis of bioactive analytes. Bacteria are fragile and environmentally sensitive

microorganisms that require a special environment to support their lifecycles during analytical

tests. Here, we develop a novel bioelectrochemical platform based on the soft hydrogel/eutectic

gallium-indium alloy interface for the detection of Streptococcus thermophilus and Bacillus

coagulans bacteria in various mediums. The soft hydrogel-based device is capable to support
1
bacteria’ viability during detection time. Current-voltage data is used for multilayer perceptron

algorithm training. The multilayer perceptron model is capable of detecting bacterial

concentrations in the 104-108 CFU/mL range of the culture medium or in the dairy products with

high accuracy (94%). Such fast and easy biodetection is extremely important for food and

agriculture industries, biomedical and environmental science.

INTRODUCTION

Currently, machine learning approaches, and in particular deep learning, are very popular when

processing and analyzing large amounts of data.1 Deep networks are flexible function

approximators that, with enough training data, can be trained to recognize patterns in data. 2 This

approach is increasingly used to analyze a large number of biological images 3–5

and has found its

application to simulate molecular mechanisms 6 and automatic annotation of animal behavior .

1,7

However, our approach is to use deep learning to analyze the current-voltage characteristics data.

Hydrogels are increasingly used in various fields: flexible electronics 8,9

, sensing , wound
10–12

dressing 13,14
and energy saving 15,16
. Soft biomimetic interfaces involving hydrogels and liquid

metals and alloys play an important role in gaining more and more information about various

complex biosystems. Among all complex biosystems, the potential of the “good” microbiome is

of greatest interest.17 Certain lactic acid bacteria have beneficial effects on human health, cure

gastrointestinal disorders and strengthen the host's immune system. 18,19 Bacterial viability is

generally considered an essential metric for optimal probiotic functionality. 20 It is believed that

they must enter the gastrointestinal tract alive and in sufficient quantities (10 7– 108 colony-

forming units (CFU)/mL) for their benefits to be evaluated. 21,22 Earlier we developed a fermented

milk product with the spore-forming probiotic Bacillus coagulans and auxiliary culture

Streptococcus thermophilus in its composition.23

2
During development, it was crucial to control the number of viable bacteria during storage.

Various methods, which have certain disadvantages, are used to count the number of viable

bacteria. For all these methods, bacteria must be grown selectively in a cultural medium. This

medium must contain certain substances, have a specific pH, and a suitable oxygen level (or lack

of oxygen for anaerobic bacteria). 24,25 However, these methods do not take into account the

possibility of transition to a viable but uncultivated state in which bacteria do not multiply, but

are metabolically active.26,27 Selective staining with membrane-permeable fluorescent dyes to

identify live and dead bacteria require manual counting. 28,29 Application of a deep learning

methods for bacteria images processing allows a great gain in a data analysis and may be

implemented for a very sensitive methods, such as holographic microscopy, however it still

requires a training with selective staining and specific microscopy setup.30 The real-time

polymerase chain reaction method is complicated and does not allow distinguishing the DNA of

living cells from dead ones.31 Some other approaches were tried, such as counting bacteria on a

microfluidic chip by implementation of differential resistive pulse sensing (RPS). 32 However, this

method has disadvantages in recognizing solutions with a high and low concentration of bacteria.

And the possibility of separating live and dead bacteria by the RPS method has not been proven.

Currently, electrochemical methods are widely applied for numerous analytical tasks in colloid

science and analysis of small biological objects. Cyclic electrochemical method combined with

artificial neural network analysis allows to estimate thin layer parameters with the focus on a

number of small point defects and their development. 33 Electrochemical Impedance spectroscopy

method applied to a biopolymer coated substrate also shows a high sensitivity for a hole-like

defects and the corrosion caused by biofilf inhomogenity. 34 Considering the bacteria behavior

resembling inhomogeneous biofilm, it can be a great achievement to apply cyclic voltammetry

3
for the detection and enumeration of bacteria are attracting considerable attention for their

intrinsic advantages, including fast response, simple equipment, and high specificity. 35–37 State of

the art research suggests that electrochemical sensors will inevitably become a viable alternative

to conventional methods in the near future. Electrochemical impedance spectroscopy (EIS) is a

method for studying conductive materials and phase boundaries. 38 This method applies a low

amplitude, variable frequency cyclic function to the transducer and uses the resulting current to

calculate the impedance.38 Initially, the EIS method was used to quantify the total biomass in

samples;39 later, its application to electrodes modified with DNA-probes or antibodies was a

breakthrough in the selectivity of the method. 40 The conductometric method, like the EIS, does

not need labels as a means of rapid detection of bacteria, which simplifies sensor preparation. 41

However, its detection rates are still low compared to traditional methods.42 There are hydrogel-

based systems that are used to detect various biological molecules.8 This technique has already

proven itself well as a selective analysis for the tick-borne encephalitis virus. 10 We propose a soft

flexible hydrogel/liquid metal actuator for bacteria monitoring in different medium and products.

Such a device provides faster and cheaper bacteria quantitative analysis in comparison with

existing techniques.

EXPERIMENTAL SECTION

Materials. Agar (powder) was purchased from Sigma-Aldrich. MRS broth (de Man, Rogosa,

Sharpe) (powder) was purchased from HiMedia, India. Commercial probiotic preparation

LactoSpore® as a source of the spores of Bacillus coagulans MTCC 5856 was used. The

authenticity of the strain of Bacillus coagulans in LactoSpore® has been established through

genotyping and phenotypic characterization. LactoSpore® contains a mixture of ellipsoidal

terminal spores and vegetative cells. During cultivation in MRS broth nutrient medium, all spores

4
germinated into vegetative cells. The starter culture of Streptococcus thermophilus TA 40 (TA 40

LYO 50 DCU) is purchased from Danisco, France. PBS tablets were purchased from Rosmedbio.

An eutectic alloy of gallium-indium was purchased from the Megaclassic company.

Methods: Preparation of the hydrogel. To prepare the gel, 0.1 g of agar was added to 9 mL of

phosphate-buffered saline, and then this solution was stirred and sterilized at 121°C. To prevent

the death of bacteria, the solution was cooled down to 45-50°C and poured into a Petri dish to 1

mL of bacteria suspension (inoculant). The inoculant was prepared in advance by separating the

overnight cultivation of Streptococcus thermophilus and Bacillus coagulans bacteria in a nutrient

medium MRS with the addition of lactose (5 g / L). Then, the inoculants with two bacteria were

mixed 1:1. To obtain various concentrations of bacteria, 1 mL of inoculant was serially diluted in

9 mL of 0.01 M PBS. 1 mL of fermented milk product was dispersed and mixed with 9 mL of

agar solution with PBS to prepare gel samples with yogurt. Counting the number of viable

bacteria in the inoculum and dilutions was made by counting the CFU on the plates.

Scanning electron microscopy images of mixed gallium oxide and lactate insoluble film.

Scanning electron microscopy (SEM) studies were performed with a Hitachi S-3400N scanning

electron microscope. Images were acquired for insoluble film formed on the eGaIn electrode

surface at an operating voltage of 5 kV. Mixed gallium oxide-lactate film was prepared by the

placing of hydrogel with bacteria on the liquid metal electrode. Under the applied voltage gallium

is oxidizing and lactic acid produced by bacteria interacts with gallium cations. Formed film is

transferred on silicon wafer with a layer of excess gallium-indium alloy which was used as a

background. Elemental compositions of acquired films were estimated by means of energy-

dispersive X-Ray spectroscopy. The spectra acquisition conditions were as follows: 5 kV

5
accelerating voltage; 0.2 nA beam current, 120 sec dwell time per spectrum. In order to avoid

film degradation, spectra were acquired from an area of 15 µm2.

DFT calculation. The quantum chemical calculations and full geometry optimization of all

model structures were carried out by the ORCA 4.2.1 program package. 42 Convergence

tolerances for geometry optimization procedure: energy change = 5.0000e-06 Eh, maximal

gradient = 3.0000e-04 Eh/Bohr, RMS gradient = 1.0000e-04 Eh/Bohr, maximal displacement =

4.0000e-03 Bohr, RMS displacement = 2.0000e-03 Bohr. Grid4 and FinalGrid5 specifications

were used in all cases. Symmetry operations have not been applied during the geometry

optimization procedure for all model structures. The Hessian matrices were calculated for all

optimized model structures in order to prove the location of correct minima on the potential

energy surfaces (no imaginary frequencies in all cases) and to estimate the thermodynamic

parameters, the latter being calculated at 298.15 K and 1.00 atm (Table S1).

Training set construction. Current-voltage (I-V) characteristic curves were measured by

Keithley 6430. For the database, the I - V curves were measured three times at 400 points per

cycle for three independent samples per concentration range in the voltage sweep mode in the

following potential values: -0.02 to 0.02 V, from -0.1 to 0.1V, -0.5 to 0.5V, -1 to 1V, -5 to 5V.

The I-V characteristics for all samples with different concentrations of bacteria are shown in

Figure S2. The database for the concentration of bacteria is compiled on the basis of the

parameters of electrochemical processes - voltage and current. Voltage ranges were chosen as

descriptors. The data themselves represent the values of the currents obtained from the entire

cyclic volt-ampere curve. Since it was required to obtain the same amount of data from each

sample, the speed was different in each voltage range. For voltage ranges of 0.02V, 0.1V, 0.5V,

1.0V and 5.0V, it was, respectively, 1 mV / s, 5.2 mV / s, 27 mV / s, 57.1 mV / s, 285.7 mV / s.

6
Thus, each sample in the database is represented by five descriptors with 2400 points in each of

them; whole database consists of 72000 experimental points.

Then, the database underwent a binary normalization process, as a result of which all current

values took values from 0 to 1. This training method was used to train the multilayer perceptron

model.

Machine learning model. For machine learning calculations Weka v3.8.4 (Waikato

Environment for Knowledge Analysis) is used. Dividing the data into a training set (90%) and a

test set (10%) allows the accuracy of the model to be checked.The predictive model is trained on

the database. Such a dataset consts of columns filled with attributes that determine the answers. 44

Since the peaks of the redox potential are individual for hydrogel compositions, currents are

attributes, and bacteria concentrations are classes. Machine learning model use attributes and find

statistical weights for them to predict classes. The multilayer perceptron model was chosen due to

its relatively low complexity and low-cost performance. For data processing by supervised

learning method, a multilayer perceptron model with the following parameters was used: learning

rate - 0.1, momentum – 0.1, random seed – 0, the number of hidden layers – 3, the number of

nodes in the hidden layer – 12, each of which is a weighting factor. The nodes alternately

generate an activation potential:

N
z=∑ x i wi +b ,
i=1

Where z is the activation potential, z is the descriptor value, w is the weighting factor of the node,

b is the bias.

The values of the activation function are assessed using the activation function, which is sigmoid.

Through this process, objects are classified. The weights (wi) during the training of the model are

7
fitted using the backpropagation method. In the course of its operation, the model uses the input

data, corresponds to the result and compares it with the original answers, after which the weights

are adjusted taking into account the error gradient (E):

∂E t −1
Δ wi , j=η + γΔ wi , j ,
∂ wi , j

Where η - training time, γ - momentum.

RESULTS AND DISCUSSION

We propose using of the current-voltage (I−V) characteristics at the interface between the

eutectic gallium-indium alloy (eGaIn) electrode and the hydrogel (Figure 1) to detect

Streptococcus thermophilus and Bacillus coagulans bacterial concentrations in the range from

104 to 108 CFU/mL.

Considering that water (or other liquid) accounts for 12% of the structure of a bacterium, its

population increment will affect the electrical conductivity and dielectric constant of a

suspension. Accordingly, any suspension with bacteria can be analyzed by statistically measuring

its electrochemical behavior using the cyclic voltammetry method. Due to the formation of

various redox states on the electrode-electrolyte interface, different signals will occur on the I-V

curves. Analysis of the data obtained may indirectly indicate the presence of a certain number of

bacteria. For the full spectrum of generation of redox states, it is necessary to use an electrode

material capable of electric current induced polymorphic transition in a given voltage range. One

such material is eGaIn alloy, which consists of 75 wt % Ga and 25 wt % In. 45 This alloy has been

successfully used to create flexible electronics due to its low melting point, high electrical

conductivity, non-toxicity, and low saturated vapor pressure. Gallium is quickly passivated in air,

forming a thin oxide film on its surface, which has a higher resistivity than the pure metal. This
8
electrochemical approach has a clear advantage over the plate counting because it is easier, faster,

and can reduce human error.

The electrochemical system presented in Figure 2a consists of hydrogels and two eGaIn

electrodes. The hydrogel was made based on 0.01 M phosphate-buffered saline. Since the

measurement is performed in a hydrogel, the composition of the gel includes the bacteria

Streptococcus thermophilus and Bacillus coagulans MTCC 5856 at various concentrations.

Before being added to the hydrogel, the bacteria were cultured in MRS liquid nutrient medium to

maintain viability. Homofermentative lactic acid bacteria carry out the microbiological synthesis

of lactic acid during fermentation (Figure 2b). The bacteria used in this study contain only L-

lactate dehydrogenase and therefore form the L-isomer of lactic acid. When a potential difference

is applied, gallium is oxidized and its Ga(OH) 2+ cations begin to diffuse into the hydrogel. Indium

cations do not exist in such a system because of their reduction by the gallium. The formal

oxidation potential of gallium is -0.529 V, however, in aqueous media, the formal oxidation

potential of gallium shifts to 0.319 V.8 In the hydrogel, gallium cations interact with bacteria,

their metabolites, and cultural medium anions (Figure 2c).

Most of the gallium complexes and compounds formed in the system are soluble. However,

gallium oxide, lactate and phosphate are insoluble. As a result, a film of Ga(CH3CH(OH)COO)3

and Ga2O3 is formed on the electrode surface. In the presence of a small amount of lactic acid,

gallium lactate crystallizes in the form of thin plates. 46 Due to the fact that the reaction in the

electrochemical cell proceeds very quickly, the formation of gallium lactate crystals of a

distinguishable size does not occur in the system. Formation of insoluble oxide-lactate layer is

proved by the scanning electron microscopy images (Figure 2d). The images clearly show the

contrast between the solid film of gallium compounds and the liquid alloy in the fissure, formed

9
during sample deposition on a wafer. EDX spectra of alloy and film surfaces show a significant

excess of the carbon and oxygen amounts in the film (Figure 2e and S1). Presumably, it verifies

the presence of the oxide and lactate, which are constituted the insoluble layer. Moreover,

corrugated surface indicates that the surface layer is solid, and the surface tension is unable to

restore the smooth surface with a minimal amount of energy. However, the transparency of a

layer at higher accelerating voltages limits the thickness by no more than tens of nanometers.

Due to the absence of gallium phosphate, we provide some quantum chemical computations to

evaluate the thermodynamic favorability of binding between gallium cation and lactate and

dihydrophosphate anions. The level of theory for calculations is B3LYP/def2-TZVPP. For

theoretical studies of reactions, the dihydrophosphate anion (H2PO4-) was considered as model

particle due to the order of phosphoric acid dissociation constants Ka1=7.5·10-3, Ka2=6.2·10-8, and

Ka3=4.8·10-13. Results of DFT calculations are presented in Table S1, S2 and reveal that formation

of gallium lactate is much more thermodynamically favorable than formation of gallium

dihydrophosphate. Difference of reactions ΔG is about 39 kJ/mol. That is the reason of

phosphorus absence in EDX spectra of insoluble film.

Resistance of the gallium lactate layer determines the I-V curve of the sample since this insoluble

layer has the higher resistance in system. The dissolution potential of gallium oxide and gallium

lactate increases with film resistance increasing. The presence of various concentrations of

bacteria in the hydrogel also changes the I-V curve. The interaction of gallium cations with

various concentrations of bacteria leads to the formation of lactate and phosphate film in different

ratios. Redox processes of the formation and decomposition of the film with different

constituents determine the shape of the I-V characteristic peaks, which differ in the potential

range of 0.02 – 5V.

10
Various machine learning methods can solve many biological and chemical issues, in particular,

for optimizing metabolic network models and their analysis, as well as predicting complications

of diabetes.47,48 We apply the feed-forward deep neural network model (multilayer perceptron) for

logical data analysis.49 And training such a machine learning model is performed by an error

backpropagation algorithm.50 Multilayer perceptron training allows the detection of individual

states at the eGaIn / hydrogel interface that depend on the hydrogel composition.

The applied voltage ranges for the database are -0.02 to 0.02V, -0.1 to 0.1V, -0.5 to 0.5V, -1 to

1V, -5 to 5V. Collected I-V characteristics (Figure 3) correspond for a two-electrode

system/hydrogel interface. Final database consists of 6 data points – PBS control and 5

concentration ranges (104 - 108 CFU/mL). Each data point includes 12000 experimental points

which are evenly distributed in 5 descriptors by voltage ranges. New experimental points from

independent experiments are sequentially added to the database until the samples are successfully

classified. The collected data was split into a training dataset (90%) and a test dataset (10%) to

test the accuracy of the model. This proportion is optimal from the point of view of preserving

large amounts of data and does not require a significant reduction in the test set. The multilayer

perceptron model is trained with the following parameters: learning rate – 0.1, momentum – 0.1,

the number of hidden layers – 3, the number of nodes in the hidden layer – 12 (Figure 4a).

Classification errors (Figure 4b) also indicate a high accuracy in determining the composition of

hydrogels. An excellent determination accuracy of 94% is achieved.

In five out of six cases, the trained multilayer perceptron model obtained correct results with the

highest probabilities (Figure 5). Certain difficulties arose with the determination of the

concentration of bacteria 104 (Figure 5a). From I-V curves it is visible that the concentration of

104 has a similar appearance to the control sample without bacteria (Figure S2). This uncertainty

11
arises from the multiple responses to phosphate anions, lactate and, other compounds at low

concentrations. This is accompanied by the presence of a weak useful signal. This disadvantage

can be compensated for by the fascination with the database. Further addition of statistically

significant values to the database will lead to a decrease in the statistical error.

The collected database was used to determine the concentration of Streptococcus thermophilus

and Bacillus coagulans bacteria in a yogurt sample as an analyte to assess the potential for

applying the proposed approach to dairy products. The multilayer perceptron model trained on

the collected database gives the correct result with the highest probability (Figure 6).

CONCLUSION

We have proposed a soft biomimetic monitoring device that can detect bacteria of various

concentrations with 15 mins time versus a traditional period of three days. The device can be

used to determine the concentration of wide range of bacteria. In its essence, our approach is

essential for microbiological analysis at various levels. The main advantage is to use a one-time

collected database and machine learning algorithm, that avoids calibration every time. The

multilayer perceptron algorithm shortens the bacteria detection time and has the highest

percentage of accuracy, which increases the statistical reliability of predictions. The proposed

reliable method can be used to solve a wide range of complex biological problems. The presented

experimental concept can be further applied to miniaturized monitoring systems.

CONFLICTS OF INTEREST

The authors declare no competing financial interest.

ASSOCIATED CONTENT

12
Supporting Information 1: Figure S1, SEM image of an insoluble film formed on the eGaIn

electrode surface with EDX analysis showing distribution map of Ga, C, O. Figure S2, example

of I-V curves of different modes and gel with various concentrations used for a database. Table

S1, calculated Gibbs free energies change for reactions proceeding in a hydrogel. Table S2,

cartesian atomic coordinates for all optimized model structures.

Supporting Information 2: Dataset used for the machine learning approach (CSV)

ACKNOWLEDGEMENTS

The authors acknowledge RSF grant No 21-13-00403 for financial support.

13
Figure 1. Brief scheme of the machine-learning-assisted electrochemical platform for

determining the concentration of bacteria in comparison with classical method.

14
Figure 2. (a) Scheme of the electrochemical system. (b) Formation of the primary metabolite of

lactic acid by bacteria Streptococcus thermophilus and Bacillus coagulans in the gel structure. (c)

Interaction of lactate anions with eGaIn with the formation of insoluble reaction products on the

electrode surface. (d) SEM image of an insoluble film formed on the eGaIn electrode surface

(view from top). (e) EDX spectra of alloy and film surfaces.

15
Figure 3. Characteristic curves of I-V characteristics with 400 points per cycle for gels with

different concentrations of Streptococcus thermophilus and Bacillus coagulans bacteria (0.01 M

PBS pH 7.4) in a voltage range of -1 to 1V. (a) I-V curve for a gel without bacteria. (b) I-V curve

for a gel with 104 bacteria per milliliter. (c) I-V curve for a gel with 10 5 bacteria per milliliter. (d)

I-V curve for a gel with 106 bacteria per milliliter. (e) I-V curve for a gel with 107 bacteria per

milliliter. (f) I-V curve for a gel with 108 bacteria per milliliter.

16
Figure 4. a) Scheme of multilayer perceptron with three hidden layers. Three hidden layers with

twelve nodes are used to describe the relationship of current-voltage data from the input layer and

sample compound data from the output layer. b) Prediction error graph - the axes show the actual

and predicted values, the correct values are indicated by crosses (C – control).

17
Figure 5. The results of the prediction of samples with different concentrations of Streptococcus

thermophilus and Bacillus coagulans bacteria, obtained on the model of a multilayer perceptron.

(a) Graph for the sample with bacteria concentration 104 CFU / mL. (b) Graph for the sample

with bacteria concentration 105 CFU / mL. (c) Graph for the sample with bacteria concentration

106 CFU / mL. (d) Graph for the sample with bacteria concentration 10 7 CFU / mL. (e) Graph for

the sample with bacteria concentration 108 CFU / mL. (f) Graph for the sample without bacteria.

18
Figure 6. Graph with the result of the yogurt sample compound prediction.

19
REFERENCES

(1) Vogt, N. Machine Learning in Neuroscience. Nat. Methods 2018, 15 (1), 33–33.

https://doi.org/10.1038/nmeth.4549.

(2) Speiser, A.; Müller, L.-R.; Hoess, P.; Matti, U.; Obara, C. J.; Legant, W. R.; Kreshuk, A.;

Macke, J. H.; Ries, J.; Turaga, S. C. Deep Learning Enables Fast and Dense Single-

Molecule Localization with High Accuracy. Nat. Methods 2021, 18 (9), 1082–1090.

https://doi.org/10.1038/s41592-021-01236-x.

(3) Belthangady, C.; Royer, L. A. Applications, Promises, and Pitfalls of Deep Learning for

Fluorescence Image Reconstruction. Nat. Methods 2019, 16 (12), 1215–1225.

https://doi.org/10.1038/s41592-019-0458-z.

(4) Weigert, M.; Schmidt, U.; Boothe, T.; Müller, A.; Dibrov, A.; Jain, A.; Wilhelm, B.;

Schmidt, D.; Broaddus, C.; Culley, S.; Rocha-Martins, M.; Segovia-Miranda, F.; Norden,

C.; Henriques, R.; Zerial, M.; Solimena, M.; Rink, J.; Tomancak, P.; Royer, L.; Jug, F.;

Myers, E. W. Content-Aware Image Restoration: Pushing the Limits of Fluorescence

Microscopy. Nat. Methods 2018, 15 (12), 1090–1097. https://doi.org/10.1038/s41592-018-

0216-7.

(5) Gómez-de-Mariscal, E.; García-López-de-Haro, C.; Ouyang, W.; Donati, L.; Lundberg, E.;

Unser, M.; Muñoz-Barrutia, A.; Sage, D. DeepImageJ: A User-Friendly Environment to

Run Deep Learning Models in ImageJ. Nat. Methods 2021, 18 (10), 1192–1195.

https://doi.org/10.1038/s41592-021-01262-9.

(6) AlQuraishi, M.; Sorger, P. K. Differentiable Biology: Using Deep Learning for

Biophysics-Based and Data-Driven Modeling of Molecular Mechanisms. Nat. Methods

2021, 18 (10), 1169–1180. https://doi.org/10.1038/s41592-021-01283-4.

(7) Kabra, M.; Robie, A. A.; Rivera-Alba, M.; Branson, S.; Branson, K. JAABA: Interactive
20
 Machine Learning for Automatic Annotation of Animal Behavior. Nat. Methods 2013, 10

(1), 64–67. https://doi.org/10.1038/nmeth.2281.

(8) Ivanov, A. S.; Nikolaev, K. G.; Novikov, A. S.; Yurchenko, S. O.; Novoselov, K. S.;

Andreeva, D. V.; Skorb, E. V. Programmable Soft-Matter Electronics. J. Phys. Chem. Lett.

2021, 12 (7), 2017–2022. https://doi.org/10.1021/acs.jpclett.1c00007.

(9) Ding, Y.; Tang, H.; Zhang, C.; Li, W.; Li, G.; Zhang, Y.; Xu, C.; Zhao, F.; Guo, Q.; Guo,

C. F.; Xiang, X. ‐D. High‐Throughput Screening of Self‐Healable Polysulfobetaine

Hydrogels and Their Applications in Flexible Electronics. Adv. Funct. Mater. 2021, 31

(18), 2100489. https://doi.org/10.1002/adfm.202100489.

(10) Ivanov, A. S.; Nikolaev, K. G.; Stekolshchikova, A. A.; Tesfatsion, W. T.; Yurchenko, S.

O.; Novoselov, K. S.; Andreeva, D. V.; Rubtsova, M. Y.; Vorovitch, M. F.;

Ishmukhametov, A. A.; Egorov, A. M.; Skorb, E. V. Tick-Borne Encephalitis

Electrochemical Detection by Multilayer Perceptron on Liquid-Metal Interface. ACS Appl.

Bio Mater. 2020, 3 (11), 7352–7356. https://doi.org/10.1021/acsabm.0c00954.

(11) Jia, Z.; Müller, M.; Schönherr, H. Towards Multiplexed Bacteria Detection by Enzyme

Responsive Hydrogels. Macromol. Symp. 2018, 379 (1), 1600178.

https://doi.org/10.1002/masy.201600178.

(12) Gevorkian, A.; Morozova, S. M.; Kheiri, S.; Khuu, N.; Chen, H.; Young, E.; Yan, N.;

Kumacheva, E. Actuation of Three‐Dimensional‐Printed Nanocolloidal Hydrogel with

Structural Anisotropy. Adv. Funct. Mater. 2021, 31 (17), 2010743.

https://doi.org/10.1002/adfm.202010743.

(13) Zhang, H.; Sun, X.; Wang, J.; Zhang, Y.; Dong, M.; Bu, T.; Li, L.; Liu, Y.; Wang, L.

Multifunctional Injectable Hydrogel Dressings for Effectively Accelerating Wound

Healing: Enhancing Biomineralization Strategy. Adv. Funct. Mater. 2021, 31 (23),

21
 2100093. https://doi.org/10.1002/adfm.202100093.

(14) Zhu, T.; Jiang, C.; Wang, M.; Zhu, C.; Zhao, N.; Xu, J. Skin‐Inspired Double ‐

Hydrophobic‐Coating Encapsulated Hydrogels with Enhanced Water Retention Capacity.

Adv. Funct. Mater. 2021, 31 (27), 2102433. https://doi.org/10.1002/adfm.202102433.

(15) Cheng, T.; Zhang, Y.; Wang, S.; Chen, Y.; Gao, S.; Wang, F.; Lai, W.; Huang, W.

Conductive Hydrogel‐Based Electrodes and Electrolytes for Stretchable and Self‐Healable

Supercapacitors. Adv. Funct. Mater. 2021, 31 (24), 2101303.

https://doi.org/10.1002/adfm.202101303.

(16) Mandal, S.; Kumari, S.; Kumar, M.; Ojha, U. Supplementary Networking of

Interpenetrating Polymer System (SNIPSy) Strategy to Develop Strong & High Water

Content Ionic Hydrogels for Solid Electrolyte Applications. Adv. Funct. Mater. 2021, 31

(26), 2100251. https://doi.org/10.1002/adfm.202100251.

(17) Lufton, M.; Bustan, O.; Eylon, B.; Shtifman-Segal, E.; Croitoru-Sadger, T.; Shagan, A.;

Shabtay-Orbach, A.; Corem-Salkmon, E.; Berman, J.; Nyska, A.; Mizrahi, B. Living

Bacteria in Thermoresponsive Gel for Treating Fungal Infections. Adv. Funct. Mater.

2018, 28 (40), 1801581. https://doi.org/10.1002/adfm.201801581.

(18) Nagpal, R.; Kumar, A.; Kumar, M.; Behare, P. V.; Jain, S.; Yadav, H. Probiotics, Their

Health Benefits and Applications for Developing Healthier Foods: A Review. FEMS

Microbiol. Lett. 2012, 334 (1), 1–15. https://doi.org/10.1111/j.1574-6968.2012.02593.x.

(19) Parvez, S.; Malik, K. A.; Ah Kang, S.; Kim, H.-Y. Probiotics and Their Fermented Food

Products Are Beneficial for Health. J. Appl. Microbiol. 2006, 100 (6), 1171–1185.

https://doi.org/10.1111/j.1365-2672.2006.02963.x.

(20) Maukonen, J.; Alakomi, H.-L.; Nohynek, L.; Hallamaa, K.; Leppämäki, S.; Mättö, J.;

Saarela, M. Suitability of the Fluorescent Techniques for the Enumeration of Probiotic

22
 Bacteria in Commercial Non-Dairy Drinks and in Pharmaceutical Products. Food Res. Int.

2006, 39 (1), 22–32. https://doi.org/10.1016/j.foodres.2005.05.006.

(21) Kurmann J.A.; Rasic J.L. The Health Potential of Products Con- Taining Bifidobacteria;

Robinson R.K., Ed.; Elsevier Applied Sciences: London, 1991.

(22) Bouhnik, Y. Survie et Effets Chez l’homme Des Bactéries Ingérées Dans Les Laits

Fermentés. Lait 1993, 73 (2), 241–247. https://doi.org/10.1051/lait:1993222.

(23) Lavrentev, F. V.; Ashikhmina, M. S.; Ulasevich, S. A.; Morozova, O. V.; Orlova, O. Y.;

Skorb, E. V.; Iakovchenko, N. V. Perspectives Of Bacillus Coagulans Mtcc 5856 In The

Production Of Fermented Dairy Products. LWT 2021, No. January, 111623.

https://doi.org/10.1016/j.lwt.2021.111623.

(24) Tortora, G.; Funke, B.; Case, C. Microbiology: An Introduction, Pearson.; London, 2016;

Vol. 12.

(25) Corry, J.; Curtis, G.; Baird, R. Handbook of Culture Media for Food Microbiology,

Elsevier Science.; Amsterdam, 2003; Vol. 2.

(26) Xu, H.-S.; Roberts, N.; Singleton, F. L.; Attwell, R. W.; Grimes, D. J.; Colwell, R. R.

Survival and Viability of NonculturableEscherichia Coli AndVibrio Cholerae in the

Estuarine and Marine Environment. Microb. Ecol. 1982, 8 (4), 313–323.

https://doi.org/10.1007/BF02010671.

(27) Lahtinen, S. J.; Ahokoski, H.; Reinikainen, J. P.; Gueimonde, M.; Nurmi, J.; Ouwehand,

A. C.; Salminen, S. J. Degradation of 16S RRNA and Attributes of Viability of Viable but

Nonculturable Probiotic Bacteria. Lett. Appl. Microbiol. 2008, 46 (6), 693–698.

https://doi.org/10.1111/j.1472-765X.2008.02374.x.

(28) Johnson, M. B.; Criss, A. K. Fluorescence Microscopy Methods for Determining the

Viability of Bacteria in Association with Mammalian Cells. J. Vis. Exp. 2013, No. 79.
23
 https://doi.org/10.3791/50729.

(29) Boulos, L.; Prévost, M.; Barbeau, B.; Coallier, J.; Desjardins, R. LIVE/DEAD®

BacLightTM: Application of a New Rapid Staining Method for Direct Enumeration of

Viable and Total Bacteria in Drinking Water. J. Microbiol. Methods 1999, 37 (1), 77–86.

https://doi.org/10.1016/S0167-7012(99)00048-2.

(30) Verduijn, J.; van der Meeren, L.; Krysko, D. v.; Skirtach, A. G. Deep Learning with

Digital Holographic Microscopy Discriminates Apoptosis and Necroptosis. Cell Death

Discovery 2021, 7 (1), 229. https://doi.org/10.1038/s41420-021-00616-8.

(31) Kramer, M.; Obermajer, N.; Bogovič Matijašić, B.; Rogelj, I.; Kmetec, V. Quantification

of Live and Dead Probiotic Bacteria in Lyophilised Product by Real-Time PCR and by

Flow Cytometry. Appl. Microbiol. Biotechnol. 2009, 84 (6), 1137–1147.

https://doi.org/10.1007/s00253-009-2068-7.

(32) Song, Y.; Zhang, H.; Chon, C. H.; Chen, S.; Pan, X.; Li, D. Counting Bacteria on a

Microfluidic Chip. Anal. Chim. Acta 2010, 681 (1–2), 82–86.

https://doi.org/10.1016/j.aca.2010.09.035.

(33) Finke, A.; Escobar, J.; Munoz, J.; Petit, M. Prediction of Salt Spray Test Results of Micro

Arc Oxidation Coatings on AA2024 Alloys by Combination of Accelerated

Electrochemical Test and Artificial Neural Network. Surface and Coatings Technology

2021, 421, 127370. https://doi.org/10.1016/j.surfcoat.2021.127370.

(34) Kumari, S.; Tiyyagura, H. R.; Douglas, T. E. L.; Mohammed, E. A. A.; Adriaens, A.;

Fuchs-Godec, R.; Mohan, M. K.; Skirtach, A. G. ANN Prediction of Corrosion Behaviour

of Uncoated and Biopolymers Coated Cp-Titanium Substrates. Materials & Design 2018,

157, 35–51. https://doi.org/10.1016/j.matdes.2018.07.005.

(35) Lazcka, O.; Campo, F. J. Del; Muñoz, F. X. Pathogen Detection: A Perspective of

24
 Traditional Methods and Biosensors. Biosens. Bioelectron. 2007, 22 (7), 1205–1217.

https://doi.org/10.1016/j.bios.2006.06.036.

(36) Huang, Y.-M.; Hsu, H.-Y.; Hsu, C.-L. Development of Electrochemical Method to Detect

Bacterial Count, Listeria Monocytogenes, and Somatic Cell Count in Raw Milk. J. Taiwan

Inst. Chem. Eng. 2016, 62, 39–44. https://doi.org/10.1016/j.jtice.2016.01.030.

(37) Zhang, X.; Jiang, Y.; Huang, C.; Shen, J.; Dong, X.; Chen, G.; Zhang, W. Functionalized

Nanocomposites with the Optimal Graphene Oxide/Au Ratio for Amplified Immunoassay

of E. Coli to Estimate Quality Deterioration in Dairy Product. Biosens. Bioelectron. 2017,

89, 913–918. https://doi.org/10.1016/j.bios.2016.09.098.

(38) Barsoukov, E.; Macdonald, J. R. Impedance Spectroscopy: Theory, Experiment, and

Applications; John Wiley & Sons: New Jersey, 2018; Vol. 560.

(39) Grimnes, S.; Martinsen, Ø. G. Bioimpedance and Bioelectricity Basics, 2nd ed.; Academic

Press: Oxford, 2008; Vol. 435. https://doi.org/10.1016/B978-0-12-374004-5.X0001-3.

(40) Mirsky, V. M.; Riepl, M.; Wolfbeis, O. S. Capacitive Monitoring of Protein

Immobilization and Antigen–Antibody Reactions on Monomolecular Alkylthiol Films on

Gold Electrodes. Biosens. Bioelectron. 1997, 12 (9–10), 977–989.

https://doi.org/10.1016/S0956-5663(97)00053-5.

(41) Enami, M.; Amin, M.; Shabani, P.; Gashtil, M. R. Design and Fabrication of a

Conductometry System for Fast Detection of Pathogenic Bacteria in Human Urine. Int. J.

Enteric Pathog. 2020, 8 (1), 32–36. https://doi.org/10.34172/ijep.2020.07.

(42) Neese, F. The ORCA Program System. WIREs Computational Molecular Science 2012, 2

(1), 73–78. https://doi.org/10.1002/wcms.81.

(43) Radke, S. M.; Alocilja, E. C. A High Density Microelectrode Array Biosensor for

Detection of E. Coli O157:H7. Biosens. Bioelectron. 2005, 20 (8), 1662–1667.

25
 https://doi.org/10.1016/j.bios.2004.07.021.

(44) Gasteiger, J.; Zupan, J. Neural Networks in Chemistry. Angew. Chemie Int. Ed. English

1993, 32 (4), 503–527. https://doi.org/10.1002/anie.199305031.

(45) Wang, C.; Wang, C.; Huang, Z.; Xu, S. Materials and Structures toward Soft Electronics.

Adv. Mater. 2018, 30 (50), 1801368. https://doi.org/10.1002/adma.201801368.

(46) Dudley, H. C.; Garzoli, R. F. Preparation and Properties of Gallium Lactate. J. Am. Chem.

Soc. 1948, 70 (11), 3942–3943.

(47) Cuperlovic-Culf, M. Machine Learning Methods for Analysis of Metabolic Data and

Metabolic Pathway Modeling. Metabolites 2018, 8 (1), 4.

https://doi.org/10.3390/metabo8010004.

(48) Dagliati, A.; Marini, S.; Sacchi, L.; Cogni, G.; Teliti, M.; Tibollo, V.; De Cata, P.;

Chiovato, L.; Bellazzi, R. Machine Learning Methods to Predict Diabetes Complications.

J. Diabetes Sci. Technol. 2018, 12 (2), 295–302.

https://doi.org/10.1177/1932296817706375.

(49) Behler, J. First Principles Neural Network Potentials for Reactive Simulations of Large

Molecular and Condensed Systems. Angew. Chemie Int. Ed. 2017, 56 (42), 12828–12840.

https://doi.org/10.1002/anie.201703114.

(50) Schneider, G.; Wrede, P. Prediction of the Secondary Structure of Proteins from the

Amino Acid Sequence with Artificial Neural Networks. Angew. Chemie Int. Ed. English

1993, 32 (8), 1141–1143. https://doi.org/10.1002/anie.199311411.

26
TOC

27