Innovative biosynthesis, artificial intelligence-based optimization, and characterization of chitosan nanoparticles by Streptomyces microflavus and their inhibitory potential against Pectobacterium carotovorum – Scientific Reports

Recent nanotechnology has occupied a unique situation in modern life, owing to the preferred features, especially when applied to biological systems. When compared to bulk materials, nanoparticles (NPs) display novel behaviors due to their small size, diverse shapes, as well as varied optical, thermal, electrical, and mechanical characteristics. Organic nanoparticles include those that are fabricated from organic components such as proteins, carbohydrates, lipids, polymers such as chitosan, and any other organic substance. Typically, these NPs are biodegradable and non-toxic²⁷. The prospective field of application for organic NPs is determined by different criteria, including composition, surface shape, stability, carrying capacity, and others²⁷.

Several protocols were stated for the manufacture of nanoparticles. Physical and chemical approaches, as well as mixture procedures, use highly concentrated stabilizers and reductants that are detrimental to both the environment and human health. The biological-based procedure has emerged with several advantages. The technique is a single-step process and so, environmentally friendly, non-toxic, and requires lower energy, further, the biological-based product has greater stability²⁸. Nanoparticles synthesis based on microbial factories came superior to other methods. Most bio-fabrication of nanoparticles is limited to metal ions, and no describes of CNPs bio-fabrication were established on microbial bases. However, microbial factory techniques follow the green chemistry principles and the biosynthesized chitosan nanoparticles are stable, non-toxic, and free of potentially harmful chemical compounds^29,30.

A total of ten morphologically dissimilar actinomycete strains were screened for their potential for chitosan nanoparticles biosynthesis. Among these isolates, 5 isolates exhibited an obvious ability to biosynthesize CNPs. Of them, the most promising isolate Streptomyces sp. strain NEAE-83 was selected for further studies. No previous effort, according to the best of the authors’ knowledge, has described the synthesis of CNPs by Streptomyces sp. strain NEAE-83. A pioneer biofabrication process of CNPs was reported in the current study using a newly isolated Streptomyces sp. strain NEAE-83.

UV/visible spectra of CNPs generated by Streptomyces sp. strain NEAE-83

The bacterial fabrication pattern of CNPs was checked and examined in the UV/VIS spectra from 200 to 400 nm (Fig. 1A). The optical features of the generated CNPs biopolymer exhibited a solitary severe peak with the greatest absorbance at 310 nm. Figure 1B,C show three vials of chitosan: Streptomyces sp. strain NEAE-83 supernatant, and the CNPs, as well as the nanoparticles after extraction and drying.

The single peak absorbance at 310 nm CNPs by Streptomyces sp. strain NEAE-83 in the UV region lies within the previously reported range of 285 nm³⁰ and 320 nm¹⁵. The UV/visible spectrum of CNPs is considered an indicator of the success of the biosynthesis process, where it is sharp in CNPs, but has a broader absorption band in chitosan^15,30.

Identification of Streptomyces sp. strain NEAE-83

The current Streptomyces sp. strain NEAE–83 has been completely classified depending on its culture, morphology, cell physiology characteristics, and the 16S rRNA sequence. It was non-motile, aerobic, and formed long well-developed substrate mycelium. The reverse side of the colony was brown and the color of the mycelium was gray without diffusible pigments (Fig. 2A). The colonial morphology was consistent with the genus Streptomyces^31,32.

Scanning electron micrograph (Fig. 3) shows that aerial vegetative mycelia are abundant and well-developed and the aerial hyphae are long. It could be noticed that the aerial hyphae distinguished into Rectiflexibiles spore chains bearing more than 30 spores with smooth-surfaced, measuring 0.33 ± 0.04 × 0.89 ± 0.10 µm. The aerial spores are elongated.

Physiological features

The data in Table 1 displayed the physiological features of Streptomyces sp. strain NEAE–83. Milk coagulation, peptonization, hydrolysis of starch, liquefaction of gelatin, protease, uricase, cellulase and chitosanase were positive, but L-asparaginase, nitrate reduction, melanoid pigment, and H₂S production were negative. Optimal microbial development was at 30–37 °C and pH 7 with NaCl tolerance up to 3% (w/v). Diffusible pigments were not detected. These identification criteria are mostly following the genus Streptomyces^2,31,32. Another, the Streptomyces sp. strain NEAE-83 exhibited no antimicrobial activity against all tested Gram-positive and Gram-negative bacteria and fungal strains.

Gene sequencing and phylogenetic analysis

The phylogenetic neighbor-joining tree (Fig. 4B), constructed based on the sequence of 16S rRNA gene (1214 bp), shows the associations between Streptomyces sp. strain NEAE-83 and interrelated species of the genus Streptomyces on the GenBank. The analysis of the phylogenetic tree specified that Streptomyces sp. strain NEAE-83 falls into a clade together with Streptomyces microflavus strain X52 (GenBank/EMBL/DDBJ accession No. MT878548.1), Streptomyces microflavus strain G7 (GenBank/EMBL/DDBJ accession No. MT355851.1), Streptomyces fulvissimus strain MPPS14 (GenBank/EMBL/DDBJ accession No. MT973973.1), Streptomyces fimicarius strain U25 (GenBank/EMBL/DDBJ accession No. MT355869.1) and Streptomyces lavendulae strain L18 (GenBank/EMBL/ DDBJ accession No. MT355862.1).

According to the study of the 16S rRNA sequence of Streptomyces sp. strain NEAE-83, together with its morphological and physiological characteristics, Streptomyces sp. strain NEAE-83 was classified as Streptomyces microflavus strain NEAE-83. The strain was deposited in the Genbank, and the accession number was obtained as MG384964.

Characterization of CNPs

Following the selection of Streptomyces microflavus strain NEAE-83, and the assurance of its efficacy in the bioconversion of bulk chitosan into CNPs, the latter was investigated through several characterization procedures.

Electron microscopy investigation

Investigation of CNPs via scanning electron microscopy (SEM) displayed details on the size and surface morphological construction. CNPs show spherical-like particles (Fig. 5A), with uniformity and homogeneity, without any visible agglomeration. The image of the morphological structure generated by transmission electron microscopy (TEM) displayed the size of CNPs at a range from 60 to 70 nm, without aggregation, and agglomeration (Fig. 5B). SEM and TEM were harmonized in CNPs identification. Both SEM and TEM are efficient tools for discovering nanoparticles, and widely accepted tools for the examination of morphological formation e.g., shape, size, and surface area³³.

The spherical-like particles obtained in this study were in agreement with the spherical or oval shapes of most of the CNPs preparations^34,35. Moreover, the noticed well-dispersed and entangled CNPs lead to the creation of a larger surface area, suggesting the suitable of CNPs for processes that depend on adsorption criteria³⁶.

CNPs that have a small particle size (60–100 nm) and a large surface area could be essential for a wide variety of applications including fabrics³⁷. The particle size of chitosan nanoparticles has a significant impact on the properties of these particles when they are used in pharmaceutical applications. A smaller particle size has the potential to encapsulate a greater quantity of therapeutic substances, improve the drug’s stability and bioavailability, and make it possible for the drug to be administered for a longer period of time³⁸. Chandrasekaran et al.³⁹ proposed that the surface area increases as the particle size decreases. That might be the determining factor in the increased antibacterial activity of nanoparticles of small size.

In disagreement with bulk materials, the current CNPs image shows the non-agglomeration. Although several nanoparticles may have a high ionic force that causes aggregation and agglomeration in the aqueous phase, the CNPs synthesized by Streptomyces microflavus strain NEAE-83 showed no aggregation or agglomeration. There is a distinct variation between them. Aggregation describes toughly bonded or even fused particles whereas agglomeration reveals faintly bonded particles, however, even little agglomeration of CNPs may be accepted⁴⁰.

Energy-dispersive X-ray spectroscopy (EDXS)

The kinds, distributions, and concentrations of elements in CNPs were explored using EDXS via TEM (Fig. 5C). The EDXS spectra of CNPs approve the existence of the main elemental component of native chitosan (carbon, nitrogen, and oxygen). During the fabrication process, CNPs may be subjected to changes in the structure, so EDXS may help detect any variation. Where the internal shell of the atom is hit by a TEM electron ray, in which its electron(s) is displaced by another electron from the external shell to fill the empty position. This process leads to variation in energy in the form of X-ray, the intensity of the X-ray is straightforwardly connected to the concentration and is unique for every element⁴¹.

Particle size analysis

The particle size distribution of chitosan nanoparticles suspension was investigated in this study using a particle size analyzer (PSA). The size distribution was determined at room temperature and included a wavelength range of 1 to 760 nm. The particle size analyzer measurement of a chitosan nanoparticles sample revealed a narrow and sharp peak at 36.6 nm diameter at θ =  90° and 150.2 nm at θ = 11.1° as shown in Fig. 6A. The sizes recorded by the particle size analyzer can only be taken as a relative value and cannot be compared with that determined by electron microscopy⁴². Electron microscopy is capable of detecting the geometric dimensions of the particles given by measuring the width of individual particles from the image, and determining their shape and surface structure (e.g. texture)^43,44. Imaging was favored because of its high-resolution visualization of particles and the minimal effect of artifacts on size determination⁴⁴.

Zeta (ζ) potential analysis

The biosynthesized CNPs were investigated to find out ζ-potential value. The depicted ζ-potential (Fig. 6B) shows that the dispersal has a sole peak, representing brilliant uniformity of CNPs, which had a positive charge with ζ-potential of + 35.6 mV, which confirms the presence of surface protonated amine groups. The current ζ-potential value suggests good stability of the biofabricated CNPs.

The analysis of ζ-potential is a unique way of exploring the characteristics of colloidal dispersals. Zeta potential can have a significant impact on the stability of particles in suspension because of the electrostatic repulsion that exists between individual particles⁴⁵. Yien et al.⁴⁶ reported that the zeta potential of chitosan nanoparticles ranged from + 22 to + 55 mV and their inhibitory effect was influenced by particle size and zeta potential of chitosan nanoparticles. Kheiri et al.⁴⁷ revealed that the zeta potentials of CNPs surfaces have a positive charge about of 45.6 mV. According to the findings of Khan et al.⁴⁸, Raza and Anwar⁴⁹ and Asal et al.⁵⁰, the zeta potential on the surface of CNPs was determined to be + 31 ± 3.14, + 31.3 and + 31 ± 2.2 mV; respectively. In contrast, Qi et al.⁵¹ reported that the surfaces of chitosan nanoparticles had a positive charge of about 51 mV. On the other hand, Iswanti et al.⁵² reported that the surfaces of chitosan nanoparticles had a positive surface charge of + 3.3 ± 0.4 mV. Despite the fact that the suspension is physically stable, Manikandan and Sathiyabama⁵³ mentioned that a zeta potential of at least ± 30 mV is necessary as a minimum for a NPs suspension to be stabilized principally by electrostatic repulsion. If the zeta potential is smaller than + 30 mV, this indicates that the CNPs have less stability due to lower electrostatic repulsion⁵⁴. Another, the positive ζ-potential may be considered another advantage i.e., as antimicrobial agent. Positive ζ-potential of the particles enables ease of interacting with the negative charges on the cell membrane and DNA, then released into the cytoplasm⁵⁵.

Mapping analysis

A mapping analysis was carried out to give a closer look at the behaviors of the CNPs community. The current mapping analysis of the individual responses of nanoparticles was carried out to investigate the pattern of locations and distribution of the elemental nanomaterial (Fig. 7). The data recovered from the electron microscopy show the whole image of the distribution of CNPs that distributed evenly. However, the individual O, C, and N elements of CNPs have the same pattern, where they are evenly scattered and distributed.

Fourier transform infrared (FTIR) investigation

FTIR analysis was carried out to detect the possible incidence of different binding functional groups with CNPs owing to the stabilization and reduction action of Streptomyces microflavus strain NEAE-83 supernatant. The FTIR spectrum of the bio-synthesized CNPs was analyzed and compared with the FTIR spectrum of a chitosan standard (Fig. 8A,B). The first group of bands appeared in the spectra between 4057 and 3750 cm⁻¹, indicating the combination of functional groups of –NH₂, –CH, C–C, and –OH.

Significant shifts of peaks in the spectrum of CNPs from peaks in the spectrum of the chitosan standard indicate a significant role of functional groups in the CNPs biosynthesis. The presence of bands at 3442 cm⁻¹ in the chitosan standard sample indicates strong dimeric OH stretch⁵⁶. Moreover, on the formation of nanoparticles, the hydroxyl (OH) stretching peak at wavelength 3442.09 cm⁻¹ found in the spectrum of chitosan standard shifted to 3326.66 cm⁻¹ which indicates the stretching vibrations of OH groups (Fig. 8A,B). A band at 3420 cm⁻¹ is attributed to combined peaks of –OH and –NH₂ groups stretching vibration in chitosan⁵⁷. The aliphatic stretching group (CH and CH₂) appeared at 2929 cm⁻¹. At 2381–2129 cm⁻¹ the stretching vibrations of C=C conjugated and C≡C has emerged.

Krishnaveni and Ragunathan^56,58 reported that the bands at 1655 cm⁻¹, and those at 1641, 1642 cm⁻¹ indicate the presence of the amide I region. Further, the band at 1654.98 cm⁻¹ (around 1655 cm⁻¹) indicates the stretching vibration of type I amide. The intense peak of the amide I region shifted from 1654.98 cm⁻¹ in the FTIR spectrum of chitosan standard to 1641.48 cm⁻¹ in the FTIR spectrum of CNPs, indicating interactions between protonated amine groups of the chitosan standard with the components of the culture filtrate of Streptomyces microflavus strain NEAE-83. In chitosan nanoparticles, this peak is sharper and shifts toward 1641.48 cm⁻¹ indicating increased interactions or bonding. So, the shifting of vibrations from higher to lower wave numbers reveals the formation of CNPs⁵⁷.

The peak around 1558 cm⁻¹ in the FTIR spectrum of CNPs is due to stretching vibrations of Amide II (CONH₂)⁵⁷. CNPs spectrum showed a band around 1412 cm⁻¹ indicating aromatic C–C stretch⁵⁷. In the FTIR spectrum of chitosan standard, Amide III region presence was indicated by a band at 1381 cm⁻¹⁵⁷. Bands at 1381, 1320 and 895 cm⁻¹ disappeared in the FTIR spectrum after CNPs biosynthesis, revealing that these groups are involved in the CNPs biosynthesis by the culture filtrate of Streptomyces microflavus strain NEAE-83. Absorption in the wavenumber around 1076 cm⁻¹ is produced from the stretch vibration of CO groups (COH and COC) in the oxygen bridge, arising from chitosan deacetylation. The small peaks located at the end of the FTIR spectra correspond to the wagging of the saccharide structure of chitosan^45,59.

FTIR analysis declares the occurrence of capping groups to the surface of CNPs, which stabilizes the CNPs, as well as prevents their coagulation and/or aggregation in the colloidal phase.

X-ray diffraction (XRD) analysis

XRD was utilized for exploring the crystallographic building of CNPs. The pattern of XRD is used in materials science as a fast and primary procedure for the phase detection of the crystallinity to give enough perspective on the unit aspects. As a consequence of this, XRD has been designated as a fingerprint for a specific substance. X-rays irradiation led to the various dispersion angles and intensities of the ray.

The XRD pattern of CNPs sample showed six distinctive peaks at 2θ which were at 11.38, 16.44, 18.27, 22.86, 24.13 and 26.17° (Fig. 8C) indicating a shift from the normal chitosan peaks. Rasaee et al.⁵⁹ reported that the CNPs exhibited diffraction peaks at 2θ = 10° (weak diffraction peak) and 20° (strong diffraction peak), revealing that the chitosan had a high degree of crystallinity. XRD spectrum of CNPs indicates the peak shift from the normal peaks of natural chitosan, which corresponds to the reduction of crystallinity and the increased amorphous structure. The amorphous structure of CNPs contributes to an increased capacity for sorption^45,60.

CNPs’ thermal characteristics

The thermoanalytical properties of the bacterially generated CNPs were tested throughout a controlled temperature range, employing two main ways: thermogravimetric analysis (TGA) and differential scanning calorimetry (DSC). Since the thermoanalytical properties of nanoparticles participate in chemical processes. The primary distinction between TGA and DSC is the technique used to quantify the changes in samples caused by heat^35,60,61.

TGA was accomplished to track the thermal behavior of CNPs as a consequence of a continual fluctuation in the heating ratio (from 25 to 500 °C) to explore mass fluctuation in physical and chemical features of CNPs with the change in temperature (Fig. 9A). TGA showed that from 32.73 up to 60.90 °C, a quick initial mass dropping (− 3.756%) can easily be detected. At higher temperatures, weight losses of CNPs were observed with multistage decomposition. The temperature range of 205.92–355.80 °C recorded the highest weight loss (− 28.683%), which was 1.474 mg, while the lowest weight loss (− 1.051%) was measured over a range of 483.73–499.90 °C.

TGA determines differential thermal analysis by determining the relationship between the alteration in the amount of the sample and different temperatures, or as a fixed temperature at a time and constant loss of mass^35,60,61. Commonly, the temperature at the start of TGA causes dryness of CNPs and can easily cause a quick drop mass of CNPs due to the loss of water-bound located in the two polar groups, this drying change does not result in chemical reactions or lead to any structural alterations⁶⁰. Another reason for the initial weight drop could be the breakdown of volatile units, dehydration of the saccharide rings, and depolymerization³⁶. Later, as the temperature rises, sequential losses in CNPs weight may occur owing to sublimation and/or evaporation. Nevertheless, at higher temperatures, a multistage decomposition owing to the thermal degradation of CNPs could be observed in the form of step-like patterns⁶¹.

Even though the phases of TGA did not interfere during dynamic behavior as seen in the graph, there is a possible unseen decomposition interference, requiring either a much lower rating of heat or using the stepwise (step-by-step) TGA approaches. As a result, TGA alone may not be enough to detect the destructed products; consequently, DSC is frequently essential in addition to TGA to ascertain the presence of intermediate destructed products³⁵.

The thermo-analytical investigation of the DSC was conducted at various rates of heating to clarify the amount (positively or negatively) of variation in the heat flow of CNPs as a result of temperature in the occurrence of the solvent as a reference. Thermo-analytical information was gathered for both CNPs and solvent reference to produce a phase schema (Fig. 9B). With the change in the thermodynamic system, two-phase transitions of CNPs were detected with certain wide endothermic peaks. One peak at 117.22 °C was identified through the range of 81–140 °C, requiring − 199.95 J of heat per gram of CNPs. At 253.95 °C, another peak emerged, resulting in 35.30 J of heat per gram of CNPs. At 229 °C a sole glass transition phase of the CNPs was detected.

The DSC was used to calculate the required or released heat flow versus the change of temperature at a specific time. The thermal analysis displayed some weight loss in the advance stages due to the CNPs decomposition. However, the nanoparticles produced by Streptomyces microflavus strain NEAE-83 did not completely disintegrate at high temperatures (500 °C) and demonstrated some constancy in the structure of the polysaccharide, which could be attributed to the elevated crosslinking feature of the CNPs, which procedures a sturdier and firmer network of the hydrogel^36,60,61.

The phase transitions of the thermal degradation mechanism for every single step were highlighted. The DSC settings are planned to enable the linear holding temperature of a sample as a function of time. So, the physical phenomena of a material can be tracked, such as thermal constancy, purity, and glass transition³⁵.

The first endothermic peak was generated at a low temperature owing to the elimination of water. Then a broad endothermic peak emerged later at 253.95 °C due to the breaking of CNPs cross-linkage. As well, one single glass transition (the phase at which material drives from a hard brittle state into a soft rubbery one) was found in the DSC heating curves at high temperature (229 °C), this is because of the existence of an extreme thermal constancy of the crosslinking, demonstrating the consistency of the CNPs³⁶. The reduction in crystallinity of CNPs after transformation may be owing to alterations in chitosan’s solid-state building caused by crosslinking, and thus CNP breakdown happens above 300 °C⁶⁰.

All previous specifications provide a precise perspective for characterizing CNPs and came into agreement with each other and previous works. Furthermore, the currently proposed microbial method is thought to be ideal for producing high-quality CNPs.

Optimization of CNPs biosynthesis by Streptomyces microflavus strain NEAE-83

Finding out the best settings for extreme biosynthesis of CNPs by Streptomyces microflavus strain NEAE-83 is the prime aim of the trial optimization policy. The bio-fabrication process by Streptomyces microflavus strain NEAE-83 was optimized and modeled based on the advanced technique of RSM, then an artificial intelligence approach was applied. The prediction models by both approaches were generated and compared, as an unprecedented procedure in this field.

Central composite face-centered design (CCFCD)

The statistical modulating approach of RSM was used to set up a four-factors matrix of CCFCD to explore the maximum combination as well as the effects of individual, interaction, and quadratic variables of the tested independent factors on CNPs biosynthesis by Streptomyces microflavus strain NEAE-83. The design matrix and the levels (actual and coded) of the four variables tested, as well as the investigational and predicted CNPs values and their residual errors are presented in Table 2. The trial runs showed various degrees of responses that reached up to 9.41 mg/mL (run  27).

Based on the data obtained from CCFCD, the linear, two-factor interaction, and quadratic models were compared to choose the best fit model (Table 3). Comparison of the model statistics such as lack of fit, and the sum of squares concluded that the quadratic model was the ideal choice, recording insignificant lack of fit error for the probability value (P > 0.05) whereas, the model P-value was significant. Both the sum of squares of prediction error (PRESS) and the standard deviation were lower in comparison with the other models. Finally, the quadratic model had higher values of determination coefficient (R²), adjusted-R², and predicted-R². Therefore, the quadratic model was chosen for the molding CNPs biofabrication process. When using the quadratic model that was created using the CCFCD, the predicted values of CNPs were found to be in line with the actual ones (Table 2).

The quadratic CCFCD model was selected based on the effectiveness of each model. The lower P-value (< 0.05) was reliably significant for the modeling CNPs biofabrication process. Another important choice criterion is R². The R² value indicates the extent to which the experimental parameters can explain the observed response values⁶². If R² is 0.9 or greater, the model is regarded as adequate and more precise for predicting the response^63,64. Generally, the closer to 1, the greater the modeling capacity of the data¹⁸. The present quadratic model has R², adjusted R², and predicted R² values that are all actually close to one. The change in the CNPs experimental response results in the variation in the amount of a factor(s) is known as the R² value, which is always ranged from zero to 1 for all types of R²⁶². Surprisingly, regardless of the significance of the factors, raising the number of predictors (factors) leads to a constant increase in the R² value. As a result, the adjusted R² is a modified version of R² that takes into account the number of model factors. In contrast to R², the adjusted R² changed wisely when adding additional factor(s) to the model. As a result, adjusted R² is a better indicator than R² for determining model fitness. Finally, the predicted R² is employed to evaluate the model’s predictive potential, such as predicting the CNPs value at new untested levels of the factors. The values of predicted-R² and Adjusted-R² should be within 20% of each other, indicating that the model is very significant and accurate, as well as a reasonable agreement between them⁶⁵. The predicted values of CNPs calculated based on the quadratic model of CCFCD were very closest to those of the experimental ones; consequently, the residuals or errors were low, representing another evidence of the accuracy of the model. Consequently, the linear and the two factors interaction models were omitted, and the quadratic model was chosen as the top-fitted one for CNPs bioprocessing by Streptomyces microflavus strain NEAE-83.

Coefficients and multiple regression analysis

The data recovered from the CCFCD were exposed to both multi-regression and variance analysis (ANOVA) (Table 4). The overall model performed significantly at P-value < 0.05. All the model terms, including the effects of quadratic, interaction, and linear terms tracked a similar significant tendency, except the two factors interaction of the initial pH and chitosan concentration, as well as the quadratic effect of both initial pH and chitosan concentration. The lack-of-fit error of the model did not reach the significance threshold, recording a higher P-value of 0.1801. Furthermore, the model standard deviation, coefficient of variation, and adequate precision exhibited decent performance.

The low P-value, together with the high F-value and the nonsignificant lack-of-fit indicated the significance of the suggested overall model. The P-value was used as a diagnosing tool for measuring the significance of both the model and the factors. The process variables with P-values ≤ 0.05 were judged to have significant effects on the response⁶⁶. Also, most of the tested terms are significant (P < 0.05), representing the importance of the tested parameters for CNPs bio-fabrication, and further suggesting that the variables, with their levels, and CCFCD, are well specified and perform optimally for the microbial fabrication of CNPs.

The signal-to-noise ratio is measured by the adequate precision value; a value greater than 4 is desired and indicates a strong model fit⁶⁶. The adequate precision ratio is greater than 4, indicating the fruitful model design space to maximize CNPs biosynthesis at the various tested ranges of the factors. Another element of trustworthiness was the minor coefficient of variation% (C.V.%) value, which is desirable for the model’s reliability. The very low value of C.V.% indicates a greater precision of the performed experiments⁶³. Furthermore, the three types of R² are adequately high to support the significance of both the model and its variables. Both adjusted R² and predicted R² were in harmony. To be in reasonable agreement, both types of R² should be no more than 20% of each other^18,67, indicating high compatibility among trial and predicted values of biosynthesized CNPs and further signifying the reasonable model’s predictive ability within the design space. Furthermore, the coefficient estimate revealed a variety of positive and negative effects. The negative coefficients of some model terms indicate the antagonistic impact at higher concentrations¹⁶, such a variable has an on the microbial construction of CNPs. Whereas a positive coefficient value designates a cooperative impact, and the high level of the variable(s) increases CNPs biosynthesis within the design area.

Checking the model adequacy

To determine the model’s suitability, the adequacy was investigated by plotting the prediction versus the actual points (Fig. 10A). The points of regression analysis are located much nearer to the perfect prediction line and display better fitting of the model predicted values and the experimental results, which confirms the accuracy of the model⁵.

Similarly, the externally studentized residual values of the model versus the predicted one were plotted (Fig. 10B). The graph shows an equivalent scattering of the residual points around the 0-axis, what is more, the residual points were found to be near the 0-axis without linearity. The graph showed that the externally studentized residual values scatter randomly around the 0-axis, suggesting that the variance of the experimental results is constant for all values⁶⁸. This distribution pattern is sufficiently perfect to support the CCFCD model’s applicability. Accordingly, the variance of CNPs data and the biosynthesis process were independent, demonstrating the model’s adequacy and generalizability. As a result, these two sufficiency tests validate the design and data points as well.

Dual relationship of the investigated factors

In order to investigate the dual impact of every pair of the factors on CNPs biosynthesis using Streptomyces microflavus strain NEAE-83, the plots of the three-dimensional (3D) surface of the four independent factors were constructed (Fig. 11). The maximum CNPs biosynthesis was situated at the low chitosan concentration with a high pH value (Fig. 11A), the high incubation temperature with high initial pH (Fig. 11B), high initial pH with low incubation period (Fig. 11C), low chitosan concentration with high incubation temperature (Fig. 11D), low chitosan concentration with low incubation period (Fig. 11E), and low incubation period with high incubation temperature (Fig. 11F), Out of these ranges, CNPs biosynthesis declined sharply. In the analysis of the 3D plots, a tight relationship between every pair of the tested factors could be noticed. This means that the factors’ ranges were wisely chosen, and the model is best appropriate for the design. Accordingly, the greatest combination of the tested in terms of the coded level was estimated based on the next CCFCD model’s prediction function:

where X₁; initial pH, X₂; chitosan concentration (%), X₃; temperature (°C), X₄; incubation period (h).

Modeling CNPs biosynthesis by ANN

The CCFCD data were subjected to more advanced modeling procedures i.e., the artificial intelligence-based approach (Table 2) through machine learning. A predictive ANN model was developed. A fully connected multilayer feed-forward ANN platform was created for modeling the CNPs biosynthesis by the candidate actinomycete.

Despite artificial intelligence has made its road into recent scientific research, no work was reported on the modulation of biofabrication of CNPs by ANN. This is the first work to demonstrate the feasibility of this type of modeling. As a result, artificial intelligence was offered as a possible model for the experimental CCFCD data.

The finest architectural structure was established, numerous hidden layers and neurons within each hidden layer were tested at numerous patterns of ANN-specific factors. The machine learning continued until the error values i.e., mean absolute deviation (MAD), the sum of squared errors (SSE), and root mean square error (RMSE), reached their minimal, complemented by the uppermost value of R², for both training and validation points (Table 5). Numerous trials learning, 5000 tours each, were performed, accordingly, the optimum ANN parameters were found to be at a learning rate of 0.1, engaging the procedure of holdback at 0.2 proportion. The activation function operated at the hidden layers’ nodes was the hyperbolic tangent sigmoid (NTanH).

The topology of ANN under such conditions was found to perform better when having one input layer comprised of the four-neurons (independent factors), one output layer of one-neuron (CNPs biosynthesis), and one in-between hidden layer with 20 neurons; (NTanH (20). Accordingly, the best architectural structure of the ANN topology was generated as 4-20-1 (Fig. 12).

The ANN is flexible enough to generate efficient and accurate models from any kind of response surface, using adequate hidden layers and nodes. The ANN platform employs an algorithm for machine learning that predicts response variables utilizing a supple function. The activation function helps the ANN learn weights and maps the complex nonlinear relationship, even if there is no apparent relation between input(s) and output(s) variables, that is why ANN can predict and fit data very well by modeling different response surfaces using the suitable architecture, and capable of learning any nonlinear function in flexibility and accuracy way^20,21,69.

The current neural network platform (4-20-1) employs an algorithm of machine learning, named a fully connected multilayer perceptron. It is worth mentioning that ANN has an intermediate layer rather instead of a direct path between the input factors and the output CNPs variable.

The intermediate (hidden) layers play a central role in managing a unique correlation between inputs and outputs, instead of the direct path; therefore, ANN is considered an exceptional predictor when the form of relationship between the response(s) and the inputs is not required or recognizable^20,21,69. The ANN activation function employed in the hidden layer nodes is in charge of discovering such linear combinations of the individual variables, without designating the relationship between both the input(s) and the response^20,21,69.

Evaluation of ANN model

Built on such a model, the predicted values of ANN for each experimental data point were estimated (Table 2). Consequently, ANN predicted values show more sensible harmony with the trial ones, and, and their residuals displayed lesser values than those gotten by the CCFCD model.

The predicted values of ANN were drawn against the actual values (Fig. 13). The analysis of linear regression demonstrates improved fitting points in relation to the actual ones. The points are located nearer to the line of the perfect forecast for both the training and validation process, endorsing the suitability of the model.

The residuals analysis plot shows a scattering of the residual data above and below the regression line. The residual analysis showed also a normal scattering of the residuals. This pattern is sufficiently perfect to support the ANN model’s appropriateness.

Comparison of CCFCD and ANN models

In comparison with the CCFCD model, the ANN model has high predictive points with lower residuals (Table 2), where the linear regression, and residual analysis display that the ANN model points (Fig. 13) are located much nearer to the slope of the best forecast than the points of CCFCD model (Fig. 10). This concludes that the ANN can fit the real investigational data precisely with a higher predictive capacity. Therefore, the generalization capability of the ANN model surpasses the CCFCD model.

The model’s accuracy to predict microbial CNPs biosynthesis by both CCFCD and ANN was further compared with a collection of statistical parameters that were estimated for the overall model, as well as, for training and validation groups of CCFCD and ANN (Table 5). The R² values of the ANN were high for both groups than the CCFCD model. In contrast to CCFCD, RMSE and MAD recorded lower values for ANN. The same trend was observed for the overall model i.e., greater R² value and inferior RMSE, MAD, and SSE of ANN in contrasted to CCFCD, deducing the more confidence of ANN model than CCFCD.

As discussed earlier, R² is used to determine the association between the CNPs response and projected values; thus, a bigger value in ANN designates an extra significant association between the two datasets than is reported in CCFCD. RMSE is engaged in regression assessment to validate trial results since a lesser value denotes that the data are aggregated around the area of best fit. MAD is another figure that signifies the average spreading of data around the mean. A minor MAD value infers a reduced spread of data around the mean. The value of SSE is an extra goodness-of-fit test, which is utilized to calculate the total divergence of the actual values from the fitted ones; a lower value designates that the model is appropriate to a greater extent. The current data agree with the recent studies that found the ANN model was superior to RSM, recording lower values of RMSE, MAD, and SSE and greater values of R^221,69. Consequently, ANN has a better generalization capability than CCFCD.

Validation of CCFCD, and ANN models

The experimental validation of CCFCD and ANN models for the maximization of CNPs biosynthesis by Streptomyces microflavus strain NEAE-83 was checked using the computed theoretical values of the four tested factors. The desirability function was used to find the best conditions for maximum CNPs biosynthesis. To validate both models, the optimum theoretical values that maximize the CNPs biosynthesis were estimated by both models to be 5.5 initial pH, 1.3% chitosan, 40 °C, and 12 h incubation period. The corresponding estimated CNPs biosynthesis based on CCFCD, and ANN were 9.29, and 9.44 mg/mL, respectively. Under these conditions, the theoretical values were assessed experimentally. The experimental value of the microbially synthesized CNPs by Streptomyces microflavus strain NEAE-83 was 9.52 mg/mL, confirming a great precision degree of both models. However, the estimated value by ANN was much closer to the experimental one than that estimated by CCFCD, proving again that ANN has an advanced predictive power than the CCFCD model. This supposition, in fact, provides a robust indication for the design’s and modeling procedure’s fitness, applying the tested arrays of the four factors in the bacterial CNPs biosynthesis process.

The current optimum pH of 5 was previously reported as its superior over that at neutral range at pH 7, nanoparticles at pH lesser than 7 tend to be smaller and globular, whereas the particles created at a pH outside 7 tend to be non-spherical and bigger in size or agglomeration of small particles⁷⁰. Our characterization studies support such results.

Regarding the current chitosan concentration of 1.295 mg/mL, chitosan intensity powerfully impacts the size and growth of the nanoparticles. Chitosan concentration at 1.0 mg/mL was among the best conditions to generate CNPs with the lowest size (95 nm) compared to the higher concentrations⁷¹.

The reaction temperature at 40 °C in the current study could be considered suitable from a biotechnological point of view. Since the reaction temperature plays a key role in particle development, and shape/size control, where, the reaction temperature can greatly influence the reaction rate, and hence the particle features⁷². Current characterization findings support the data recovered from the optimization process regarding shape and size.

The current incubation period of 12 h is a reasonably suitable time since, Saifful and Shahidan⁷³ reported that the prolongation of incubation time (18 h) generated a higher average size of nanoparticles compared with the shorter time (2 h). The same conclusion was reported also by Vaezifar et al.⁷¹.

In contrast to the single-step calculation models, the high precision of ANN predictive capability may be owing to its wide ability to estimate the nonlinearity of the system^20,21. However, there are some drawbacks of ANN modeling, of which ANN devours a comprehensive computational period through iterations of estimates, and the organized nature cannot demonstrate the contributions and the significance of each factor in the model, thus factors in the model cannot be reduced or eliminated from the model²¹.

Antimicrobial activity assay of chitosan and the prepared CNPs

The antibacterial activity of chitosan and the prepared CNPs was investigated against the plant pathogenic bacteria Pectobacterium carotovorum using agar diffusion and growth inhibition assay. CNPs showed a significant inhibition zone (19 mm) against the pathogenic bacteria when compared with chitosan solution (11 mm), as shown in Fig. 14A,B. CNPs markedly suppressed the bacterial growth upon increasing its concentration in a liquid medium. The lowest concentration of CNPs which can prevent the growth of bacteria was recorded at 2 mg/mL.

Chitosan is well-known as a natural, non-toxic polymer, that has antibacterial activity. In the agriculture filed, chitosan can be used as a natural pesticide for preventing and controlling plant diseases^74,75. Chitosan in nanoscale provided a novel and unique activities including antimicrobial, antitumor effects, and drug delivery system⁷¹. In the current study, the CNPs showed higher antibacterial activity than chitosan against Pectobacterium carotovorum. This is due to the uniform spherical shape and smaller size of CNPs in comparison to chitosan. The polycationic character of CNPs enables them to interact with the negatively charged bacterial cell wall greater than chitosan itself. Besides, the CNPs have large surface area which facilitates it to be tightly absorbed onto the bacterial surface. This causes disruption to the cell membrane, and leakage of intercellular compounds which in turn affects cell viability⁵¹. Penetration of the CNPs into the bacterial cell causes DNA damage and prevents protein synthesis from RNA⁷⁶. In agreement with other investigations, our results showed that CNPs inhibit the growth of Pectobacterium carotovorum with a diameter of 19 mm. Ahmed et al.⁷⁷ reported that the CNPs inhibit the growth of four isolates of gram- positive (Staphylococcus aureus, Bacillus subtilis) and gram-negative bacteria (Escherichia coli, Pseudomonas aeruginosa) with a diameter range from 9 to 16 mm. In another study by Divya et al.⁷⁸, they reported that the diameter of the inhibition zone produced by CNPs against klebsiella pneumoniae, Staphylococcus aureus, Pseudomonas aeruginosa and Escherichia coli ranged between 12 and 16 mm.

The antimicrobial activity of CNPs various depending on a number of factors like type, molecular weight, degree of polymerization of chitosan, the solvent used, temperature and pH value. CNPs also exhibit dose-dependent and species-specific antimicrobial properties⁷⁹. The gram-positive bacteria showed more sensitivity against CNPs than gram-negative bacteria. Where, the cell wall of gram-positive bacteria is composed of a thick layer of lipoprotein, and phospholipids, that lead to increasing the negative charge on the cell surface⁸⁰. In this study, the minimum inhibition concentration of CNPs against Pectobacterium carotovorum was 2.0 mg/mL (Fig. 14C). Aliasghari et al.⁸¹ reported that the minimum inhibitory concentration (MIC) of CNPs for four strains of Streptococcus was ranged from 1.25 to 2.5 mg/mL. Similarly, CNPs have a bactericidal effect against seven bacterial strains at concentrations various from 0.125 to 5 mg/mL⁷⁷.