Artificial intelligence-assisted optimization of Lentinula edodes extracts for enhanced bioactive profile and therapeutic potential

Abstract

In this study, extraction parameters to increase the biological activities of Lentinula edodes extracts were optimized using both Response Surface Methodology (RSM) and Artificial Neural Network-Genetic Algorithm (ANN-GA) hybrid models. Antioxidant, anticholinesterase, antiproliferative activities, and phenolic contents of the extracts obtained under optimum conditions were determined. According to the findings of the study, it was determined that the extracts obtained with ANN-GA optimization had higher TAS (6.612 mmol/L), FRAP (176.25 mg TE/g), and DPPH (133.00 mg TE/g) values ​​compared to RSM. In addition, TOS (4.167 µmol/L) and OSI (0.063) levels were lower. In anticholinesterase assays, ANN-GA extracts were found to be more effective than RSM in inhibiting AChE (72.47 µg/mL) and BChE (132.13 µg/mL). Furthermore, in antiproliferative assays on A549, MCF-7, and DU-145 cancer cell lines, ANN-GA-optimized extracts were observed to have stronger cytotoxic effects. LC-MS/MS analyses revealed that ANN-GA optimization enriched biologically important phenolic compounds such as gallic acid, protocatechuic acid, and caffeic acid at higher levels. Consequently, AI-assisted optimization offers a powerful strategy for enhancing the biological value of mushroom extracts.

Introduction

Mushrooms, which possess medicinal properties and can be consumed as food, play an important role in both healthy nutrition and disease prevention thanks to their rich nutritional and bioactive compounds¹. In addition to their high protein, fiber, vitamin, and mineral content, they can also exhibit immune-supporting, antioxidant, anticancer, antimicrobial, and cholesterol-lowering effects, particularly through secondary metabolites such as polysaccharides, phenolic compounds, and triterpenes. With these properties, mushrooms are not only a valuable part of the daily diet but also hold great potential as functional foods and natural therapeutic supplements^2,3. Mushroom extract optimization is a process that aims to maximize the biologically active compounds found in mushrooms (e.g., polysaccharides, phenolics, triterpenes, and sterols). In this context, parameters such as extraction temperature, time, solvent type, and solvent ratio are carefully determined and optimized using various statistical methods. This allows the biological activities of mushroom extracts to be revealed more effectively, increasing their usability in the pharmaceutical, nutraceutical, and food industries^4–6.

Lentinula edodes (commonly known as the shiitake mushroom) is one of the most consumed edible mushroom species worldwide due to its high nutritional value and medicinal potential. Recent studies have revealed the beneficial health effects of this species and innovations in cultivation. Rich in nutrients, L. edodes contains polysaccharides (especially β-glucans and lentinan), terpenoids, sterols, lipids, vitamins, and minerals. These bioactive components exhibit immunomodulatory, antioxidant, antitumor, antiviral, antimicrobial, and cholesterol-lowering effects^7–10. The substrates, nutrient media, and breeding methods used are constantly being improved, and modern techniques such as crossbreeding, mutation, and genetic engineering make it possible to obtain new, high-yielding, disease-resistant varieties^11,12. Furthermore, L. edodes, particularly β-glucans and lentinan, has been reported to play a supportive role in cancer treatments by activating the immune system. Furthermore, antioxidant properties prevent cellular damage through the neutralization of free radicals, and polysaccharides are reported to suppress tumor growth. Additionally, water and alcohol extracts have been reported to exhibit antiviral and antimicrobial effects^7–10. With its rich nutritional content and versatile biological activities, L. edodes has significant potential not only as a food but also as a functional food and natural therapeutic supplement. In our study, extract optimization was conducted to maximize the biological activity of L. edodes. The antioxidant, anticholinesterase, antiproliferative activities, and phenolic contents of the optimized extracts were then determined. Although L. edodes has been extensively investigated for its nutritional and medicinal properties, studies integrating extraction optimization with artificial intelligence-based hybrid modeling and comprehensive biological activity evaluation remain limited. The present study addresses this gap by comparatively applying Response Surface Methodology (RSM) and an Artificial Neural Network–Genetic Algorithm (ANN-GA) approach to optimize extraction parameters and directly relate these conditions to antioxidant, anticholinesterase, antiproliferative activities, and phenolic profiles. This study is a systematic investigation of the superiority of ANN-GA based optimization over classical statistical models in enhancing the multifunctional biological activity of L. edodes extracts.

Material and method

The fruiting bodies of the mushroom samples evaluated in this study were obtained from the mycology laboratory of the Department of Biology, Plant Protection and Quarantine of S. Seifullina Kazakh Agrotechnical Research University. Fungal materials are kept at the Department of Biology at Osmaniye Korkut Ata University.

Extraction procedure

In this study, the effects of three key parameters were investigated in detail to optimize the extract yield from L. edodes samples. Samples were dried at 45 °C for 48 h using a laboratory fruit dryer (WMF KITCHENminis, Germany) and then ground to a fine powder using a mechanical grinder. For each experimental condition, 5 g of dried mushroom sample was extracted with 100 mL of solvent. The experimental design used extraction temperature, processing time, and solvent ratio (ethanol/water) as independent variables, and each of these variables was evaluated at three levels. Thus, a full factorial experimental plan consisting of 27 conditions was implemented. Temperatures were set at 30, 45, and 60 °C, extraction times were set at 30, 45, and 60 min, and solvent ratios were set at 0%, 50%, and 100% ethanol. All extraction processes were carried out in an ultrasonic bath (WIGGENS UA22MFDN, Germany) with a frequency of 40 kHz, 100% power capacity, and 400 W. Following these processes, the extracts were filtered through Whatman No. 1 filter paper, and the solvents were removed under vacuum at 40 °C. All experiments were performed in triplicate (n = 3), and the resulting extracts were stored at + 4 °C for later analysis.

Extract optimization

Experimental data were analyzed using Response Surface Methodology (RSM). This statistical approach evaluated interactions between parameter combinations and determined the optimal processing conditions. Moving beyond the modeling phase, a hybrid optimization method combining Artificial Neural Networks (ANN) and Genetic Algorithms (GAs) was applied. This integrated approach revealed the simultaneous effects of the parameters in more detail, allowing for the understanding of complex relationships between variables. This provided a strong scientific basis for AI-based decision support systems.

Response surface methodology (RSM)

This study utilized the Response Surface Methodology (RSM) approach to determine the optimal levels of key parameters affecting the extraction process. In the experimental design process, extraction temperature, application time, and ethanol/water ratio were defined as independent variables. Furthermore, the total antioxidant status (TAS) values ​​of the extracts obtained from different combinations of parameters were analyzed as dependent (response) variables.

Modeling and optimization steps were performed using Design Expert 13 statistical software. The obtained experimental data were evaluated within the framework of a parametric statistical approach based on a quadratic polynomial regression model. The mathematical modeling used in this context is structured based on the general formulation presented below:

In this model, the dependent variable Yₖ reflects the TAS values ​​of the extract samples. The terms X_i are modeled as coded independent variables representing parameters such as extraction temperature, processing time, and ethanol/water ratio, respectively. The β₀ coefficient included in the equation serves as the constant term associated with the center point of the experimental setup.

The statistical validity and reliability of the created model were comprehensively evaluated using basic statistical indicators such as the coefficient of determination (R²), analysis of variance (ANOVA), and the significance level (p-value). Derivatives of the model were taken to determine the optimal response conditions. Critical points were calculated, and its solution capacity was tested analytically. Additionally, three-dimensional response surface plots were created to more clearly visualize both the individual and mutual effects of the independent variables. The findings were interpreted from a scientific perspective using these plots and subjected to detailed analysis.

Artificial neural network-genetic algorithm (ANN-GA)

In this study, the Artificial Neural Network (ANN) method was chosen as part of the predictive modeling process. The input variables were the extraction temperature, application time, and ethanol/water ratio. The output variable was the TAS value. To increase the model’s predictive accuracy and test its generalization ability, the entire dataset was randomly divided into 80% training, 10% validation, and 10% test subsets and analyzed. The training process was based on the Levenberg–Marquardt (LM) algorithm, which is known for its high computational speed and low error generation.

To optimize the network’s architectural structure, different topologies were created and systematically tested, with the number of neurons in the hidden layer ranging from 1 to 20. The learning rate and momentum coefficient were fixed at 0.5 in all architectural experiments. The maximum iteration number during model training was 500, the validation stopping criterion was 50, and the error tolerance was 1 × 10⁻⁵. Each topology was run with 1000 independent replicates, and the resulting data were evaluated using comparative performance metrics.

Two key statistical indicators were used to quantitatively assess the model’s predictive power: Mean Square Error (MSE) and Mean Absolute Percentage Error (MAPE). These error metrics were calculated according to the following formulas:

1

2

In this study, the term e_i represents the experimentally obtained observation values ​​in the formula used to calculate the error levels of the modeling process. p_i represents the predicted results of the artificial intelligence-based model. n represents the total number of data analyzed.

In the computational optimization phase to determine the optimal conditions, the Genetic Algorithm (GA) was chosen as the analysis algorithm. The overall performance of the algorithm was tested using different population sizes. The effect of this parameter on the accuracy of the results was evaluated comparatively. A probability-based roulette wheel selection strategy was applied during the selection process of new individuals. A single-point crossover mechanism was used to preserve genetic diversity and expand the search space. The iterative behavior of the algorithm was visually monitored using convergence plots obtained throughout the process. These graphical analyses detailed the contribution of the number of iterations to optimization efficiency. In addition, each GA run was conducted with at least 45 independent repetitions to increase the probability of reaching the global optimum solution and to ensure the overall stability and reliability of the algorithm.

Extraction for bioactivity

In this study, extracts were prepared under conditions determined by extraction optimization, and the biological activities of the resulting products were evaluated. The optimization process using RSM revealed the most favorable conditions as a temperature of 41.728 °C, a processing time of 39.795 min, and a 35.416% ethanol/water ratio. In contrast, the ANN-GA-based hybrid optimization model revealed the optimal values ​​as a temperature of 44.247 °C, a processing time of 45.826 min, and a 34.155% ethanol/water ratio. The conditions closest to the predictions obtained from both methods were applied using an ultrasonic bath device (WIGGENS UA22MFDN, Germany) with a frequency of 40 kHz, 100% power, and a capacity of 400 W. Extracts prepared under these parameters were analyzed in biological activity tests conducted within the scope of the study.

Antioxidant Activity Tests

Total antioxidant and oxidant analysis

The total antioxidant capacity of mushroom extracts obtained under optimized extraction conditions was measured using the TAS Assay Kit (Rel Assay Diagnostics, Mega Tıp, Gaziantep, Turkey; Product Code: RL0017). Results were expressed in mmol Trolox equivalents/L¹³. Total oxidant levels were determined using the TOS Assay Kit (Rel Assay Diagnostics, Mega Tıp, Gaziantep, Turkey; Product Code: RL0024), and the results were expressed as µmol hydrogen peroxide equivalents/L¹⁴. The Oxidative Stress Index (OSI) was calculated by dividing the TOS value by the TAS value and converting this result to a percentage¹⁵.

DPPH free radical scavenging activity

The antioxidant capacities of mushroom extracts obtained under optimized extraction conditions were evaluated using the DPPH (2,2-diphenyl-1-picrylhydrazyl) radical scavenging method. For this purpose, stock solutions of each extract were prepared at a concentration of 1 mg/mL in dimethylsulfoxide (DMSO, Sigma-Aldrich, St. Louis, MO, USA). 1 mL of the prepared stock solutions was taken and homogeneously mixed with a 4 mL mixture containing 160 µL of 0.267 mM DPPH solution (Sigma-Aldrich, St. Louis, MO, USA) and 0.004% methanol (Merck, Darmstadt, Germany). The mixtures were incubated in the dark at room temperature for 30 min. Following incubation, absorbance values ​​were measured using a UV-Vis spectrophotometer at a wavelength of 517 nm. Results were calculated and reported as mg Trolox equivalent/g extract (mg TE/g)¹⁶.

Ferric Reducing antioxidant power assay

The ferric reducing power of mushroom extracts obtained under optimized extraction conditions was assessed using the FRAP (Ferric Reducing Antioxidant Power) method. In this context, 100 µL of stock solution was taken from each extract and homogeneously mixed with 2 mL of FRAP reagent. FRAP reagent was prepared by mixing 20 mM FeCl₃·6 H₂O (Iron(III) chloride hexahydrate, Sigma-Aldrich, St. Louis, MO, USA) and 10 mM 2,4,6-tris(2-pyridyl)-s-triazine (TPTZ, Sigma-Aldrich, St. Louis, MO, USA) solutions dissolved in 300 mM acetate buffer (pH 3.6) [Acetic acid and Sodium acetate, Merck, Darmstadt, Germany], 40 mM HCl at a ratio of 10:1:1. The mixtures were incubated at 37 °C for 4 min and then absorbance measurements were carried out at a wavelength of 593 nm using a UV-Vis spectrophotometer. Analysis results were calculated and reported in mg Trolox equivalent/g extract (mg TE/g)¹⁶.

Anticholinesterase activity tests

Anticholinesterase activities of mushroom extracts obtained under optimized extraction conditions were determined using the colorimetric method developed by Ellman and colleagues¹⁷. Galantamine hydrobromide (Sigma-Aldrich, St. Louis, MO, USA) was used as a reference inhibitor for comparison in the analyses. For this purpose, stock solutions were prepared from mushroom extracts at concentrations ranging from 200 to 3.125 µg/mL. In the experimental setup, 130 µL of phosphate buffer (Merck, Darmstadt, Germany), at a concentration of 0.1 M and adjusted to pH 8.0, was first added to each microplate well. Then, 10 µL of extract solution and 20 µL of acetylcholinesterase (AChE, Type VI-S, EC 3.1.1.7) or butyrylcholinesterase (BChE, EC 3.1.1.8) enzyme solutions (Sigma-Aldrich, St. Louis, MO, USA) were added. The mixtures were incubated at 25 °C in the dark for 10 min. Then, 20 µL of DTNB [5,5’-dithiobis-(2-nitrobenzoic acid), Sigma-Aldrich, St. Louis, MO, USA] solution and 20 µL of substrate (acetylcholine iodide or butyrylcholine iodide, Sigma-Aldrich, St. Louis, MO, USA) were added to each well to initiate the reaction. Absorbance values ​​were measured with a microplate reader at 412 nm wavelength. Using the obtained data, enzyme inhibition percentages were calculated, and half-maximal inhibitory concentration (IC₅₀) values ​​were determined in µg/mL.

Antiproliferative activity tests

The antiproliferative activities of the optimized mushroom extracts were evaluated on three different human cancer cell lines (A549 (lung adenocarcinoma; ATCC CCL-185), MCF-7 (breast adenocarcinoma; ATCC HTB-22), and DU-145 (prostate carcinoma; ATCC HTB-81)). The cell lines used were obtained from the American Type Culture Collection (ATCC, Manassas, VA, USA). Experimental solutions of the extracts were prepared at concentrations of 25, 50, 100, and 200 µg/mL. When the cells reached 70–80% confluence, they were detached from the culture surface with 3.0 mL of Trypsin-EDTA solution (Sigma-Aldrich, St. Louis, MO, USA; Cat. No. T3924) and transferred to appropriate plates. They were incubated for 24 h to ensure cell adhesion to the surface. Then, the extract solutions were added to the cultures and a 24-hour incubation period was applied. After the incubation was completed, the culture medium was removed and MTT solution [3-(4,5-dimethylthiazol-2-yl)−2,5-diphenyltetrazolium bromide; Sigma-Aldrich, St. Louis, MO, USA; Cat. No. M5655] was added to each well at a concentration of 1 mg/mL. The cells were incubated at 37 °C until purple formazan crystals formed. These crystals were then dissolved with Dimethylsulfoxide (DMSO) (Sigma-Aldrich, St. Louis, MO, USA; Cat. No. D8418). DMSO was also used as a negative control. Equal volumes of DMSO were applied to all groups. Absorbance values ​​for cell viability were measured using an Epoch spectrophotometer (BioTek Instruments, Winooski, VT, USA) at a wavelength of 570 nm¹⁶.

Phenolic analysis

The phenolic compound profile of mushroom extracts obtained under optimized extraction conditions was analyzed using a Shimadzu LC–MS/MS-8030 system (Shimadzu Corporation, Kyoto, Japan). Chromatographic separation was carried out on an Inertsil ODS-4 C18 analytical column (2.1 × 50 mm, 2 μm; GL Sciences, Tokyo, Japan), and the column temperature was maintained at 40 °C using a column oven (CTO-10ASvp). The LC system was equipped with binary pumps (LC-20ADXR) operating in binary gradient mode, with a total flow rate of 0.40 mL/min. The autosampler model was SIL-20ACXR, and the maximum system pressure was set to 660 bar. The mobile phase consisted of solvent A (ultrapure water containing 0.1% formic acid) and solvent B (methanol containing 0.1% formic acid). The gradient elution program was applied as follows: 5% B at the initial condition, increased to 95% B at 4.0 min and maintained until 7.0 min, followed by a return to 5% B at 7.01 min, with a total analysis time of 12 min. The injection volume was 2 µL. For sample preparation, 0.5 g of dried mushroom residue was weighed into 15 mL centrifuge tubes, and 10 mL of methanol was added. After brief vortex mixing, the samples were subjected to ultrasonic extraction for 30 min and subsequently centrifuged at 6000 rpm for 15 min. The supernatant was collected, and the extraction procedure was repeated three times. The combined supernatants were evaporated to dryness and reconstituted in methanol to a final volume of 10 mL. Prior to LC–MS/MS analysis, all samples were filtered through 0.45 μm syringe filters and transferred into vials. A total of 24 phenolic compounds were quantitatively analyzed using authentic analytical standards. The investigated compounds included acetohydroxamic acid, catechin hydrate, vanillic acid, syringic acid, thymoquinone, resveratrol, myricetin, and kaempferol, which were analyzed in positive ionization mode, as well as fumaric acid, gallic acid, protocatechuic acid, 4-hydroxybenzoic acid, caffeic acid, salicylic acid, phloridzin dihydrate, 2-hydroxycinnamic acid, oleuropein, 2-hydroxy-1,4-naphthoquinone, naringenin, silymarin, quercetin, luteolin, alizarin, and curcumin, which were analyzed in negative ionization mode. Quantification of individual phenolic compounds was performed using external calibration curves constructed from standard solutions prepared at different concentration levels. Calibration curves were generated using linear regression analysis. The limits of detection (LOD) and limits of quantification (LOQ) for each compound were determined and are reported in the phenolic composition table.

Statistical analysis

Statistical analysis of the experimental findings obtained in this study was performed using SPSS 21.0 software (IBM Corp., Armonk, NY, USA). An Independent Samples t-test was used to test the significance of the difference between groups when comparing mean values ​​between two independent groups. For comparisons involving three or more groups, one-way analysis of variance (One-Way ANOVA) was used to accurately analyze the multigroup structure of the data set. In the ANOVA tests conducted, a value of α = 0.05 was used as the threshold for statistical significance. If statistically significant differences were detected between groups, advanced analyses were conducted using Duncan’s multiple comparison test to determine which groups accounted for these differences.

Results and discussions

Extract optimization

This study investigated the effects of three main extraction parameters on Total Antioxidant Status (TAS). The independent variables examined were extraction temperature (30, 45, and 60 °C), application time (30, 45, and 60 min), and the ethanol/water ratio (0%, 50%, and 100%), defined as the solvent composition. Data from the measurements of TAS values ​​obtained from extraction processes applied at different levels of combination of these variables are presented in detail in Table 1. The collected experimental data were evaluated using both comparative statistical tests and multivariate analysis techniques.

Table 1.

TAS values ​​obtained for extract optimization of Lentinula edodes.

Experiment number	Extraction temperature (°C)	Extraction time (min)	Ethanol/water ratio (%)	TAS (mmol/L)
1	30	30	0	5.671 ± 0.019k
2	30	45	0	5.970 ± 0.024n
3	30	60	0	4.656 ± 0.030d
4	30	30	50	5.939 ± 0.021n
5	30	45	50	6.144 ± 0.041p
6	30	60	50	4.858 ± 0.023f
7	30	30	100	5.759 ± 0.022l
8	30	45	100	6.076 ± 0.020o
9	30	60	100	4.737 ± 0.032e
10	45	30	0	5.944 ± 0.025n
11	45	45	0	6.338 ± 0.024r
12	45	60	0	4.934 ± 0.019g
13	45	30	50	6.069 ± 0.022o
14	45	45	50	6.541 ± 0.014t
15	45	60	50	5.029 ± 0.021h
16	45	30	100	5.954 ± 0.034n
17	45	45	100	6.444 ± 0.020s
18	45	60	100	4.790 ± 0.019e
19	60	30	0	5.340 ± 0.014i
20	60	45	0	5.659 ± 0.030k
21	60	60	0	4.286 ± 0.026a
22	60	30	50	5.554 ± 0.029j
23	60	45	50	5.856 ± 0.017m
24	60	60	50	4.449 ± 0.013c
25	60	30	100	5.352 ± 0.023i
26	60	45	100	5.749 ± 0.022l
27	60	60	100	4.374 ± 0.022b

^aMeans having the different superscript letter(s) in the same column are significantly different (p < 0.05) according to Duncan’s multiple range test.

In this study, the effects of different extraction conditions on TAS were evaluated, and the results revealed that process parameters directly affect the effectiveness of antioxidant compounds. In particular, the highest TAS values ​​of 6.541 ± 0.014 mmol/L were determined in the extraction process performed using a 50% ethanol/water mixture at 45 °C, a time of 45 min, and a temperature range of 45 °C. This combination represents the optimal polarity and temperature range that maximizes the dissolution of phenolic and other antioxidant compounds. When the extraction time was extended to 60 min under the same temperature conditions and the solvent was completely converted to ethanol, the TAS value (4.790 ± 0.019 mmol/L) obtained suggests that the interaction of time and solvent ratio may limit biological activity. A general decreasing trend was noted when the temperature was increased to 60 °C. Values ​​measured at this temperature did not exceed 5.856 mmol/L even after 45 min, and values ​​as low as 4.286 ± 0.026 mmol/L were obtained in extracts containing 0% ethanol after 60 min. This supports the notion that high temperatures can disrupt the structural integrity of phenolic compounds, leading to thermal degradation, resulting in significant decreases in TAS. Furthermore, increasing the extraction time to 60 min had a suppressive effect on TAS at all temperature levels, with this negative effect most pronounced at 60 °C. On the other hand, TAS values ​​obtained at 30 °C ranged from 4.656 to 6.144 mmol/L, indicating that a slower but more stable extraction process is possible at lower temperatures. Two different optimization approaches were applied in this study based on the experimental findings. During the modeling phase, conducted using Response Surface Methodology (RSM), four separate regression models were compared and analyzed: linear, two-factor interaction (2FI), quadratic, and cubic. Statistical evaluations revealed an R² value of 0.988, and the quadratic model stood out as the best fit for the available data set.

The findings demonstrate that the model can establish statistically significant and reliable relationships between the independent variables and the response variable. They also support the model’s accurate representation of the patterns in the data set. In this context, it was concluded that the quadratic model offers an effective approach in terms of predictive accuracy and statistical reliability in applications aimed at optimizing extraction parameters. The mathematical expression for the quadratic regression model developed for the prediction of TAS data for L. edodes is presented below:

In the quadratic regression model, the independent variables defined as X₁ represent the extraction temperature, X₂ represents the application time, and X₃ represents the ethanol/water ratio. The effects of these variables on TAS values ​​obtained from L. edodes were comprehensively analyzed both at the main (single) factor level and in the context of pairwise interactions between variables. The outputs from the modeling process were converted into three-dimensional response surface plots (Fig. 1) to visualize and more thoroughly interpret multivariate interactions. These plots offer significant advantages in visually representing the individual and reciprocal effects of independent variables on TAS. They also serve as a guide for determining optimal parameter combinations.

Fig. 1.

Response surface plots.

In the second phase of the research, the ANN method was chosen for the predictive modeling process based on experimental data. After extensive performance analyses conducted on different neural network structures, the architecture that provided the highest predictive accuracy was selected. To further increase the optimization level, this structure was integrated with a GA to create a hybrid system. In this context, an ANN model with six neurons in a single hidden layer in a 3–6-1 structure was identified as the most suitable architecture based on statistical performance criteria. The model’s predictive power was determined as Mean Square Error (MSE): 0.001, Mean Absolute Percentage Error (MAPE): 0.360%, and Correlation Coefficient (R): 0.998. These statistical results clearly demonstrated the high accuracy and reliability of the created ANN model.

In the GA-assisted optimization phase, optimal solutions were sought based on the most successful prediction outputs provided by the ANN. During this process, population size, one of the primary factors affecting the algorithm’s success, was tested at various levels and subjected to systematic comparisons. As a result of the evaluations, the optimal number of individuals was determined to be 10. The data of the convergence curve (Fig. 2) show that the algorithm reached a balanced solution point after approximately 5 iterations and the optimization process was completed successfully.

Fig. 2.

Convergence Graph.

Antioxidant activity

Thanks to their antioxidant properties, mushrooms contribute to cell protection by reducing oxidative damage caused by free radicals, playing an important role in both nutrition and health practices¹⁸. In this study, the antioxidant capacities of extracts obtained from L. edodes under optimal extraction conditions were examined, and the findings are presented in Table 2.

Table 2.

Antioxidant values of Lentinula edodes.

Parameters	ANN-GA extract values	RSM extract values
TAS (mmol/L)	6.612 ± 0.016b	6.506 ± 0.011a
TOS (µmol/L)	4.167 ± 0.033a	4.835 ± 0.027b
FRAP (mg Trolox Equi/g)	176.25 ± 1.34b	159.10 ± 1.64a
DPPH (mg Trolox Equi/g)	133.00 ± 1.09b	122.20 ± 1.13a
OSI (TOS/(TAS*10))	0.063 ± 0.001a	0.074 ± 0.001b

In this study, the antioxidant activities of L. edodes extracts were determined using RSM and ANN-GA based optimization approaches. It was observed that the extracts obtained under ANN-GA conditions exhibited higher antioxidant activity. The higher TAS and FRAP values, and lower TOS and OSI values ​​in the ANN-GA model indicate that artificial intelligence-based optimization increases biological activity by more accurately modeling the complex interactions between parameters. These results reveal that the ANN-GA approach improves extraction efficiency compared to the classical RSM model not only in terms of quantity but also in terms of functional quality. There are numerous studies in the literature on DPPH, FRAP, and ABTS activities of L. edodes, and these studies reported high antioxidant capacities depending on the extraction method and solvent system^9,19–22. However, no direct findings were found regarding the TAS, TOS and OSI values ​​of this species. In this respect, our study provides an original contribution to the literature on L. edodes. On the other hand, studies on different wild mushroom species enable comparative evaluation of our findings. In the literature, it has been reported that Paralepista flaccida has a TAS value of 4.054 mmol/L, a TOS value of 10.352 µmol/L and an OSI value of 0.255; while Lactarius deliciosus has a TAS value of 7.468 mmol/L, a TOS value of 13.161 µmol/L and an OSI value of 0.176^23,24. Additionally, the TAS value for Hericium erinaceus was reported as 5.426 mmol/L, TOS as 6.621 µmol/L, and OSI as 0.122; and for Cantharellus cibarius, TAS was reported as 5.511 mmol/L, TOS as 7.289 µmol/L, and OSI as 0.132^25,26. Considering these data, the higher TAS value and lower TOS and OSI values ​​obtained for L. edodes compared to many wild mushrooms in our study support the species’ strong antioxidant potential and its ability to maintain oxidative stress balance. In conclusion, the L. edodes extracts obtained under ANN-GA-optimized extraction conditions represent a novel methodological contribution, as this study is the first to report TAS, TOS, and OSI parameters for this species while demonstrating the effectiveness of artificial intelligence-based optimization in enhancing biological activity. In this respect, the study reveals that artificial intelligence-based optimization methods are a more reliable and powerful alternative compared to classical statistical models in biological activity studies.

Anticholinesterase activity

Bioactive compounds contained in mushrooms may contribute to neuroprotective effects by inhibiting the activity of acetylcholinesterase and butyrylcholinesterase enzymes²⁷. In this study, the anticholinesterase activities of extracts obtained from L. edodes under optimal conditions were investigated, and the calculated IC₅₀ values ​​are presented in Table 3.

Table 3.

Anticholinesterase activity of Lentinula edodes.

Sample	AChE µg/mL	BChE µg/mL
Galantamine	6.79 ± 0.16	15.49 ± 0.25
ANN-GA extract	72.47 ± 0.96	132.13 ± 0.85
RSM extract	88.91 ± 0.99	143.08 ± 1.18

In our study, optimized L. edodes extracts were found to exhibit significant inhibitory activity against both acetylcholinesterase (AChE) and butyrylcholinesterase (BChE). Galantamine, used as a positive control, exhibited very low IC₅₀ values ​​for both enzymes. However, it is noteworthy that the extracts obtained with the optimization strategies presented different enzyme inhibition profiles. The extract optimized with the ANN-GA approach showed 72.47 and 132.13 µg/mL values ​​against AChE and BChE enzymes, respectively. In contrast, these values ​​were found to be 88.91 and 143.08 µg/mL, respectively, in extracts obtained with the traditional statistical optimization method, RSM. These results indicated that ANN-GA-based optimization was more effective than RSM in inhibiting both AChE and BChE. Literature has reported that L. edodes exhibits inhibitory properties against acetylcholinesterase activity²⁸. This supports the conclusion that the AChE inhibition observed in our study stems from the mushroom’s natural biochemical potential. However, in our current study, we observed that the activity levels in the extracts obtained with the application of different optimization methods differed significantly. Therefore, the ANN-GA approach was found to be more successful in enhancing enzyme inhibition. The results obtained in our study indicate that artificial intelligence-assisted optimization techniques may be more advantageous than classical methods in maximizing biological activity. In conclusion, although the anticholinesterase activity of L. edodes extracts remained at lower levels compared to a strong standard inhibitor such as galantamine, it was demonstrated that this activity could be significantly improved with the application of optimization techniques.

Antiproliferative activity

Mushrooms, through their bioactive compounds, can contribute to limiting tumor development by suppressing the proliferation of cancer cells^29,30. In this study, the biological activities of L. edodes extract obtained under optimal conditions on A549, MCF-7, and DU-145 cell lines were investigated, and the results are shown in Fig. 3.

Fig. 3.

Antiproliferative activity of Lentinula edodes optimized extract. (Control group: Cells were cultured in standard medium only and no chemical treatments were applied. DMSO group: DMSO was added to the cell medium as a solvent, but no extract was applied. Extract treatment groups: Cells were treated with extract at different concentrations of 25, 50, 100, and 200 µg/mL.).

In our study, optimized extracts of L. edodes were determined to exhibit dose-dependent antiproliferative activity on three different human cancer cell lines (A549, MCF-7, and DU-145). The high preservation of cell viability in the control and DMSO groups confirmed that the solvent did not cause toxicity in the experimental system. A steady decrease in optical density was observed with increasing concentration in the extract applications between 25 and 200 µg/mL. This trend was particularly pronounced at 100 and 200 µg/mL levels. The optimized extracts of the mushroom generally exhibited the highest activity against A549, followed by MCF-7 and DU-145, respectively. Extracts optimized with ANN-GA exhibited generally stronger antiproliferative activity compared to those optimized with RSM. This difference can be explained by the fact that ANN-GA more accurately models multivariate and nonlinear relationships, optimizing the affinity of intermediate-polar/phenolic compounds and sterol/tirpenoid derivatives associated with anticancer activity. Furthermore, AI-assisted optimization is thought to better capture the chemical composition window that maximizes biological activity compared to conventional RSM. High cytotoxicity of L. edodes in the MCF-7 breast adenocarcinoma line has been previously reported in the literature³¹. In a separate study, potent cytotoxicity of L. edodes in the T47D line, another model of breast cancer³². Furthermore, the observations of significant cytotoxic effects in HepG2 (liver), HCT116 (colorectal), and PANC-1 (pancreas) lines support the broad-spectrum antiproliferative potential of L. edodes against different tissue origins³³. The sensitivity trend observed in MCF-7 is consistent with these reports. The further enhancement of the effect with ANN-GA suggests that optimization enriches the balance of components that enhance the therapeutic signal. In conclusion, optimized L. edodes extracts exhibited dose-dependent antiproliferative effects in three different cancer cell lines. Furthermore, extracts obtained with ANN-GA optimization were found to exhibit higher cytotoxic potential compared to the RSM method.

Phenolic contents

Thanks to their rich phenolic compound content, mushrooms are among the valuable natural resources prominent in biological and pharmacological research³⁴. In this study, the phenolic compound profile of the extract obtained from L. edodes under optimum conditions was determined by LC–MS/MS analysis, and the results are presented in Table 4.

Table 4.

Phenolic contents of Lentinula edodes.

Phenolic compounds	LOD	LOQ	ANN-GA extract (mg/kg)	RSM extract (mg/kg)
Gallic acid	2.17	7.24	7421.54 ± 2.47	7011.50 ± 1.70
4-hydroxybenzoic acid	7.51	25.02	2650.04 ± 1.88	2516.21 ± 1.52
Caffeic acid	2.19	7.29	1962.43 ± 2.06	1576.94 ± 1.17
Quercetin	4.18	13.93	583.89 ± 1.33	511.97 ± 0.72
2-hydoxycinamic acid	1.37	4.56	1439.55 ± 1.48	1496.08 ± 2.12
Vanillic acid	24.74	82.46	1039.35 ± 1.13	959.67 ± 3.49
Protocatechuic acid	2.21	7.37	3469.88 ± 1.61	2881.83 ± 1.11
Catechinhyrate	5.47	18.22	345.75 ± 1.60	166.53 ± 2.04

In our study, the amounts of phenolic compounds in the optimized extracts of L. edodes varied depending on the optimization method used. Generally, higher concentrations of most phenolic compounds were detected in extracts obtained with the ANN-GA approach compared to extracts optimized with the RSM method. This suggests that AI-assisted optimization strategies can increase the extraction efficiency of phenolic metabolites. Among the compounds, gallic acid stood out as the phenolic with the highest concentration and was detected in ANN-GA extracts. Similarly, the amounts of protocatechuic acid and caffeic acid were significantly increased with ANN-GA optimization. In contrast, the level of 2-hydroxycinnamic acid was found to be higher in extracts optimized with RSM. This suggests that each phenolic compound responds differently to extraction conditions, and ANN-GA does not provide absolute superiority for all compounds. The presence of phenolic compounds such as p-hydroxybenzoic acid, trans-cinnamic acid, protocatechuic acid, caffeic acid, ferulic acid, xomaric acid, vanillic acid, cinnamic acid, 3,4-dimethoxybenzoic acid, syringic acid, chlorogenic acid, rutin, 2-hydroxycinnamic acid, and catechin has been reported in the literature in Lentinula edodes^35–38. Recent studies continue to support the association between phenolic profiles and antioxidant capacity in L. edodes, including investigations on spent mushroom substrates and processing/biotechnological treatments that modulate phenolic fractions and antioxidant performance. These reports frequently highlight phenolic acids such as gallic, protocatechuic, and vanillic acids among abundant contributors to antioxidant-related endpoints^39–41. The compounds identified in L. edodes in our study are consistent with these literature data, and it appears that the ANN-GA approach allows these natural compounds to be obtained at higher concentrations. Therefore, ANN-GA-based optimization not only increased the extraction efficiency but also enriched the phenolic profile, which is associated with biological activity. This once again demonstrates that artificial intelligence-based optimization is a powerful tool in increasing the therapeutic and functional food potential of natural products. Although individual phenolic compounds were quantified, attributing the observed multifunctional bioactivities to a single “responsible” phenolic is not scientifically definitive without bioassay-guided fractionation and testing of isolated constituents. In complex extracts, antioxidant- and enzyme-related endpoints typically reflect additive or synergistic effects among multiple phenolic acids and flavonoids, and their relative abundances can vary with extraction conditions^42,43. Therefore, the present LC–MS/MS dataset should be interpreted as a quantitative profile supporting extract-level activity, while pinpointing specific causative molecules remains a priority for future work using fractionation/isolation strategies.

Conclusion

Our study demonstrated the differences obtained by optimizing the biological activities of L. edodes extracts using classical statistical methods (RSM) and artificial intelligence-based hybrid approaches (ANN-GA). Extracts optimized with the ANN-GA method were found to have higher antioxidant capacity (TAS, FRAP, DPPH), lower oxidative stress parameters (TOS, OSI), stronger anticholinesterase activity, and more pronounced antiproliferative effects in three different cancer cell lines compared to RSM. Furthermore, in phenolic compound analyses, ANN-GA optimization yielded higher levels of biologically critical phenols, particularly gallic acid, caffeic acid, and protocatechuic acid. These results demonstrate that AI-assisted optimization not only increases extraction efficiency but also enhances the biological quality and therapeutic value of the resulting extracts. The ANN-GA approach overcomes the limitations of classical methods, accurately modeling the complex interactions between parameters and maximizing biological activity. In this context, our study offers new opportunities for the development of functional foods, the extraction of natural antioxidant and neuroprotective agents, and supportive anti-cancer treatment approaches. Future research will test ANN-GA-based optimization not only with fungal species but also with bioactive compounds derived from different plants and algae, paving the way for large-scale applications in the biotechnology, pharmaceutical, and food industries. Furthermore, in vivo and clinical studies will further demonstrate the reliability and effectiveness of this optimization strategy. Therefore, this study demonstrates that AI-assisted bioprocess optimization can be an indispensable tool for achieving high accuracy and high biological efficacy in natural product research.

Author contributions

The authors confirm contribution to the paper as follows: study conception and design: V.T.K., A.G., M.S., I.A.; data collection: V.T.K., A.G., M.S., I.A. analysis and interpretation of results: V.T.K., A.G., M.S., I.A. draft manuscript preparation: V.T.K., A.G., M.S., I.A. All authors reviewed the results and approved the final version of the manuscript.

Data availability

The datasets used and/or analysed during the current study available from the corresponding author on reasonable request.

Declarations

Competing interests

The authors declare no competing interests.