Intelligent predictive neural network analysis on microbes in bioconvection flow of hybrid nanoliquid under diffusion thermo and thermo diffusion effects Authors: Humaira Kanwal, Shouki A Ebad, Fuad Alsarari, Abed Saif Ahmed Alghawli, Dhouha Choukaier, Ruzimurod Abiyev Published: 2026 ABSTRACT AbstractThe model optimizes microfluidic and lab-on-a-chip systems by predicting gyrotactic microbe-driven mixing and transport in the presence of linked thermal and concentration gradients. It aids in the design and operation of bioreactors and wastewater treatment units by simulating microbe dispersal, nutrient transport, and nanoparticle-assisted mass transfer for better biodegradation. It aids in targeted medication administration and biomedical experiments by predicting microbe and nanoparticle behavior in thermally heterogeneous solutions. Simulating microbial mobility in thermo-stratified, nanoparticle-laden fluids helps to inform environmental monitoring and cleanup techniques. It improves heat-exchanger and cooling technologies by utilizing bioconvective enhancement and Marangoni surface-tension phenomena. It optimizes biofuel and fermentation processes by improving microbe aggregation and stability in hybrid nanofluids. It aids in the creation of biosensors and advanced materials processing, where thermo-diffusive transport and microbial motility interact to effect performance. Artificial neural network (ANN) modeling optimized with the Bayesian Regularization technique is used in this study. The method avoids over-fitting and improves prediction accuracy. The outcomes of thermo-diffusion and diffusion-thermo properties on bioconvection flow of a hybrid nanofluid around a disk with gyrotactic microbes are investigated using homotopy analysis modeling. The temperature and concentration profiles become more pronounced as the Soret and Dufour numbers increase.Graphical Abstract Keywords: Gyrotactic microorganisms, Hybrid nanofluid, Bioconvection flow, Artificial neural network, Surface tension gradient Introduction Artificial neural networks and regression-based techniques are now common machine learning (ML) technologies that may effectively replace traditional computational fluid dynamics (CFD) simulation. Conventional CFD techniques can need a substantial amount of computational time and resources, especially for complex heat transfer and fluid flow models. However, when trained using high-fidelity datasets like as those from direct numerical simulation, ML models may accurately predict critical flow variables like temperature, pressure, and velocity at a far reduced processing cost. Additionally, ML-driven approaches enhance turbulence prediction by learning intricate nonlinear flow features that are challenging to capture with conventional turbulence models. By outcome important components that maximize efficiency and minimize energy wounded, ML augments fluid flow investigation in uses ranging from industrial cooling systems to aerodynamic optimization. Research and engineering uses of fluid dynamics are revolutionized by this integration of machine learning with CFD, which bridges the gap among correctness and computing cost. Abdal et al. [1] examine the machine learning investigation utilizing linear regression for Casson fluid flow in 3D using CCHFM (Cattaneo-Christov heat flux model). Pemmaraju et al. [2] inspect the computational study of ternary EMHD nanofluid with heat source using ML. Shaaban et al. [3] use ML based on the Bayesian regularized technique to investigate enzyme-based biosensor uses for the temperature bioconvection flow of the BNF (Boger nanofluid) when microbes that are present. Mahboobtosi et al. [4] examined MHD convection that is unsteady in micropolar fluids with Lorentz and suction impacts is modeled using a mixed physics–machine learning approach. Abbas et al. [5] inspected the convective flow of a THNF (trihybrid nanofluid) with heat radiation using an ANN enhanced by the Bayesian-regularized technique. Ahmad et al. [6] inspect the generation of a ML technique for penta-HNF flow simulation to improve the heat handover in a magnetic field. Abbas et al. [7] examine Levenberg–Marquardt back-propagation neural networks’ computational power to investigate the bioconvection flow of a THNF having oxytactic germs. Hussein et al. [8] studied the evolution of ANN for the Darcy-Forchheimer flow of HNF. Abbas et al. [9] conducted ML research for thermal improvement in Boger HNF comprising microbes. Okasha et al. [10] investigate the characteristics of SB (Stephan blowing) and heat radiation on nanofluid with LTNE using a BBO (Bayesian-Regularization optimizer) deep neural network. Abbas et al. [11] analyse the thermodynamic control of heat transportation in THNF containing motile microbes using a Bayesian-regularized neural network. Shaaban et al. [12] used an INN (intelligent neural network) analysis to study the impact of Stephan blowing on the flow of BNF with thermophoresis and Brownian motion. The transfer of species propelled by a temperature gradient, when thermal and concentration effects are closely related, is referred to as thermo-diffusion. The Soret and Dufour impacts become important in flows with discernible density fluctuations, resulting in simultaneous mass and heat transfer. Non-uniform species concentration during chemical reactions, especially in binary fluid systems, is the cause of this coupling. The significance of these phenomena in energy storage, catalytic reactors, geothermal reservoirs, nuclear waste management, drying technologies, and thermal insulation encourages further research into coupled thermal-solutal transport in porous and reactive media. Sivanandam and Alqurashi [13] examine the outcomes of radiation, viscosity dissipation, Dufour and Soret impacts, and MHD spinning nanoliquid flow with Joule heating and slip. Srinivasa et al. [14] examine Joule heating, Dufour, and Soret significance all have an impact on HNF flow via a moving thin needle in a permeable surface with heat radiation. Hamida et al. [15] examined a thermal engineering examination of the MHD Maxwell nanofluid flow driven by solar radiation with gyrotactic microorganisms and Soret-Dufour influences. Salahuddin et al. [16] explored the importance of the impacts of Soret and Dufour as well as solutal and thermal study for MHD tangent hyperbolic liquid stagnation point flow with double stratification. Multiple nanoparticles are distributed across a single base fluid in hybrid nanofluids, a sophisticated type of nanofluids. These fluids perform better thermally than conventional and binary nanofluids because of their combined physicochemical characteristics. Hybrid nanofluids are appropriate for a variety of applications, such as industrial production, thermal management, and biomedical systems, because different nanoparticles work in concert to boost thermal conductivity and improve heat transport efficiency. Hafeez et al. [17] scrutinized melting transfer of heat affects the radiative hybrid nanofluid’s MHD flow over a wedge. Mumtaz et al. [18] inspected the evaluation of the temperature profile of a HNF in MHD flow via a surface with radiative and Joule heating impacts. Anwar et al. [19] analyze the study of heat transportation properties in HNF with different heat radiations and viscosities. Malik et al. [20] inspected a combined ANN and numerical study of the features of radiation and MHD on flow of HNF and heat transportation in permeable cylinders. Rehman et al. [21] scrutinized the squeezing between two porous surfaces, a HNF based on blood using mathematical modeling while taking viscous dissipation and MHD into account. Aoudia et al. [22] examined the improved thermal evaluation of THNF with Hall current and gyrotactic bacteria: enzyme biosensors and bacteria-powered micromixers use. Kanwal et al. [23] researched into the thermodynamic case evaluation of the LTNE impacts on temperature radiative THNF with Stephan blowing effects. Okasha et al. [24] examine the electrophoresis particle deposition in Boger nanofluid: Marangoni convection. Mohsin et al. [25] studied the impacts of elastic deformation and heat radiation on the flow of a HNF containing gyrotactic microbes and Stefan blowing influence. Aoudia et al. [26] analysed the use of ANN to advance heat and mass transportation in HNF with gyrotactic microbes and LTNE. Abbas et al. [27] observe the use of THNF in industrial cooling systems: An examination of LTNE properties and elastic deformation. Due to its significance in agricultural, biomedical, and industrial applications such pollution removal, food processing, and pharmaceutical systems, the interaction of microorganisms with viscous fluid flow has drawn more attention. While bio-convection resulting from the upward swimming of motile microorganisms can cause flow instability and stratification, suspended microorganisms improve fluid mixing and mass transfer. For effective design, it is crucial to comprehend the interdependent effects of heat transport, magnetic fields, and bioconvection in microfluidic devices. Furthermore, developing technologies such as bio-galvanic devices, micro-reactors, and energy-related fluid transport systems depend heavily on the combined impact of nanofluid transport and microbial flow. Ashfaq et al. [28] inspected magnetohydrodynamics BLF (boundary layer flow) with mobile bacteria via porous medium under varying wall thermal conditions: a mass and heat transmission study. Khan et al. [29] scrutinized the micropolar nanofluid’s flow containing gyrotactic microbes past a sheet that is inclined in a permeable medium were designed using ANN. The CCHFM and the EMHD flow of HNF such as gyrotactic microbes was analyze by Gupta et al. [30]. Zeeshan et al. [31] investigate the computational assHNF essment of ANN are used by bioconvective gyrotactic germs to wedge the flow of magnetohydrodynamics Carreau fluid with heat radiation. Afzal et al. [32] observed the use of multi linear regression to represent microbial behavior in an Casson liquid flow under heat radiation across a stretched sheet simplifies study with a non-uniform source. This study uses the Yamada-Ota model to predict the outcomes of diffusion-thermo and thermo-diffusion on the flow of a hybrid nanofluid via a disk with gyrotactic bacteria. The constitutive equations of the flow are resolved using the Homotopy analysis technique with ANN. The back-propagation-based Bayesian Regularization approach is employed to analyze the fluid’s training performance, error distribution through histograms, and regression behavior. The study’s objective has been described as: The objective of this study is to ascertain how thermal radiation affects the hybrid nanofluid’s velocity and thermal fields. The objective of this study is to ascertain how thermal radiation affects the hybrid nanofluid’s velocity and thermal fields. The back-propagation-based Bayesian Regularization approach is employed to analyze the fluid’s training performance, error distribution through histograms, and regression behavior. The back-propagation-based Bayesian Regularization approach is employed to analyze the fluid’s training performance, error distribution through histograms, and regression behavior. To scrutinize the consequence of diffusion-thermo and thermo-diffusion properties on the boundary layer flow of HNF. To scrutinize the consequence of diffusion-thermo and thermo-diffusion properties on the boundary layer flow of HNF. To examine the impact of MHD on the bioconvection flow. To examine the impact of MHD on the bioconvection flow. Mathematical modelling Consider the diffusion-thermo and thermo-diffusion significance on the 3D flow of HNF (hybrid nanofluid) in the occurrence of microorganisms using the Yamada-Ota model. The back-propagation-based Bayesian Regularization approach is employed to analyze the fluid’s training performance, error distribution through histograms, and regression behavior. Figure 1 displays the equation of a problem employing coordinates (). In the axial, radial, and tangential directions denoted by (), the flow components are , and w. Based on the previously stated hypothesis, the stable BL (boundary layer) equations are as given [33–35]: Flow structure. Assumption Boundary conditions [33–35] Similarity transformation [33] Surface tension [33] ODEs , BCs Where Microbe difference factor bioconvection Schmidt number Marangoni convection parameter Porosity parameter Magnetic parameter, Schmidt number, Peclet number, Prandtl number, Dufour number Soret number, Marangoni number. The thermos-physical aspects of HNF are show in Tables 1 and 3. Table 1Characterizes the features of HNF [36]:PropertiesHNF Characterizes the features of HNF [36]: Table 3Outcome of Machine learning algorithm.ScenarioPerformanceGradient TrainingTestingTimeEpoch11.51×E-112.19×E-081.19×E-113.95×E-113223124.61×E-121.05×E-084.11×E-128.45×E-120917935.81×E-098.19×E-065.80×E-095.80×E-942100043.43×E-129.93×E-103.64×E-121.69×E-1220519 Outcome of Machine learning algorithm. The Nusselt number ( Nu), local density of microorganisms ( Nn), rate of mass transport ( Sh), and skin-friction coefficient ( Cf), are well-defined as follows: Hybrid nanofluid thermal conductivity model by Yamada Ota [25]: Table 2The thermos-physical features of based fluids and the NPs [37, 38].φ1+φ2=φComponents of properties k 690 8.953 4250 Cu 385 400 8933 4179 0.6071 0.05 997.1 The thermos-physical features of based fluids and the NPs [37, 38].φ1+φ2=φ Supervised Learning Algorithms (SLAs) This study’s artificial neural network is based on a feed forward multilayer perceptron (FFNN) design with nine neurons in a single hidden layer. The hidden layer used a sigmoid activation function, whereas the output layer used a pure linear (purelin) activation function. The dataset consisted of 300 samples, with 80% used for network training and the remaining 20% used to test the model’s prediction ability. Before network training, all output and input data were normalized to the range ([0,1]) to ensure numerical stability, promote convergence, and prevent variables of bigger magnitudes from dominating the learning process. To solve difficult real-world problems, MLAs have been created in a variety of fields, engineering, including science and mathematics. According to their training, MLAs are frequently divided into three major classes: Reinforcement Learning Algorithms (RLAs) Reinforcement Learning Algorithms (RLAs) Supervised Learning Algorithms (SLAs) Supervised Learning Algorithms (SLAs) Unsupervised Learning Algorithms (ULAs) Unsupervised Learning Algorithms (ULAs) By completing tasks or examples, the supervised learning algorithm (SLA) simulates how humans learn. By comparing the anticipated outputs with the known desired outputs, weight modifications are made for this work, which employs the SLA. To aid in this process, the neural network (NN) is provided with a training dataset, also known as a teaching signal. By back-propagating the error between the actual and anticipated outputs during training, the model can modify its internal weights to enhance performance and guarantee that the neural network learns the intended mapping. This section provides a concise overview of the outcomes for the suggested back-propagated Bayesian Regularization approach with neural network scheme (BBRT-NNS). Figures 2a, 3a, 4a, and 5a, which display the ideal training performance at the proper epochs for the training and testing datasets, demonstrate the NN (neural network's) amazing generalization ability over unknown data. Based on the performance trends, the network successfully learns the physics of the problem and maintains a strong prediction ability outside of the trained data range. Figures 2b, 3b, 4b, and 5b display the variations in training performance, gradient, and mean squared error for four representative cases from epochs 231, 179, 1000, and 519. These pictures show the progressive and adaptive learning behavior of the ANN and its ability to use the back-propagation Bayesian-Regularization technique to iteratively remove training errors. The correctness of the ANN model is further illustrated by the solution fitting graphs displayed in Figures 2c, 3c, 4c, and 5c. It is clear that the ANN has learned the nonlinear mapping between the input and output variables because the predicted curves closely resemble the real data. Good data-fitting qualities and low variances are the outcomes of this. Further information on the numerical stability and convergence properties of the suggested ANN model may be found in the error histograms shown in Figures 2d, 3d, 4d, and 5d. The mainstream of the error values concentrates around zero, there are most likely few departures from the predictions, indicating that the network has a high degree of accuracy and precision. Finally, Figures 2e, 3e, 4e, and 5e display the comprehensive regression investigation for the testing and training datasets. A high degree of connection between the actual and expected results is demonstrated by a regression constant (R = 1) and the regression line's nearly perfect arrangement with the target data. As demonstrated by the combined results from Figures 2e, 3e, 4e, and 5e, the back-propagation Bayesian-Regularization technique, when combined with the well-trained ANN model, offers a computational framework for modeling complex heat transportation phenomena that is exceptionally precise, stable, and reliable. Regression analysis also enhances the interpretability of the model by ensuring higher fidelity and forecast constancy when compared to experimental and mathematical reference datasets. This approach provides a number of advantages for contemporary thermal-fluid analysis, allowing engineering systems to be optimized with greater computing efficacy and physical realism. Figures 6(a&b) and 6(c&d) illustrate the relationship between φ affects f' (η)and θ(η). These graphs demonstrate how the temperature profile of the HNF increases as the principal velocity fields decrease in proportion to the volume friction of the nanoparticles. Figs. 7 (b&a) demonstrations the importance of Du on the θ. The Dufour outcome is a heat flow produced by a solutal gradient. Thermal profiles with the Dufour effect have higher temperatures than those without. Dufour influences considerably set the TBL (thermal boundary layer) by increasing the energy of boundary layer flow. The concentration profile climbs as the Soret number upsurges, as Figs. 7(c&d) demonstrate. A larger concentration profile is the outcome of augmented molar mass diffusivity, which is indicated by the increased Soret impact. Figures 8(a&b) show how the Peclet number Pe affects the concentration of microorganisms (Θ). It has been shown that the concentration of microorganisms reduces as Pe increases. Because Pe describes the relative impacts of advective and diffusive transport, a rise in Pe implies that advective transport is more effective than microbiological diffusion. As a result, microorganisms are carried away from the surface more quickly, resulting in a decrease in microbe concentration and a thinner microorganism boundary layer close to the surface. This trend suggests that enhanced advective transport predominates microbial diffusion and influences the geographic distribution of motile bacteria. Scenario 1 Scenario 2 Scenario 3 Scenario 4 (a&b) Effect of φ on f' (η). (c&d), Influence of φ on θ(η) (a&b) Effect of Du on θ(η),(c&d) Influence of Du on θ(η) (a&b) Effects of Pe on Θ(η),(c&d) Influence of Sr on Sh Figs. 8(c&d) show the importance of Sr on Sh. As the Soret number increases, so does the mass transfer rate. Figs. 9(a&b) show the importance of Du on Nu. The rate of heat transmission rises with the Dufour number. Figs. 9(c&d) show the relevance of φ on Nu. The volume friction of nanoparticles upsurges as the rate of heat transportation upsurges. The influence of Pe on Nm field 3D and2D are demonstration in Figs. 10(a&b). The local density of microbe declines as increase the values of Pe. Table 4 show a comparison of the existing study's results with those of earlier research. The most recent outcomes are reliable with previous findings. (a&b) Effects of Du on Nu,(c&d) Influence of Du on Nu (a&b) Significance of Pe on Nn Table 4The Sherwood number, skin friction, and Nusselt number were compared to previously published data after the other components were set to zero. Present resultsDevi and Mabood [33]RdMaNuShCfNuShCf0.10.322784−0.0257910.322506−0.0257720.20.367464−0.0244640.367230−0.0244530.30.411889−0.0228670.411237−0.0228450.40.1119980.1379890.1118350.1379920.11.1990981.1990150.31.2281991.2281740.51.2577871.257761 The Sherwood number, skin friction, and Nusselt number were compared to previously published data after the other components were set to zero. Conclusion Examine the diffusion-thermo and thermo-diffusion properties on bioconvection flow of a hybrid nanofluid employing Yamada-Ota model. The primary conclusions from this study are as follows: As the volume fraction of NPs rises, the main velocity field decreases. As the volume fraction of NPs rises, the main velocity field decreases. The fluid flow model’s approximate solutions are estimated using the BBRT-NNs on the reference dataset. The fluid flow model’s approximate solutions are estimated using the BBRT-NNs on the reference dataset. The concentration of mobile microbes in the flow decreases as the Peclet number rises. In a variety of micro- and bio-systems, including microfluidic devices, biosensors, microbial fuel cells, and micromixers, gyrotactic microorganisms are used in conjunction with nanoparticles to improve thermal performance. This method is especially useful in industrial settings, where the best design of reactors and rotating gear depends on efficient mixing and enhanced heat transmission. The concentration of mobile microbes in the flow decreases as the Peclet number rises. In a variety of micro- and bio-systems, including microfluidic devices, biosensors, microbial fuel cells, and micromixers, gyrotactic microorganisms are used in conjunction with nanoparticles to improve thermal performance. This method is especially useful in industrial settings, where the best design of reactors and rotating gear depends on efficient mixing and enhanced heat transmission. The thermal and solutal profiles increase as increase the values of Dufour and Soret numbers. The thermal and solutal profiles increase as increase the values of Dufour and Soret numbers. Acknowledgment This work was supported by the Deanship of Scientific Research with Northern Border University, Arar, Saudi Arabia, under Project NBU-FFR-2026-1564-07. Nomenclature Electrical Conductivity Specific heat (Jkg-1) k-1) Magnetic parameter Dufour number Marangoni number Mass diffusivity coefficient (m2s-1) Local density of motile microbes Volume fraction of nanoparticles Dynamic viscosity (kgm-1s-1) Soret number Density of motile microbe(kgm-3) Diffusivity of microorganisms(m2s-1) Darcy parameter Uniform magnetic field(kgs-2A-1) Kinematic viscosity (m2s-1) Marangoni convection factor Surface tension (Nm-1) Flow fields of fluid(m s-1) Schmidt number Eckert number Fluid thermal for liquid phase(k) Constants Microbe difference parameter Fluid ambient temperature Nusselt number (kgm-1s-1) Nusselt number Concentration Coordinates Sherwood number Prandtl number Bio-convection Schmidt number Thermal conductivity (Wm-1k-1) Fluid density (kgm-3) Sherwood number (kgm-1s-1) Maximum cell swimming speed Microbes’ concentration difference parameter Hybrid nanofluid Surface tension coefficients Boger nanofluid Local thermal non-equilibrium Ternary hybrid nanofluid Stephan blowing Cattaneo-Christov heat flux model Trihybrid nanofluid Author contributions Humaira Kanwal and Shouki A. Ebad: Supervision, Methodology, Conceptualization. Fuad Alsarari: Methodology, Formal analysis. , Abed Saif Ahmed Alghawli and Dhouha Choukaier: Methodology, Data curation. Ruzimurod Abiyev: Writing – review & editing. Data availability All data used in this manuscript have been presented within the article. Declarations Ethics declaration Not applicable Clinical trial This study is not a clinical trial. Consent to publish declaration Not applicable. Consent to participate declaration Not applicable Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. References Associated Data Data Availability Statement All data used in this manuscript have been presented within the article.