2.1 Predictive modeling and simulation

In TE, predictive models are used to simulate scaffolds and biological processes,^{28, 29} predict the behavior of cells,³⁰ and understand how tissues respond to various stimuli in a virtual and controlled environment.⁸ Predictive models with the help of ML algorithms and AI use the structure-property relationship of the existing data to predict the performance of the scaffolds.³ These models can predict how biomaterials interact with cells, tissues, and the surrounding environment to provide the mechanical and biochemical signals necessary for tissue regeneration. Furthermore, these models are used in the design of bioreactors similar to the body environment,³¹ simulating the growth of blood vessels,³² predicting the integration of scaffolds with body tissues and the reaction of the immune system to them,¹⁷ understanding complex biological processes,¹⁵ personalizing treatment approaches for the patient,³³ and predicting the performance of new drugs and their effects on tissues.³⁴ Xu et al.³² used different ANNs with more than 30 variables to achieve the best predictive model of vascular TE strategies and were able to accurately predict 94.24% (for a NN with two layers) and 91.45% (for a single-layer NN). Some of the models and algorithms used for this purpose are summarized in Table 1 and Figure 2.

The selection of each of these algorithms and models depends on the nature of the data and the desired result. Zhou et al.⁴⁰ used the MeD-P ML model to predict the interaction between biomaterials and mesenchymal stem cells (MSCs, based on RNA sequence data) and optimize the performance of biomaterials on cell fate. Xu et al.⁴¹ developed a design engine using ANNs and decision trees to provide better TE designs in the treatment of laboratory animals.⁴²

2.2 AI-driven optimization of biomaterial properties

Biomaterial design should be optimized according to the necessary physical and biological characteristics⁴³ and includes the step-by-step selection of bioactive ligands based on biological studies and natural tissue structure, requires a lot of primary resources and time.^44-46 AI networks (due to their ability to analyze a large set of data and provide predictive algorithms and models) can accelerate this and reduce costs.^{47, 48}

Choosing the optimal algorithm (supervised and unsupervised learning) for biomaterials development is done according to the type and amount of available data.⁴⁹ These algorithms predict the effect of motifs, functional groups, and their combination on the performance and properties of the resulting composite by evaluating molecular weight, rheological properties, chemical spectrum, synthesis methods, and characteristics of various polymers and biomaterials in databases such as Reaxys, PubChem, and Chem Spider, and lead to facilitating the discovery and design.^50-52 There are various database libraries for various biomaterials, some of which are summarized in Table 2.

ML can help with the presentation of a set of optimal reproducible biomaterial designs, express the impact of their chemical composition on their biological activity and performance, and accelerate their entry into the market.⁵³ To date, ML has been used in investigating the effect of biomaterials on peptide hydrogel formation,⁵⁴ cell adhesion and protein absorption,⁵⁵ and foreign body response.⁵⁶ ML can be used to identify interactions between heterogeneous inputs such as chemical composition, processing conditions, concentration of additives, and others.⁵⁷

To design biomaterials with the help of ML, a large library of experimental data must first be created. Li et al.⁵⁴ created a library of 2000 self-assembled hydrogels based on Fmocamino acids, aldehydes, isocyanides, and amines. They were able to predict the relationship between the type and sequence of peptides and hydrogel properties by evaluating their molecular structure, rheological properties, cell adhesion, and morphology, with the help of regression models. This library played a significant role in determining the optimal sequence for various applications. Pugar et al.⁵⁷ used the combination of inputs related to the composition and stoichiometry of polyurethane monomers with thermodynamic and physicochemical properties predictors to predict the mechanical properties and Young's modulus of different polyurethane formulations.

Barrett et al.⁵⁸ evaluated the effect of amino acid motifs on the antifouling and antimicrobial properties of the resulting peptide using Bayesian networks. This model helps them to find that the effect of the background distribution of amino acids on the above properties is greater than that of sequence motifs. Kwaria et al.⁵⁵ used a combination of SVM, Regression, and NN models and the chemical and structural information of the constituent molecules of 145 monolayers to predict the wettability and protein absorption of self-assembled monolayers. Their results showed the ability of the resulting models to predict the wettability and protein absorption of layers and the effect of each of the structural parameters on it, which helped to design optimal materials. Using an artificial NN, Echezarreta-López et al.⁵⁹ examined the available data on bioactive glasses to determine the relationship between motif design, bioactive glass concentration, silver doping, and silicon doping on antimicrobial activity. Their results show the ability of NNs to predict interactions between parameters that traditional statistical analyses are unable to do.

2.3 AI-based design of scaffolds and bioinks for 3D printing

Scaffold fabrication methods are divided into two categories: conventional methods (including electrospinning, solvent casting, and phase separation) and modern 3D printing methods or additive manufacturing (AM, such as fused deposition modeling, extrusion, stereolithography, and VAT photopolymerization).^{60, 61} 3D printing, as a revolutionary technology, has been widely used in the fabrication of drug release devices, scaffolds, artificial organs, and 3D drug screening models by eliminating the disadvantages of traditional methods.⁶⁰ 3D printing with layer-by-layer deposition of materials under computer control provides the possibility of reproducible construction of cost-effective, accurate, and patient-specific scaffolds with complex architecture, controllable pores, and high efficiency from a wide range of biomaterials.⁶¹ However, the fabrication of these scaffolds requires the optimization of fabrication parameters such as speed, temperature, nozzle diameter, and pressure.^{62, 63} In addition, the properties of the biomaterial used as bioink, such as printability, viscosity, and curing mechanism, are also effective in this optimization, which requires adjustment between a large number of different but highly correlated variables. bioink generally includes ceramics, polymers, cells, growth factors, etc., whose optimization requires exhausting, time-consuming, and expensive experimental tests that have delayed the introduction of this technology to the bedside.^{63, 64}

Using AI to distinguish between low-quality and more promising print conditions is the first step to developing a suitable print condition detection system.²³ AI can be helpful by using algorithms and ML networks to increase speed and dimensional accuracy, optimize bioink, and improve various stages of scaffold fabrication and 3D printing.^{23, 65} Algorithms predict the properties and printability of different bioink formulations and the correct ratio of various components in bioink based on input data (and not fixed program instructions).⁶³ ML recognizes incompatible biomaterials in bioink preparation and provides an optimal protocol for different stages of printing (preprocessing, processing, and postprocessing).⁶⁶ Menon et al.⁶⁷ were able to predict the behavior of complex systems, increase the printing speed (2.5 times), and provide optimal formulations for silicone elastomer printing. ML can provide new proposals for scaffold design and upgrade existing designs based on customized recommendations and can be useful in maximizing repeatability, structural properties, printability, and print quality assessment (minimizing the use of materials and services) by predicting and optimizing printing parameters.⁶⁸ For example, by providing the nozzle size, temperature, and pressure used for 3D printing and their corresponding outputs such as cell damage, time, mechanical properties, and cost, algorithms can be trained to predict the performance of new input data. Gu et al.⁶⁹ used a combination of artificial NNs and various geometric data to predict the geometric configuration of different materials to print scaffolds with optimal mechanical properties (high toughness and strength). Conev et al.²³ used regression and classification-based models (based on Random Forests) to predict the effect of poly(propylene fumarate) concentration and printing speed and pressure on the dimensional accuracy of materials. Their results indicated the ability of both models to correctly label most of the tested configurations.

Data sources used for AI-aided 3D printing are provided through clinical diagnostic, histological, and immunohistochemistry images. Layered images of scaffold printing steps are also involved in creating large-scale data libraries for ML analysis. With the help of these images, AI can be involved in the quality control of scaffolds and the detection of defects such as microstructure errors, wrongly positioned cells, and curved layers.⁷⁰ By developing a DL method, Francis et al.⁷¹ predicted the amount of distortion of printed pieces with a laser-based method. Using sparse representation, naïve Bayes, decision tree, support vector machine, NN, and KNN ML algorithms, Tootooni et al.⁷² predicted the dimensional changes in the structure obtained from 3D printing and showed the best classification performance belongs to sparse representation.

Although there are few studies on the use of ML in 3D bioprinting processes, the combination of printing and ML can help evaluate drugs, reduce animal testing, reduce costs, predict the in vivo behavior of scaffolds, and accurately printing of damaged organs to help in replacement of damaged tissue and accelerates the progress of regenerative medicine.

2.4 ML for image analysis

Today, with the development of TE and the complexity of the resulting structures, various non-invasive medical imaging methods such as magnetic resonance imaging (MRI), computed tomography (CT), and optical coherence tomography (OCT) are needed for a more accurate evaluation of the tissue, each of which has various spatial resolution and use.⁷³ The use of AI along with these imaging techniques can lead to the improvement of diagnostic information, the reduction of operator errors when reading images, and the improvement of image analysis (such as classification, localization, regression, and segmentation), which will be evaluated in detail in this section (Figure 3).

AI uses special algorithms for each step of image analysis. Classification algorithms lead to the classification of images (e.g., diseased and non-diseased images) into completely separate categories, regression algorithms lead to assigning a specific number to an image (for example, bone age estimation), and localization algorithms lead to determining specific tissue or part location in the image as the location of the center pixel, and the segmentation algorithms result in the segmentation of each pixel in the image as part of a specific object.⁷⁷

MRI is one of the common soft tissue contrast imaging methods used to image implanted scaffolds and structures inside the body based on nuclear magnetic resonance.⁷⁸ MRI has been used in various areas of TE, such as evaluation of the maturity of implanted cartilage scaffolds,⁷⁹ continuous imaging of transplanted tissue,⁸⁰ measuring the depth of skin tumors, the degree of involvement of underlying tissues, and the possibility of recurrence in grafts,⁸¹ prediction of abnormal brain structure in atopic dermatitis,⁸² and assessment of vascular growth within the scaffold.⁸³ MRI image analysis is extensive and multifaceted. AI can help improve image reconstruction, noise removal and resolution improvement, object detection, image segmentation, and disease diagnosis and prediction based on MRI.^{39, 84} Wang et al.⁸⁵ used functional MRI and AI software to compare brain responses to mechanical scratching stimuli (a nylon thread) in various types of skin [hairy (forearm) and glabrous (fingertip)]. Their results predicted and inferred that a scratch on the fingertip's skin turns into tactile information. On the skin of the forearm by stimulating the precuneus part of the brain, it turns into sensory information and itch, resulting in a stronger burning sensation. AI-assisted compressed sensing is an advanced DL NN that shortens the scanning time required for MRI, accelerates reconstruction and improves the quality of images.⁸⁴ Hammernik et al.⁸⁶ proposed a network of variables using ML, which accelerated the reconstruction of MRI images by examining free parameters, reducing the reconstruction time to 193 ms.

Image segmentation is done to accurately describe the contours of organs, tissue, and lesion structures, but generally, noise and artifacts of MRI images interfere with it.⁸⁷ Various AI networks, such as U-Net, Mask RCNN, V-Net, and UNet++ have been designed to overcome this problem. Zhang et al.⁸⁸ used Mask RCNN to segment breast tumor MRI images and achieved an accuracy of 0.75. Ronneberger et al.⁸⁹ also used U-Net (with U encoder and decoder structure and a unique connection) to segment 2D medical images.

Object recognition is an important part of medical image processing, which uses a frame to mark the desired areas (such as lesions) and leads to improved segmentation accuracy. MRI slices include 2D and 3D sets. Despite the larger volume of information in 2D methods, the 3D ones provide background information and higher accuracy.⁹⁰ Using convolutional NNs, Qi et al.⁹¹ were able to improve the diagnosis of cerebral microbleeds based on MRI images to an accuracy of more than 93.16%.

ML tools are also used to diagnose diseases based on MRI images. Various networks such as AlexNet and NNs have been used for more accurate classification and diagnosis of Parkinson's,⁹² Alzheimer's,⁹³ survival of brain tumors,⁹⁴ and schizophrenia⁹⁵ based on MRI images.

CT is used as one of the typical diagnostic methods in clinics to image body tissues based on the attenuation of X-rays in different layers of the body and 3D reconstruction of the layers by computer.⁹⁶ Various AI networks play a role in improving the above-mentioned aspects of CT image analysis. For example, the cone-beam back projection and the combination of filtered back projection algorithm with NN can be used to overcome limited-angle tomography and the inability of back projection to complete the end-to-end network in CT images.⁹⁷ Solomon et al.⁹⁸ used an AI reconstruction algorithm to reduce noise when reconstructing CT images. Generative adversarial networks (GAN) and sharpness loss, perceptual loss, and adversarial loss functions have been used to improve the quality of low-dose CT images and reduce their noise.⁹⁹ In addition, ML models based on CT images were also used for the diagnosis of COVID-19¹⁰⁰ and lung cancer.¹⁰¹

OCT, as a non-invasive, high-resolution optical imaging technique, can evaluate different degrees of absorption and scattering of light by different tissues to create images.¹⁰² Neural and GAN networks lead to noise removal from OCT images. Zhou et al.¹⁰³ used the CycleGAN and conditional GAN networks to unify the style of OCT images and remove noise, respectively. In addition, AI helps to diagnose diseases related to retinopathy¹⁰⁴ and glaucoma¹⁰⁵ based on OCT images. OCT angiography can be used with the help of AL to segment and observe retinal blood vessels that cannot be seen with other methods. In general, ML networks help improve the analysis of medical images by improving the segmentation results or increasing the ability of a single network to extract data.

2.5 AI-enabled TE and therapeutics development

Considering the potential role of AI in various parts of TE, it is not far from expected that AI plays a role in the RM in clinical steps. When transplanting engineered tissue or constructs, AI can help choose an optimal immunosuppressive regimen and quickly assess the matching of the transplanted organs.¹⁰⁶ In addition, by using AI in monitor systems, it is possible to observe the reaction in a real-time of the tissue repair process and help to maximize the repair efficiency with timely reactions.¹⁰⁷ The use of AL-based approaches, especially cranioplasty, shows their increasing importance.¹⁰⁸

On the other hand, the development of new drugs and therapies with conventional methods is a multi-step and expensive process (in three phases), and post-marketing monitoring, which costs more than 2 billion dollars.^109-112 AI can speed up this processes and stimulate tissue repair by using databases and predicting effective therapeutic combinations.^{113, 114} Several ML algorithms have been used in each of these steps, such as SVM, convolutional NNs, DL, and deep NNs (for predicting drug-target interactions) and random forest, artificial NNs, and simple Bayesian (to evaluate the relationship between the physicochemical qualities of materials and their biological activity).^{109, 115}

3.1 Robotic platforms and automated systems

TE plays an important role in prolonging life not only on Earth but also in space exploration; long space flights would benefit from TE/organ growth with robotic surgery to extend astronaut life.¹³⁸

Robotics can improve TE steps performance.^139-142 For example, recent soft robotics research has raised awareness of the importance of material properties,^{142, 143} shifting the focus from rigid elements to soft materials.¹⁴³ Although recent advances have promised to close the performance gap with animals, current soft-robot actuators based on, for example, electroactive polymers, shape memory alloys or pressurized fluids are not yet mature enough to mimic the complex with high resolution.^{141, 142}

3.2 Scaffolds as robots

Other features will also be useful for future scaffolds and bring them closer to today's robots. In addition to movement/reconfiguration enabled by guided activity, other applications of scaffold robots can be used in detection, computation/control, and communication and sensing physical and chemical parameters, such as pH, temperature, and stress. Computing even simple algorithms for feature detection mapping is complex and could benefit from good models. It defines the controls for effective/optimal release and movement of drugs. It seems useful to equip the scaffold with a communication capability: as a receiver, for remote control (motion, drug release) or as a transmitter of information detected or determined by the scaffold, scaffold status, etc., interactions with other tissues/organs, interaction with extracorporeal instruments, and remote access, from outside the human body.¹³⁸

3.3 Computational optimization

Traditionally, scaffolds were made based on trial-and-error method in a very time-consuming, expensive, and sometimes lacks precise control manner. The use of computational methods such as computer-aided design (CAD), finite element analysis (FEA), and computational fluid dynamics (CFD) can significantly reduce the number of test iterations to improve scaffold properties. With reasonable accuracy of simulation results, CAD has also become a tool to predict scaffold properties before production and testing.^145-147 It is particularly useful to predict some properties and behaviors of the framework that are rarely experimentally investigated, such as the stress-strain distribution. All this means the advantages of using computational methods in the design of the scaffold. Overall, computational methods for scaffold design not only improve the traditional scaffold manufacturing process, but more importantly, open a new path in tissue engineering, providing researchers with a wide and flexible choice for scaffold design and optimization.¹⁴⁵

It is important to note that the spread of computer-aided frame design cannot be separated from the development of 3D printing or AM. Production technology places strict limits on frame design. For example, pore size cannot be precisely controlled by traditional manufacturing methods such as gas foaming, solvent casting, and freeze drying.¹⁴⁸ As a result, if computationally designed and optimized scaffolds cannot be easily and accurately produced, this may undermine the importance of designing such scaffolds. Fortunately, using 3D printing, scaffolds can be manufactured according to a CAD in less time, with greater flexibility and precision than traditional scaffolding techniques.^{148, 149} Compared to traditional scaffold fabrication techniques, 3D printing can produce scaffolds with highly complex microstructures with precise control and easy design modification.^{145, 150} Although current 3D printing techniques are not yet perfect for producing scaffolds of all desired shapes and sizes, their advantages over traditional production techniques and the great potential of scaffold fabrication still make it promising for the manufacture of computer-designed scaffolds.¹¹¹

Different loads on transplanted scaffolds include compressive, tensile, shear, torsional, bending, and biomechanical/physiological loads.^151-153 Compression is the most common in vivo loading.¹⁵⁴ The loading sizes vary greatly between apps. For example, the forces experienced in a bone scaffold are hundreds or even a few thousand Newton.¹⁵⁵ However, some polymer scaffolds can only withstand up to 100 Newton.¹⁵⁴ Therefore, it is very important to consider the mechanical properties and behavior of scaffolds.

Topology optimization is a much more powerful tool in scaffold design than size and shape optimization.^{156, 157} The purpose of topology optimization is to determine the best distribution of a limited number of materials in a given area under certain loading conditions to achieve certain desired properties.^158-161 Porosity, structural modulus, and compatibility are some of the important properties that are often studied in scaffold topology optimization. Many studies have introduced topology optimization to design unit cells and support structures with desired porosity and mechanical properties. Typically, the objective function is to maximize stiffness (equal to minimizing compliance or strain energy) with constraints such as minimum porosity (equal to maximum volume fraction). In some cases, stiffness is also limited to achieve maximum porosity/permeability. Based on previous work, Hollister and Lin,¹⁵⁶ one of the earliest researchers to introduce topology optimization, optimized a maximum permeability unit cell with a limited tolerance between targeted and effective flexibility and minimum porosity. In other early works, the unit cell was optimized to meet the desired stiffness by limiting its porosity and/or pore size.^{157, 162} Recently, Kang et al.¹⁶³ optimized unit cell with an objective function to minimize the error between the targeted and effective bulk moduli and diffusivities by limiting its porosity. Almeida and Bártolo¹⁵⁹ proposed a topology optimization method to obtain an optimal unit cell topology with minimal compatibility under different porosity and loading conditions. They further investigate optimized structures based on material interconnectivity to determine valid designs.¹⁵⁹ Dias et al.¹⁶⁰ introduced topology optimization to explore the best possible unit cell structure with maximum permeability limiting the elastic strain energy and volume fraction; they prepared the optimized structures using SLS and compared the experimental and computational results. Youssef et al.¹⁶⁴ combined topology optimization with CFD and optimized frame structure to achieve a uniform shear stress distribution.

4.1 ML-based integration of tissue engineering data for clinical decision support

ML refers to computer systems that create learning patterns using existing data and patterns. These patterns assist in better analysis, problem-solving, and decision-making.¹⁶⁵ This data can encompass various topics, including laboratory data, particularly data derived from experimental work in TE.¹⁶⁶ Today, ML is being utilized in clinical decision support systems and assist physicians at different levels of clinical decision, such as prediction, interpretation, diagnosis, and management of diseases.¹⁶⁷

This technology can combine information and experimental data from laboratory research with clinical data to create new learning patterns. AI and ML reasoning techniques can assist physicians in the diagnosis and treatment of diseases, prediction of probable responses, increasing speed and accuracy, reducing medical errors, and selecting the best therapeutic strategies from larger data sets, more precise decision-making, and the choice of more cost-effective approaches.^{168, 169} As data is collected and stored using different objectives and methods results in increases in the level of dispersion, they must be stored in a common format in information systems to be utilized in AI-based systems and clinics. For example, digital imaging standards and communications in medicine (DICOM), picture archiving, and communication systems (PACS) are employed in medical imaging management. Therefore, establishing a platform for storing laboratory data (such as TE), clinical data, and information related to patients' clinical history in a suitable format for AI systems, can be the facilitator.¹⁷⁰ Furthermore, the novelties of the field of TE and the diversity and complexity of human cells, tissues, and organs have resulted in insufficient data and strategies for conducting new experiments. Therefore, conducting new experiments requires examining various scattered resources and employing trial-and-error strategies, which can also increase human error. Based on input data, AI and ML are capable of creating algorithms to predict experimental results and the body's response to the methods employed in the experiments.¹⁷¹

Several experiments have been conducted to investigate the mechanisms of stem cells and molecular genetics in different organism, resulting in diverse data in this field. In a study, AI was utilized to develop insights and patterns from diverse datasets regarding the mechanisms of regeneration in worms. The created model holds the potential to be a valuable asset in advancing the field of TE.¹⁷² Consequently, selecting the optimal combination based on tissue type and patient requirements presents a challenging task. The presence of an intelligent system to integrate experimental data and facilitate appropriate selection becomes imperative.¹⁷³ In this regard, AI was employed in a study to predict the outcomes of using various strategies for vascular TE.³² In another study, AI based on ANNs and decision trees was utilized to integrate experimental data in the field of tendon TE and establish a decision support system for selecting safe and cost-effective approaches. The results of this research indicated that 28 out of 30 models with tendon defects were successfully treated using the proposed AI-driven therapeutic strategies based on established patterns¹⁷³

AI can provide a scheme for categorizing experimental data that enables searching between different datasets. In this context, a group employed AI aggregated experimental data of cartilage TE in a centralized system. They successfully identified suitable treatment schemes for 18 animals of 20 examined animals.⁴¹ Dhondalay et al. constructed an AI system based on ANNs and gathered experimental data on the impact of various factors on the quality, efficiency, and surface markers of cells used in cell therapy. Their work demonstrated that ANNs can help in selecting influential factors in large-scale cell processing and production.¹⁷⁴ Thus, it can be concluded that AI, through DL approaches, enables the establishment of connections between new and old data in TE and help to more accurate and faster clinical decisions.⁴¹

4.2 AI-assisted analysis of patient-specific data

ML and other subsets of AI not only can be beneficial in the analysis of medical and biomedical data and disease prediction, analysis, and enhancing the resolution of medical imaging and analyzing histopathological images, but also can help selecting appropriate therapeutic methods with lower error rates based on specific patient data. These systems help physicians with high accuracy and without fatigue, enabling the comparison of individual patient data with large volumes of laboratory and clinical data.¹⁷⁵ In image classification, the objective is to classify or predict a specific class of object in an image, which can aid in disease prediction based on existing patterns. In object detection, it identifies and classifies specific regions within an image, enabling the search for specific objects, such as the detection of specific cell types.¹⁷⁶ Histopathological images are utilized for diagnosing diseases, including cancers. Analyzing histopathological images and comparing them with existing data is time-consuming and subject to disagreement. AI and ML methods for analyzing images and diagnosing patient diseases assists in more accurate and faster decision-making.¹⁷⁷

4.3 Personalized medicine

Precision medicine or PM provides a new approach to targeted treatment based on individual genetic, biochemical, and physiological characteristics, including genes and proteins.^{178, 179} For this purpose, the use of clinical data and omic data together can be a way forward, and the use of clinical trials alone is not effective enough due to the lack of attention to individual genetic differences. Omic data includes various information including genomics, transcriptomics, proteomics, epigenomics, metagenomics, metabolomics, and nutrigenomics.¹⁸⁰ To investigate and development of new drugs in the treatment of diseases, various experiments and clinical trials in different phases of preclinical (laboratory) and clinical stages are required. Pre-clinical phases are conducted to determine the pharmacokinetic properties, optimal dosage, and the drug's effects based on the type of disease by using cell and animal studies or expanding methods based on TE such as organ-on-chip and microfluidics.

Clinical trials are divided into four phases. In Phase I, the drug's effects and the optimal effective dosage are examined in a small group. Then, in Phases II and III, the drug's effects are studied in a larger population of patients.¹¹¹ In Phase II, clinical trials evaluate the drug's efficacy, while in Phase III compare the efficacy of the new drug with previous standard treatments. Finally, Phase IV involves the introduction of the drug to the market and use in the general population and monitoring of unforeseen side effects of the drug. The integration of AI with each of these phases can enhance the accuracy of assessing and evaluating the efficacy of drugs and interventions, based on individual characteristics and contribute to the advancement of PM. In Phase I clinical trials, the potential impact and possible side effects of the new drug based on genetic factors in the patient are investigated. For this purpose, N-of-1 trials can be used where studies are conducted on one person.¹⁸¹ Also, the use of organoids together with cells isolated from the patient for pathology and drug tests is increasing as an effective approach in helping PM. The use of AI in these researchers for analysis can be helpful.¹⁸² In Phase II of studies, drugs and effective interventions must be examined in larger populations. The use of AI in this phase can assist in selecting the most suitable intervention based on available information and patterns. For example, several different drugs may be under investigation for a particular disease, and AI can propose the best treatment based on the patient's genetic characteristics and the mechanism of action of the drugs. If the safety and efficacy of drugs are approved, they can be used in Phase III clinical trials after approval by the FDA. By utilizing AI and creating “learning systems” based on the information obtained from the drug's effects on consuming individuals in clinical trials of Phases I, II, and III, as well as clinical histories and pharmacogenetic characteristics of patients, we can take another step towards PM. Recording and monitoring the effects of drugs or therapeutic interventions, as well as registering feedback or potential side effects, enables the support of advancing or replacing new therapeutic methods in the designed learning system during the subsequent phase.¹⁸¹