Exploiting Pharma 4.0 Tools in Non-Biological Complex Drugs: Comparison
View Latest Version
Please note this is a comparison between Version 2 by Catherine Yang and Version 1 by Vera Malheiro.

The pharmaceutical industry has entered an era of transformation with the emergence of Pharma 4.0, which leverages cutting-edge technologies in manufacturing processes. These hold tremendous potential for enhancing the overall efficiency, safety, and quality of non-biological complex drugs (NBCDs), a category of pharmaceutical products that pose unique challenges due to their intricate composition and complex manufacturing requirements.

  • Pharma 4.0
  • non-biological complex drugs

1. Introduction

It is widely recognized that bringing a new pharmaceutical drug to the market is a complex, lengthy, and costly process associated with high uncertainty. This process is known as drug development and encompasses various stages, including preclinical research, drug design and production, regulatory filling, clinical trials in humans, obtaining regulatory approval, and the subsequent steps of manufacturing and marketing [1]. Throughout the years, the pharmaceutical industry has undergone significant changes and advancements, progressing from Pharma 1.0 to Pharma 4.0, and more recently reaching the era of Pharma 5.0. Even though Pharma 4.0 is a relatively recent development, the truth is that certain pharmaceuticals are already venturing into the 5.0 era [2].
During Pharma 1.0, the processing of materials derived from minerals, animals, and plants underwent a significant transformation. The use of basic hand-operated tools gave way to the introduction of commercial-scale equipment capable of crushing, blending, milling, and pressing a larger quantity of medicines. In fact, certain key machines developed during the Pharma 1.0 era continue to be widely used in the present day, showcasing their durability and effectiveness [3]. Subsequently, electricity and early electronic machines ushered in a new era in the evolution of the pharmaceutical industry. This phase witnessed the integration of digital tools into various aspects of pharmaceutical research, development, manufacturing, and distribution [4]. For the pharmaceutical sector, this technological incorporation marked a significant milestone because it allowed for a more data-driven and patient-centric approach, besides providing a larger-scale production and more efficient quality control. However, these process controls were far from perfect, since they were confined to pre-determined and static settings, which only allowed for the monitoring of process performance and passive control strategies [3]. The third industrial revolution was enabled by the emergence of computers and communication technologies, including networked computing, the internet, and wireless communications [4]. This development resulted in increased automation of both processes and equipment, fostering the implementation of continuous manufacturing and enhanced active control. This revolution, known as Pharma 3.0, enabled the adoption of advanced control strategies, elevating product and process quality and reducing the need for human operators. This decrease in reliance on human operators also streamlined the tracking of parameters and production metrics. Although some industries are well into Industry 3.0, the pharmaceutical industry is still very much transitioning into it [3]. Pharma 3.0 also brought the implementation of advanced process analytical technology (PAT), which provides real-time data on process and product quality. It also enhanced quality-by-design (QbD) processes, which focus on controlling product quality within specific parameters [3]. While Pharma 3.0 already enables a much-improved understanding of how to capture, analyze, and secure large amounts of data in pharmaceutical manufacturing, there is still room for further technological advancements to achieve the full potential of PAT and QbD [5]. Later on, with the appearance of artificial intelligence (AI), cloud computing, machine learning (ML), big data analytics, in silico modeling, 3D printing, and other advanced manufacturing technologies, the manufacturing process was actualized, forcing the emergence of another industrial revolution, known as Pharma 4.0 [6]. These advanced manufacturing technologies empower autonomous and self-organizing systems capable of operating independently without human intervention [3].
In the pharmaceutical sector, the fourth industrial revolution allowed for a shift in the paradigm of formulation development [7]. The potential to integrate diverse data sources facilitates the connection of both external and internal information, establishing a comprehensive network. In the context of pharmaceutical manufacturing, this integration entails merging external data, including patient experiences, market demands, supplier inventories, and public health emergencies, with internal data encompassing energy and resource management, modeling and simulation results, and laboratory data. This fusion enables unparalleled real-time responsiveness, monitoring, control, and predictive capabilities [3].
In an ever-evolving marketplace characterized by a fast-changing and smartly integrated supply chain, coupled with more active participation of patients in their healthcare decision-making, pharmaceutical companies face an urgent imperative to maintain their competitive advantage. In this setting, the key factor that sets businesses apart is their ability to meet the expectation of Pharma 4.0, which, despite being a complex process to implement, provides enhanced resources for ensuring product safety compliance, safeguarding the supply chain, and fostering pharmaceutical development [6]. This concept of Pharma 4.0 is closely linked to artificial intelligence (AI), a technology-based system involving various advanced tools and networks that can mimic human intelligence [3]. It possesses systems and software with the ability to interpret and learn from the input data, enabling them to autonomously make decisions in order to achieve specific objectives [8]. AI plays a crucial role in Pharma 4.0 by enabling smart, data-driven decision-making and optimization throughout the entire pharmaceutical value chain [9].

2. Applying Pharma 4.0 Tools in the Production of Non-Biological Complex Drugs

2.1. Additive Manufacturing: 3D Printing

2.1.1. Polymeric Micelles

3D printing has exhibited promising results in the manufacturing of diverse drug delivery systems, such as polymeric micelles. In a recent investigation, chitosan-based polymeric micelles containing camptothecin (CPT) were integrated into 3D printing systems and coated with an enteric layer. This approach aimed to safeguard the nanosystems from the harsh conditions of the gastrointestinal tract [77][10]. The manufacturing process involved the use of a bioprinter that combines fused deposition modeling (FDM)—a widely used 3D printing method that involves melting and extruding filament materials layer by layer—and injection volume filling (IVF)—a technique that enables the incorporation of solutions or dispersions at room temperature into the extruded scaffold [77,78][10][11]. The in vitro drug release profile showed that both printfills containing the free drug and those with micelles effectively managed drug release. Consequently, there was no drug release from the printfills within the initial 2-h period at pH 1.2. However, when pH changed to 6.8, the retarding polymer began to dissolve, permitting the entry of water. Researchers also concluded that both polymeric micelles and the free drug present in the dissolution media did not exhibit any cytotoxic effects on Caco-2 cells, a colorectal cancer cell line. In fact, these cells showed an increase in metabolic activity of up to 100%, which could be attributed to the presence of simulated gastrointestinal fluids. The permeability of CPT from the micelles was observed to be higher than that of the free drug in both Caco-2 standard model (a) and a 3D intestinal model (b), with significant differences becoming evident during the final incubation periods. In the 3D model, CPT permeability reached approximately 27%, representing an enhancement compared to the standard model´s permeability of 20%. This heightened permeability in the 3D model aligns more closely with the in vivo human intestine, reflecting a drug permeability that is more akin to real-world conditions. The experiment also demonstrated consistent transepithelial electrical resistance (TEER) maintenance for both micelles and the free drug, reflecting the monolayer integrity and consequently suggesting that the tested formulation is safe. Besides, the apparent permeability coefficients exhibited notable disparities between the micelles and the free drug. In the 3D model, the apparent permeability coefficients of CPT from the micelles indicate a significant increase in CPT permeability and, consequently, bio-availability. On the other hand, the free drug maintained an apparent permeability coefficient that was similar in both models, signifying that the drug´s permeability was nearly half that of the micellar drug, specifically within the context of the 3D model. Furthermore, to ensure the structural integrity of the membrane following the permeability experiment, hematoxylin, and eosin (H&AND) staining was conducted for both models and under all conditions. This staining process employed two dyes: hematoxylin, a basic dye, and eosin, an acidic dye. Hematoxylin imparts a purple hue to acidic structures, such as the nucleus, while eosin imparts a pink color to basic structures, such as the cytoplasm and extracellular matrix. These observations indicate that the membrane remained intact and well-formed throughout and following the permeability assay. Overall, this study’s findings showed that the printfills were able to keep the micelles intact until they reached the intestinal pH, increasing the CPT intestinal absorption and, consequently, its oral availability. Furthermore, the combination of 3D printing and nanotechnology holds considerable potential for the targeted release of polymeric micelles in the colon. This advancement can enhance the absorption of drugs in the intestines while safeguarding them from degradation as they go through the gastrointestinal tract [77][10].
In another study, researchers focused on creating a bio-ink suitable for 3D printing a hydrogel implant with controlled drug release capability. To achieve this, simvastatin was loaded into polymeric micelles composed of polylactide/poly (ethylene glycol) triblock copolymers (PLA-PEG-PLA). These micelles were then incorporated into hydrogels via a photo-cross-linking 3D printing process. The resulting simvastatin-loaded triple-network hydrogel demonstrated remarkably long-term drug release for over 14 weeks, consistently maintaining a therapeutic concentration. These findings indicate that these micelles hold great promise as a bio-ink material, providing long-term hydrogel stability, biodegradability, and sustained delivery of hydrophobic drugs, such as simvastatin [79][12].

2.1.2. Liposomes/Niosomes

Due to their numerous benefits, liposomes have garnered significant attention, especially in the fields of cancer treatment and vaccinology. However, their development remains a challenge, making it crucial to seek out innovative approaches that enable fast, safe, and consistent production with high-level batch-to-batch reliability [80][13].
In an experimental study, researchers used 3D printing technology to create a 3D-printed niosomal hydrogel (3DP-NH) containing cryptotanshinome (CPT) as a topical delivery system for acne therapy. To formulate the CPT-loaded niosomal hydrogel, the CPT-loaded niosomes were carefully added, drop by drop, into the hydrogel. Subsequently, the resulting mixture was printed using an extrusion-based 3D printer to produce a 3D-printed CPT-loaded niosomal hydrogel (3DP-CPT-NH) with a specific drug dosage, shape, and size. The findings were that the 3DP-CPT-NH exhibited a significant anti-acne effect without causing any skin irritation [81][14].
Another investigation was conducted, combining microfluidic technology (MF) and 3D printing, leading to the formulation of “diamond-shaped” devices designed for the production of liposomes loaded with lysozyme as a model drug. Computer-aided design software was used to design microfluidic devices with diverse geometries, which were then printed using high-resolution digital light processing (DLP)—3DP. Stability tests confirmed the consistency of the developed formulations, and an encapsulation efficacy study showed positive results. Overall, this study showcased the effectiveness of combining MF and 3DP, highlighting the potential for synergistic growth in this field [80][13].

2.1.3. Nanocrystals

In the pharmaceutical field, 3D printing has also been making an impact in the manufacturing of nanocrystals, allowing for more precise control over drug dosage and release profiles [82][15]. For instance, additive manufacturing has been used to encapsulate nanocrystals within polymeric matrices, creating drug-loaded filaments or tablets, and improving the solubility and bio-availability of poorly soluble drugs. The precise control over the placement of nanocrystals in the printed structure allows for enhanced drug delivery and therapeutic efficacy. In fact, a recent study aimed to develop fast-dissolving oral polymeric film formulations loaded with indomethacin nanocrystals using 3D printing technology, and the outcomes demonstrated that this offers a promising approach for enhancing solubility in immediate-release formulations [83][16].
Another experiment focused on the development of an in situ forming robust injectable and 3D printable hydrogel based on cellulose nanocrystals. The results demonstrated that the hydrogels exhibited excellent injectability and maintained their shape fidelity without the need for additional cross-linking steps. The interlayer bonding between the printed layers was strong, resulting in the formation of sTable 3D structures, even up to 10 layers [84][17].
Additive manufacturing also allows for the creation of personalized drug delivery systems by incorporating nanocrystals. For instance, researchers have used 3D printing to fabricate patient-specific tablets containing nanocrystals of poorly soluble drugs. This enables customized dosages and controlled release profiles to be tailored to individual patient needs. In the first-ever study that included nanocrystals within 3D-printed tablets, albendazole nanocrystals were successfully incorporated into tablets, achieving a concentration of up to 50% w/w, which is not typically attainable with conventional tablets. Moreover, the printlet formulation with nanocrystals exhibited superior efficacy in improving drug dissolution in HCL 0.1N when compared to nanocrystals in hard gelatin capsules. The nanocrystals exhibited consistent particle size, crystallinity, and chemical stability both before and after 180 days of storage. Overall, the findings demonstrated the promising pharmaceutical potential of combining 3D printing and nanocrystals for the development of stable, fast-release, oral solid dosage forms of poorly soluble drugs. This experiment employed propylene glycol as a carrier and demonstrated that this technique holds promise for printing objects utilizing various types of nanocrystals embedded in low-melting-temperature polymers [85][18].

2.2. In Silico Modeling

2.2.1. Polymeric Micelles

A recent study focused on developing a novel technology called MeltDrops, which used hot-melt extrusion (HME) for continuous manufacturing of in situ gelling systems (ISGS) known to prolong the retention time and improve the bio-availability of ophthalmic drugs. This is relevant because the traditional manufacturing of ISGS has been challenging and costly, hindering their industrial scale-up and clinical implementation. However, MeltDrops technology offers a one-step extrusion process to develop these systems, which overcomes the limitations of batch manufacturing. Based on in silico modeling, researchers employed a molecular dynamics (MD) simulation to analyze the difference in physical properties of two types of MeltDrops—loaded with timolol maleate (TIM) or dorzolamide hydrochloride (DRZ)—under two different temperature conditions, 300 K (room temperature) and 308 K (physiological temperature). These simulations offered evidence of heightened interactions among drug, polymer, and water molecules at a higher temperature (308 K), suggesting the formation of ISGS with desired properties, including a solution–gel transition at physiological temperatures. Researchers also concluded that the in vitro drug release from MeltDrops technology demonstrated sustained and controlled release behavior, while marketed eyedrops showed complete drug release in less than 30 min. Besides, the results demonstrated a percentage decrease in intraocular pressure (IOP) following the administration of MeltDrops and marketed eyedrops, highlighting the superior IOP-reducing potential of MeltDrops compared to conventional options. Finally, a HET-CAM test was conducted in order to evaluate the potential ocular irritancy of MeltDrops. The results showed no signs of irritation, indicating that they are safe and well tolerated for ocular use [86][19].
Previously, a continuous manufacturing technique that utilizes a coaxial turbulent jet in co-flow was established for the production of paclitaxel-loaded polymeric micelles. More recently, researchers have employed coarse-grained molecular dynamics simulations to gain deeper insights into the impact of material attributes (specifically, the drug-polymer ratio and ethanol concentration) and process parameters (such as temperature) on the self-assembly process of polymeric micelles. Additionally, these simulations provided molecular-level information on micelle instability. The findings demonstrated a clear correlation between the micelle shape and drug encapsulation. As the paclitaxel content increased, the micelles transformed from spherical to ellipsoidal structures. From the simulation data, researchers were able to identify the critical aggregation number, which represents the minimum number of polymer and drug molecules required for this shape transition. Moreover, this investigation indicated that larger micellar size and reduced solvent accessibility contributed to the enhanced structural stability of the micelles. Additionally, researchers conducted an evaluation of the micellar dissociation free energy using steered molecular dynamics simulations across various temperatures and ethanol concentrations. The simulations showed that higher ethanol levels and temperatures led to micellar destabilization, resulting in a more significant release of paclitaxel. This increased drug release was attributed to the solvation of the hydrophobic core, promoting micellar swelling and reducing hydrophobic interactions, ultimately leading to a loosely packed micellar structure. In general, the computational predictions provided valuable insights into the micelle self-assembly process, morphological changes, drug release, and thermodynamic instability and showed excellent agreement with experimental results, underscoring its efficacy in studying the impact of material attributes and process parameters on the polymeric micelle formulation during continuous processing [87][20].
In another study, researchers used coarse-grained molecular dynamics simulations to investigate the behavior of a specific type of block copolymer called poly (ethylene oxide)-poly (propylene oxide)-poly (ethylene oxide) (PEO-PPO-PEO), commonly known as pluronics or poloxamers. They studied the effect of polymer and surfactant concentration on the morphology of these block copolymers and ionic surfactants, namely sodium dodecyl sulfate (SDS), in aqueous solutions. The results showed that when pluronics and SDS are present together in the solution, they tend to form mixed micelles and that the shape of those micelles depends on the relative concentrations of pluronics and SDS in the solution. The core of the mixed micelles consists of PPO chains from the pluronics, the alkyl tail of SDS, and some water molecules. The surrounding shell is composed of PEO chains, water molecules, and the sulfate headgroups of the SDS. Notably, with an increasing amount of added SDS, the observed morphology of the mixed micelles undergoes a transition from spherical to wormlike–cylindrical geometry. Overall, the molecular insights gained from studying the co-assembly of an ionic surfactant and an amphiphilic triblock copolymer in aqueous media have potential applications in various complex fluid mixtures. However, the accuracy of the results relies on the coarse-grained force field used, which can be improved with more computationally expensive atomistic simulations for quantitative comparisons with experimental data [88][21].

2.2.2. Liposomes/Transfersomes

In silico modeling has been employed in the manufacturing of liposomes because it allows researchers to simulate and predict the behavior of these structures, such as their stability, size, composition, drug encapsulation efficiency, and release kinetics. Furthermore, these models can help to optimize the manufacturing process, predict the performance of liposomal formulations, and guide experimental design [89][22].
In a recent study, the authors constructed computational models to identify active pharmaceutical ingredients (APIs) that can achieve the desired high concentrations in nano-liposomes via remote loading. The models aimed to predict the suitability of APIs for nano-liposomal delivery by considering fixed main experimental conditions, such as liposome lipid composition and size. The researchers also added a prediction of drug leakage from the nano-liposomes during storage, which is crucial for ensuring the development of pharmaceutically viable nano-drugs. More so, by using “load and leak” models, this group screened two large molecular databases to identify candidates’ APIs for delivery by nano-liposomes. Through the screening process, the researchers identified 667 molecules that showed positive results in both the loading and leakage models, indicating high-loading and stable characteristics. Among these molecules, 318 received high scores in both properties and, notably 67 of them are FDA-approved drugs [90][23]. These findings underscore the significance of computational modeling in optimizing liposomal formulations. By narrowing the search to molecules exhibiting high-loading and stability characteristics, researchers can concentrate their efforts on candidates with a higher likelihood of success. This approach facilitates efficient use of time, resources, and cost savings compared to traditional trial-and-error methods.
These computational approaches can—and should—be combined with experimental studies because it allows for a better understanding of the mechanisms that are being investigated. Computational modeling not only aids in explaining experimental results, but has also the potential to guide and inspire new directions for experimental research in the development of liposomal drug delivery systems [91][24].
In a study that aimed to develop active targeting liposomes to deliver anticancer agents to the treatment of hepatocellular carcinoma, computational modeling was used to gain insight into the structure and behavior of the intended targeted liposomal drug delivery systems within the bloodstream. This research showcases the complementary nature of these simulations alongside experimental research, often offering valuable mechanistic context [92][25].
Another example of the predictive power of in silico modeling in the pharmaceutical industry is an investigation that explored the use of thermosensitive liposomes (TSL) for targeted drug delivery to tumors. The researchers created a three-dimensional computer model to simulate the delivery of the TSL-encapsulated doxorubicin to mouse tumors. To do so, a mouse hind limb was scanned using a 3D scanner, and the resulting geometry was imported into finite element modeling software. A virtual tumor was added to the model, and the authors simulated the heating process using a surface probe. In addition to the heat transfer model, the researchers also developed a drug delivery model that simulates the kinetics of drug release. It is important to mention that the computed model was validated by performing experimental studies using gel phantoms and in vivo fluorescence imaging studies in mice with lung tumor xenografts. By comparing the results of the computer model with the experimental studies, the researchers can assess the accuracy of the model. The results showed that in silico modeling accurately reproduces the temperature profile observed in the phantom experiments and that the drug delivery profile simulated by the model also aligns with the results of the in vivo studies. Overall, it demonstrates the feasibility of using a computer model to accurately simulate drug delivery in preclinical studies [89][22].
To investigate the distribution of three drugs with different polarities (5-fluorouracil, ligustrazine, and osthole) within liposomes and transfersomes, researchers conducted a study using molecular dynamics simulation. To understand the drug distribution, these authors employed the radial distribution function—which calculates the probability of finding a drug molecule at a specific distance from a reference drug molecule within the vesicle—and the potential of mean force—which describes the potential energy between a drug molecule and the surrounding lipid molecules, indicating the strength of their interactions. By using these measures, the authors were able to characterize the distribution of drugs within the lipid vesicles. The results highlight the potential of molecular simulation technology in understanding the characteristics of lipid vesicles and their interactions with drugs [93][26].

2.2.3. Nanocrystals

A study aimed to develop and evaluate an advanced in silico modeling for understanding the pharmacokinetics of Foscan®, a formulation containing temoporfin that has received approval for palliative photodynamic therapy of squamous cell carcinoma of the head and neck. The researchers conducted precipitation experiments in the presence of biorelevant media, thereby simulating conditions akin to those encountered in the human body. This approach aimed to observe the behavior of Foscan® under these physiologically relevant circumstances. When introduced in these media, the drug underwent a process of precipitation, forming nanocrystals. Moreover, nanoparticle tracking analysis was employed to investigate these nanocrystals, providing the means to measure their size and analyze the distribution of these structures within the sample. Incorporating the data from these precipitation experiments and nanoparticle tracking analysis, the model predicted how nanocrystals of Foscan® were formed, their size distribution, and how they interacted with biological fluids in the body. This information could help them explain and predict the Foscan® pharmacokinetics more accurately, as nanocrystals can significantly impact how a drug is absorbed, distributed, and eliminated [94][27].
In another study, to evaluate the impact of polymers in the production of stable dexibuprofen (Dexi) nanocrystals with improved therapeutic potential, researchers combined in silico modeling techniques (namely AutoDockVina, Marven Sketch, and Maestro) with experimental studies. The results provided molecular insight into the mechanisms of binding of the optimal polymers to the surface of Dexi nanocrystals, showing that the combination of hydroxypropyl methylcellulose (HPMC)-polyvinyl pyrrolidone (PVP) and HPMC-Eudragit (EUD) was the most effective in stabilizing Dexi nanocrystals. Overall, the combination of computational modeling with experimental studies allows researchers to save time and resources by focusing on the most promising polymer combinations, thereby expediting the drug development process. Additionally, this integrated approach provides a deeper understanding of the molecular mechanisms underlying the stabilization of nanocrystals, helping researchers make more informed decisions in their pursuit of developing better pharmaceutical formulations [95][28].

2.3. Machine Learning

2.3.1. Polymeric Micelles

Machine learning algorithms can be utilized to predict various properties and behaviors of polymeric micelles. For example, models can be trained using data on polymer structure, composition, molecular weight, and other relevant parameters, along with experimental outcomes such as micelle size, stability, drug loading, and release profiles. These models can then be used to predict the behavior of new polymeric systems, guiding the design and selection of optimal micellar formulations [96][29].
Researchers used an artificial neural network (ANN) to create a model for the release of a chemotherapeutic drug—doxorubicin—from polymeric micelles (specifically Pluronic P105) under two different ultrasound frequencies. Although the exact number of samples used in the study was not explicitly mentioned, the model was trained using experimentally obtained input–output data concerning the release of doxorubicin from the micelles. The developed ANN model was then employed to optimize the application of ultrasound in order to achieve the desired drug release at the tumor site. The ANN method accurately predicted the release behavior and demonstrated maximum prediction errors of 0.002 and 0.001 at ultrasound frequencies of 20 and 70 kHz, respectively. The results demonstrate the successful design and testing of a controller capable of adjusting ultrasound frequency, intensity, and pulse length to maintain a constant release of Dox, potentially enhancing targeted drug delivery to tumor sites [97][30].

2.3.2. Liposomes/Niosomes

By analyzing large datasets of liposomal properties and characteristics, machine learning models can identify patterns and correlations between various liposome components (lipids and encapsulated substances, among others) and their properties (size, stability, drug release profile). This information can guide the selection of optimal liposome compositions and improve formulation success rates [98][31]. In addition, ML models can also be trained on existing data to predict important liposomes properties, which may include encapsulation efficiency, drug release kinetics, stability under different conditions, and targeting capabilities. By utilizing historical data and relevant features, machine learning algorithms can provide valuable insights and predictions, enabling more efficient and targeted liposome development [99][32].
In fact, there are some trials applying machine learning for liposome formulation optimization or prediction. For instance, one study proposed a machine learning framework to address the challenges associated with optimizing the drug entrapment efficiency of niosomal vesicles, showing that these algorithms allow for the synthesis of niosomal systems with optimal entrapment efficiency at a lower cost and time (Kashani-Asadi-Jafari et al., 2022). In another study, scientists built an artificial neural network (ANN) and advanced machine learning model to optimize the percentage of cytarabine entrapped in the liposome, showing that the ANN provided more accurate prediction formulations when compared with the multiple regression analysis method [100][33]. An ANN model was also developed to predict the size and polydispersity index of liposomes made of DOPC (1,2 Dioleoyl-sn-glycero-3-phosphocholine), cholesterol, and DSPE-PEG 2000 1,2 Distearoyl-sn-glycero-3-phosphoethanolamine-N [amino (polyethylene glycol)-2000] (ammonium salt)) using a microfluidic system. The results demonstrated that microfluidic-based preparation techniques assisted by computational tools can accelerate the development and clinical translation of nano-based pharmaceutical products [101][34].
Recently, machine learning has been combined with molecular descriptors, which are a set of quantitative values or features that represent various properties of a molecule’s structure, composition, and behavior. These are used to encode complex chemical information into numerical data, which can then be used as input for various computational analysis and machine learning models. Fundamentally, ML models leverage patterns within data to predict the properties of novel molecules, eliminating the need for physical synthesis or testing [98,102][31][35]. To illustrate this, an ANN was constructed to develop computational models focused on optimizing a continuous liposome manufacturing system. In this system, the liposomes were generated using a co-axial turbulent jet within a co-flow technology. This means that two phases were used—an ethanol phase with lipids and an aqueous phase—to create liposomes of uniform sizes. The ANN was used to optimize this manufacturing process and so, it took various input parameters known as critical material attributes (CMAs) and critical process parameters (CPPs). CMAs include characteristics of the raw materials, such as the length of the hydrocarbon tail in lipids, the percentage of cholesterol, and the type of buffer used. CPPs include process conditions, such as solvent temperature and flow rate. The ANN’s purpose was to predict critical quality attributes (CQAs) of the liposomes. In this study, the CQAs were the particle size and polydispersity index (PDI), which indicate how uniform the liposome sizes are. Thus, two types of ANN architectures were evaluated, namely a multiple-input–multiple-output (MIMO) model—which takes multiple inputs and produces multiple outputs—and a multiple-input–single-output (MISO) model—which takes multiple inputs but produces a single output. The study found that the MISO architecture outperformed the MIMO architecture in terms of accuracy for the task at hand. Apart from the ANN model, a graphical user interface was also created to help end-users perform interactive simulated risk analysis and visualize the predictions made by the ANN model. Evaluations demonstrated that the developed graphical user interface yields accurate predictions for both liposome particle size and PDI as long as the chosen inputs fall within the scope of the studied conditions during the initial ANN training. These predictions have the potential to contribute to the formulation of a control strategy designed to mitigate the impact of process disturbances on liposome particle size. Utilizing the five input features mentioned earlier, an ANN was trained with the primary goal of minimizing the mean relative error (MRE), which was successfully achieved at a level below 5%. It was notably a very low error for predicting particle size. Basically, despite the successful predictions for particle size, the model encountered challenges in accurately forecasting PDI values. To mitigate the training error, researchers introduced molecular descriptors as supplementary inputs to the ANN. These were obtained using PaDEL-Descriptor software and helped the ANN understand the characteristics of the raw materials. A combination of CMAs, CPPs, and molecular descriptors was used to train the MISO ANN model and allowed for the reduction of errors during both training and testing, indicating improved model performance. Overall, via a combination of critical material attributes, process parameters, and molecular descriptors, this study improved the accuracy of predicting the quality attributes of liposomes [103][36].
In another study, ML techniques were used to create prediction models capable of individually predicting crucial parameters of liposomes, such as size, PDI, zeta potential, and encapsulation efficiency. To validate the predictive prowess of these models, liposome formulations were created for two distinct compounds: naproxen (NAP) and palmatine HCL (PAL), representing insoluble and water-soluble molecules, respectively. In order to evaluate the significance of drug properties in liposome behavior, further investigation into the molecular interactions and behaviors of NAP and PAL within liposomes was undertaken via coarse-grained molecular dynamics simulations. These simulations demonstrated that NAP molecules tend to integrate into the lipid layer, while a majority of PAL molecules aggregate within the inner aqueous phase of the liposome. The marked disparity in the physical states of NAP and PAL underlines the pivotal role of drug properties in formulating liposomes. Additionally, formulation attributes were ranked to offer significant insights for designing effective formulations. Given that logS (logarithm of a compound’s aqueous solubility), molecular complexity (an assessment of the intricacy of a structure), and XLogP3 (represent a predictive estimation of the octanol–water partition coefficient, determined via a specific algorithm) of the drug molecules held significant influence over encapsulation efficiency, their correlation was illustrated using a heatmap. This heatmap employed color visualization in a two-dimensional format to depict the data relationship effectively. Basically, drug molecules with certain properties, such as a logS value between −3 and −6, a molecular complexity between 500 and 1000, and a XLogP3 value greater than or equal to 2, are considered a priority for formulating liposomes with better encapsulation. Finally, it is possible to observe a congruence between predicted and experimental outcomes, which serves as confirmation of the ML model’s satisfactory accuracy. In summary, the researchers established comprehensive prediction models for anticipating liposome formulations, and the influences of key factors were dissected by combining ML techniques with molecular modeling. The study successfully validates the availability and rationality of these intelligent prediction systems, offering promising applications for the future development of liposome formulations [98][31].
Based on all of that, it is safe to say that machine learning plays a valuable role in the development of liposomes by assisting in formulation design, predicting liposome properties, optimizing drug loading and release, analyzing characterization data, and optimizing manufacturing processes. By leveraging machine learning techniques, researchers can expedite the development and improve the performance of liposomal formulations for drug delivery and other biomedical applications [104][37].

2.3.3. Nanocrystals

Machine learning techniques have been increasingly employed in the field of nanocrystal development. For instance, they can be trained to predict the properties of nanocrystals by using data from a variety of sources, including experimental measurements, theoretical calculations, and molecular descriptors. These predictive models can assist in the design and selection of nanocrystals with desired properties, saving time and resources by reducing the need for extensive experimental testing. In fact, to address this issue, researchers collected data on nanocrystal size (910 data points) and polymer dispersity index (341 data points) using three different preparation methods—ball wet milling (BWM), high-pressure homogenization (HPH), and antisolvent precipitation (ASP)—in order to construct prediction models [105][38].
The results indicated that the machine learning performed well in predicting those properties for BWM and HPH methods but showed relatively poor predictions for the ASP method. Within the BWM subsets, the predicted values closely matched the experimental values in the size range of 0–500 nm. Nevertheless, for data points outside of this range, particularly those exceeding 500 nm, predicted values displayed significant disparities from the experimental values, especially in the validation and test datasets. The findings suggest that the constructed learning model exhibits superior predictive accuracy for data points falling within the 0–500 nm range, particularly where the data density is higher in the BWM size dataset. It is evident that the uneven distribution of data within the dataset significantly influences the model’s construction and predictive performance. On the other hand, due to the limited availability of input data within the size range of 500 to 1000 nm, learning algorithms struggled to discern the underlying data patterns, leading to less accurate predictions. The scatter plots for the HPH and ASP subsets showed comparable results, with predicted values closely aligning with experimental values in regions where data density was higher [105][38].
On the other hand, displays scatter plots comparing the predicted values to the experimental values of PDI data. The machine learning algorithm performed admirably in predicting this property values within the BWM and HPH subsets, as indicated by the close alignment of data points with the black line. Despite the relatively small datasets with only 133 and 119 data points in the BWM and HPH PDI datasets, respectively, the algorithm still demonstrated accurate predictions, likely due to the concentrated data distribution. Conversely, in the ASP subsets, the predictive performance in the training set was notably less accurate. This could be attributed to the smaller volume of data and the lower data quality within these subsets [105][38].
The researchers speculated that the poor prediction for the ASP method might be due to the lower quality of data resulting from the poor reproducibility and instability of nanocrystals prepared using this method. It was also found that the majority of commercialized nanocrystals products were manufactured using BWM and HPH approaches. ML helped rank the factors influencing nanocrystal properties, indicating that milling time, cycle index, and stabilizer concentration were crucial factors for the nanocrystals prepared by BWM, HPH, and ASP methods, respectively. The accuracy of these predictions was further confirmed by experiments with newly prepared nanocrystals. Concerning the nanocrystals size, the findings indicate that the algorithm delivered accurate predictions for most of the nanocrystals produced through the BWM and HPH methods. However, the predicted performance for ASP nanocrystals fell short, with predicted values being roughly twice as large as the experimentally measured sizes. Regarding PDI predictions, the algorithm also demonstrated effective forecasting for BWM and HPH datasets, but showed comparatively poorer performance for ASP. The most likely reasons for this subpar predictive performance were the limited volume of data available for ASP size and PDI datasets used in constructing machine learning models, and the lower data quality stemming from the constraints of the preparation methods. These issues ultimately led to the failure of predictions within the ASP dataset.
Overall, the results highlights the potential of using machine learning in nanotechnology manufacturing, providing a promising alternative to traditional, labor-intensive approaches in nanocrystal formulation development [105][38].

2.4. Digital Twins

Digital twins have emerged as a transformative technology in the pharmaceutical industry, offering a powerful tool for improving drug development processes and product lifecycle management [106][39]. In the realm of traditional pharmaceuticals, their significance is already well recognized, as digital twins enable the creation of virtual replicas of physical drugs and manufacturing processes, facilitating real-time monitoring, optimization, and quality control [39,40][40][41]. However, it is important to highlight the lack of experimental studies on the application of these technologies in the NBCD domain. This is an exciting frontier for innovation and has the potential to significantly advance the development, manufacturing, and quality assurance of these pharmaceutical products.

References

  1. Hariry, R.E.; Barenji, R.V.; Paradkar, A. Towards Pharma 4.0 in clinical trials: A future-orientated perspective. Drug Discov. Today 2022, 27, 315–325.
  2. Sharma, M.; Sehrawat, R.; Luthra, S.; Daim, T.; Bakry, D. Moving Towards Industry 5.0 in the Pharmaceutical Manufacturing Sector: Challenges and Solutions for Germany. IEEE Trans. Eng. Manag. 2022, 1–18.
  3. Arden, N.S.; Fisher, A.C.; Tyner, K.; Yu, L.X.; Lee, S.L.; Kopcha, M. Industry 4.0 for pharmaceutical manufacturing: Preparing for the smart factories of the future. Int. J. Pharm. 2021, 602, 120554.
  4. Sharma, D.; Patel, P.; Shah, M. A comprehensive study on Industry 4.0 in the pharmaceutical industry for sustainable development. Environ. Sci. Pollut. Res. 2023, 30, 90088.
  5. Grangeia, H.B.; Silva, C.; Simões, S.P.; Reis, M.S. Quality by design in pharmaceutical manufacturing: A systematic review of current status, challenges and future perspectives. Eur. J. Pharm. Biopharm. 2020, 147, 19–37.
  6. Inuwa, H.M.; Ravi Raja, A.; Kumar, A.; Singh, B.; Singh, S. Status of Industry 4.0 applications in healthcare 4.0 and Pharma 4.0. Mater. Today Proc. 2022, 62, 3593–3598.
  7. Wang, W.; Ye, Z.; Gao, H.; Ouyang, D. Computational pharmaceutics—A new paradigm of drug delivery. J. Control. Release 2021, 338, 119–136.
  8. Mak, K.K.; Pichika, M.R. Artificial intelligence in drug development: Present status and future prospects. Drug Discov. Today 2019, 24, 773–780.
  9. Paul, D.; Sanap, G.; Shenoy, S.; Kalyane, D.; Kalia, K.; Tekade, R.K. Artificial intelligence in drug discovery and development. Drug Discov. Today 2021, 26, 80–93.
  10. Almeida, A.; Linares, V.; Mora-Castaño, G.; Casas, M.; Caraballo, I.; Sarmento, B. 3D printed systems for colon-specific delivery of camptothecin-loaded chitosan micelles. Eur. J. Pharm. Biopharm. 2021, 167, 48–56.
  11. Linares, V.; Casas, M.; Caraballo, I. Printfills: 3D printed systems combining fused deposition modeling and injection volume filling. Application to colon-specific drug delivery. Eur. J. Pharm. Biopharm. 2019, 134, 138–143.
  12. Cisneros, K.; Chowdhury, N.; Coleman, E.; Ferdous, T.; Su, H.; Jennings, J.A.; Bumgardner, J.D.; Fujiwara, T. Long-Term Controlled Release of Simvastatin from Photoprinted Triple-Networked Hydrogels Composed of Modified Chitosan and PLA–PEG Micelles. Macromol. Biosci. 2021, 21, 2100123.
  13. Sommonte, F.; Weaver, E.; Mathew, E.; Denora, N.; Lamprou, D.A. In-House Innovative “Diamond Shaped” 3D Printed Microfluidic Devices for Lysozyme-Loaded Liposomes. Pharmaceutics 2022, 14, 2484.
  14. Wang, Z.; Liu, L.; Xiang, S.; Jiang, C.; Wu, W.; Ruan, S.; Du, Q.; Chen, T.; Xue, Y.; Chen, H.; et al. Formulation and Characterization of a 3D-Printed Cryptotanshinone-Loaded Niosomal Hydrogel for Topical Therapy of Acne. AAPS PharmSciTech 2020, 21, 159.
  15. Groetsch, A.; Stelzl, S.; Nagel, Y.; Kochetkova, T.; Scherrer, N.C.; Ovsianikov, A.; Michler, J.; Pethö, L.; Siqueira, G.; Nyström, G.; et al. Microscale 3D Printing and Tuning of Cellulose Nanocrystals Reinforced Polymer Nanocomposites. Small 2023, 19, e2202470.
  16. Germini, G.; Peltonen, L. 3D Printing of Drug Nanocrystals for Film Formulations. Molecules 2021, 26, 3941.
  17. Phan, V.H.G.; Murugesan, M.; Huong, H.; Le, T.T.; Phan, T.H.; Manivasagan, P.; Mathiyalagan, R.; Jang, E.S.; Yang, D.C.; Li, Y.; et al. Cellulose Nanocrystals-Incorporated Thermosensitive Hydrogel for Controlled Release, 3D Printing, and Breast Cancer Treatment Applications. ACS Appl. Mater. Interfaces 2022, 14, 42812–42826.
  18. Lopez-Vidal, L.; Real, J.P.; Real, D.A.; Camacho, N.; Kogan, M.J.; Paredes, A.J.; Palma, S.D. Nanocrystal-based 3D-printed tablets: Semi-solid extrusion using melting solidification printing process (MESO-PP) for oral administration of poorly soluble drugs. Int. J. Pharm. 2022, 611, 121311.
  19. Tambe, S.M.; Jain, D.D.; Hasmukh Mehta, C.; Ashwini, T.; Yogendra Nayak, U.; Amin, P.D. Hot-melt extruded in situ gelling systems (MeltDrops Technology): Formulation development, in silico modelling and in vivo studies. Eur. J. Pharm. Biopharm. 2023, 188, 108–124.
  20. Duran, T.; Costa, A.; Gupta, A.; Xu, X.; Zhang, H.; Burgess, D.; Chaudhuri, B. Coarse-Grained Molecular Dynamics Simulations of Paclitaxel-Loaded Polymeric Micelles. Mol. Pharm. 2022, 19, 1117–1134.
  21. Bhendale, M.; Singh, J.K. Molecular Insights on Morphology, Composition, and Stability of Mixed Micelles Formed by Ionic Surfactant and Nonionic Block Copolymer in Water Using Coarse-Grained Molecular Dynamics Simulations. Langmuir 2023, 39, 5031–5040.
  22. Ramajayam, K.K.; Wolfe, A.M.; Motamarry, A.; Yost, J.; Yost, M.J.; Haemmerich, D. Computer Modeling of Drug Delivery with Thermosensitive Liposomes in a Realistic Three-Dimensional Geometry. Proc. Annu. Int. Conf. IEEE Eng. Med. Biol. Soc. EMBS 2020, 2020, 5021–5024.
  23. Cern, A.; Marcus, D.; Tropsha, A.; Barenholz, Y.; Goldblum, A. New drug candidates for liposomal delivery identified by computer modeling of liposomes’ remote loading and leakage. J. Control. Release 2017, 252, 18–27.
  24. Bunker, A.; Magarkar, A.; Viitala, T. Rational design of liposomal drug delivery systems, a review: Combined experimental and computational studies of lipid membranes, liposomes and their PEGylation. Biochim. Biophys. Acta—Biomembr. 2016, 1858, 2334–2352.
  25. Pathak, P.; Dhawan, V.; Magarkar, A.; Danne, R.; Govindarajan, S.; Ghosh, S.; Steiniger, F.; Chaudhari, P.; Gopal, V.; Bunker, A.; et al. Design of cholesterol arabinogalactan anchored liposomes for asialoglycoprotein receptor mediated targeting to hepatocellular carcinoma: In silico modeling, in vitro and in vivo evaluation. Int. J. Pharm. 2016, 509, 149–158.
  26. Wu, X.; Dai, X.; Liao, Y.; Sheng, M.; Shi, X. Investigation on drug entrapment location in liposomes and transfersomes based on molecular dynamics simulation. J. Mol. Model. 2021, 27, 111.
  27. Jablonka, L.; Ashtikar, M.; Gao, G.; Jung, F.; Thurn, M.; Preuß, A.; Scheglmann, D.; Albrecht, V.; Röder, B.; Wacker, M.G. Advanced in silico modeling explains pharmacokinetics and biodistribution of temoporfin nanocrystals in humans. J. Control. Release 2019, 308, 57–70.
  28. Ullah, N.; Khan, S.; Ahmed, S.; Govender, T.; Faidah, H.S.; De Matas, M.; Shahid, M.; Minhas, M.U.; Sohail, M.; Khurram, M. Dexibuprofen nanocrystals with improved therapeutic performance: Fabrication, characterization, in silico modeling, and in vivo evaluation. Int. J. Nanomed. 2018, 13, 1677–1692.
  29. Jones, D.E.; Ghandehari, H.; Facelli, J.C. A review of the applications of data mining and machine learning for the prediction of biomedical properties of nanoparticles. Comput. Methods Programs Biomed. 2016, 132, 93–103.
  30. Husseini, G.A.; Mjalli, F.S.; Pitt, W.G.; Abdel-Jabbar, N.M. Using ANN and MPC to optimize acoustically assisted Doxorubicin release from micelles. Proc. IASTED Int. Conf. Model. Simul. 2009, 8, 320–325.
  31. Han, R.; Ye, Z.; Zhang, Y.; Cheng, Y.; Zheng, Y.; Ouyang, D. Predicting liposome formulations by the integrated machine learning and molecular modeling approaches. Asian J. Pharm. Sci. 2023, 18, 100811.
  32. Kashani-Asadi-Jafari, F.; Aftab, A.; Ghaemmaghami, S. A machine learning framework for predicting entrapment efficiency in niosomal particles. Int. J. Pharm. 2022, 627, 122203.
  33. Subramanian, N.; Yajnik, A.; Murthy, R.S.R. Artificial neural network as an alternative to multiple regression analysis in optimizing formulation parameters of cytarabine liposomes. AAPS PharmSciTech 2004, 5, E4.
  34. Rebollo, R.; Oyoun, F.; Corvis, Y.; El-Hammadi, M.M.; Saubamea, B.; Andrieux, K.; Mignet, N.; Alhareth, K. Microfluidic Manufacturing of Liposomes: Development and Optimization by Design of Experiment and Machine Learning. ACS Appl. Mater. Interfaces 2022, 14, 39736–39745.
  35. Baptista, D.; Correia, J.; Pereira, B.; Rocha, M. Evaluating molecular representations in machine learning models for drug response prediction and interpretability. J. Integr. Bioinform. 2022, 19, 20220006.
  36. Sansare, S.; Duran, T.; Mohammadiarani, H.; Goyal, M.; Yenduri, G.; Costa, A.; Xu, X.; O’Connor, T.; Burgess, D.; Chaudhuri, B. Artificial neural networks in tandem with molecular descriptors as predictive tools for continuous liposome manufacturing. Int. J. Pharm. 2021, 603, 120713.
  37. Rathore, A.S.; Nikita, S.; Thakur, G.; Mishra, S. Artificial intelligence and machine learning applications in biopharmaceutical manufacturing. Trends Biotechnol. 2023, 41, 497–510.
  38. He, Y.; Ye, Z.; Liu, X.; Wei, Z.; Qiu, F.; Li, H.F.; Zheng, Y.; Ouyang, D. Can machine learning predict drug nanocrystals? J. Control. Release 2020, 322, 274–285.
  39. Attaran, M.; Attaran, S.; Celik, B.G. The impact of digital twins on the evolution of intelligent manufacturing and Industry 4.0. Adv. Comput. Intell. 2023, 3.
  40. Schütt, M.; Stamatopoulos, K.; Batchelor, H.K.; Simmons, M.J.H.; Alexiadis, A. Development of a digital twin of a tablet that mimics a real solid dosage form: Differences in the dissolution profile in conventional mini-USP II and a biorelevant colon model. Eur. J. Pharm. Sci. 2022, 179, 106310.
  41. Moreno-Benito, M.; Lee, K.T.; Kaydanov, D.; Verrier, H.M.; Blackwood, D.O.; Doshi, P. Digital twin of a continuous direct compression line for drug product and process design using a hybrid flowsheet modelling approach. Int. J. Pharm. 2022, 628, 122336.
More
© Text is available under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/)
Video Production Service
Feedback