Artificial Intelligence for Academic Advanced Therapy Medicinal Products

Version	Summary	Created by	Modification	Content Size	Created at	Operation
1		Cristobal Aguilar-Gallardo	--	3094	2024-02-21 16:05:21	\|
2	format	Jason Zhu	Meta information modification	3094	2024-02-23 03:16:17	\|

This entry is adapted from the peer-reviewed paper 10.3390/app14031303

Advanced therapy medicinal products (ATMPs), encompassing cell and gene therapies, hold immense promise in revolutionizing treatment options for numerous diseases. However, the translation of these innovative therapies from research to clinical practice faces formidable challenges, necessitating the establishment of specialized manufacturing facilities within hospital settings. Hospital-based Good Manufacturing Practice (GMP) facilities offer a unique advantage by enabling rapid bench-to-bedside development and direct patient access, yet their implementation necessitates significant adaptation within healthcare infrastructures, constrained by spatial limitations, regulatory requirements, and resource allocation. Key considerations and solutions for overcoming these challenges are paramount for realizing the full potential of ATMPs. Proximity afforded by on-site manufacturing facilitates direct delivery to patients, mitigating risks associated with transportation and ensuring timely access to cutting-edge therapies. Moreover, aligning production schedules with clinical demands enhances flexibility and responsiveness to patient needs. However, the integration of pharmaceutical manufacturing within hospital environments necessitates addressing critical gaps in staff training, documentation practices, and oversight, which are inherent to the highly regulated pharmaceutical industry. Significant investments in infrastructure, specialized equipment, personnel training, and multi-departmental coordination are indispensable for establishing and maintaining robust hospital ATMP facilities. Embracing technological advancements, such as process analytical technology (PAT), continuous manufacturing, and artificial intelligence (AI), holds immense potential in bolstering the efficiency, quality, and safety of ATMP production processes. AI, in particular, offers the capability to analyze vast datasets generated during manufacturing, facilitating real-time prediction of product quality attributes and enabling automated adjustments through feedback control mechanisms. Despite these technological advancements, challenges persist in integrating AI into pharmaceutical manufacturing due to concerns regarding data security, regulatory compliance, and the requisite multidisciplinary collaboration. Successful adoption of AI technologies necessitates simultaneous investment in human capital to ensure effective implementation and governance. Ultimately, the convergence of innovative manufacturing technologies and synergistic partnerships across disciplines is paramount for realizing the transformative potential of ATMPs, ensuring their responsible translation from research to clinical impact while maximizing therapeutic safety and efficacy.

bioprocess advanced therapy medicinal products (ATMPs) artificial intelligence (AI) Cell and Gen Therapy

1. Process Monitoring

The personalized nature of autologous ATMPs results in inherent variability between patient cell samples used for manufacturing. This variability in starting material can propagate through the process, making achieving consistency between individualized batches challenging. Conventional bioprocess monitoring typically relies on manual and periodic offline sample analysis. AI enables improved real-time process oversight through advanced sensor instrumentation coupled with intelligent algorithms continuously analyzing data for patterns predictive of critical quality attributes.

Multivariate AI models imply that machine learning algorithms can integrate data streams from arrays of in-line sensors and bioreactors probe analyzers, to identify inter-relationships between bio-parameters and product quality. This higher-resolution view allows slight process deviations to be detected earlier and corrections to be made through automated feedback control, rather than periodic and manual sampling. Thus, AI process monitoring provides complete tracking and mapping of the multidimensional course of cell bioproduction. ML techniques can also incorporate spectroscopic ^[1], imaging, and other process data, to build predictive models of final product critical quality attributes such as identity, purity, and potency ^[2]. This real-time predictive capability facilitates standardized quality and reduced variability within and between batches, as well as reduced batch failures.

In the exploration of AI’s impact on ATMP production, researchers can observe distinct advantages and challenges, as outlined in Table 1. Enhanced process monitoring and the automation of manufacturing tasks are among the key benefits, leading to increased efficiency and reduced workload for biotechnologists. Moreover, AI’s dynamic control and adaptability allow fine-tuning cell culture conditions, thus ensuring batch consistency, which is critical for patient-specific treatments. On the other hand, issues such as data security concerns and the requirement for specialized personnel present noteworthy obstacles, potentially hindering AI’s integration.

Table 1. Advantages and disadvantages of AI in ATMP production.

Advantages of AI in ATMP Production	Disadvantages of AI in ATMP Production
Enhanced Process Monitoring	Data Security Concerns
Real-time oversight and data analysis for maintaining consistency and quality.	Risks associated with patient data privacy and data integrity in AI systems.
Automation of Manufacturing Tasks	High Initial Investment
Increases efficiency, precision, and scalability, reducing the workload of biotechnologists.	Significant upfront costs for AI integration and infrastructure development.
Dynamic Control and Adaptability	Limited Data in Early Development
AI systems can adjust to the variability in patient cell samples, optimizing batch consistency.	AI models may lack accuracy in early stages due to insufficient data for machine learning.
Improved Data Management	Requirement for Specialized Personnel
Effective handling of large datasets, enhancing process comprehension and decision-making.	Need for staff with expertise in AI and bioprocessing, who can be scarce.
Systems Biology Integration	Regulatory Challenges
AI aids in understanding complex biological relationships, enhancing predictive medicine.	Ambiguity in regulatory compliance for emerging AI applications in healthcare.

A further analysis is presented in Table 2 of the ATMP production process with and without AI integration. It is evident that AI integration significantly enhances process monitoring and manufacturing efficiency, offering a more sophisticated approach to data management, which traditional methods lack. The enhanced quality and consistency afforded by AI contribute to better control of product integrity; nonetheless, there are higher investment and costs required for implementing such advanced technologies. The need for personnel training and expertise remains an essential consideration to fully realize the potential of AI in this field.

Table 2. Comparison of ATMP production with and without AI integration.

Aspect of ATMP Production	With AI Integration	Without AI Integration
Process Monitoring	Enhanced real-time monitoring and analysis using AI algorithms.	Relies on manual and periodic offline sample analysis.
Manufacturing Efficiency	Higher efficiency and scalability due to automation and dynamic control.	Less efficient, often limited by manual operations and static processes.
Data Management	Advanced handling of large and complex datasets, facilitating better decision-making.	Traditional data management, potentially leading to slower and less informed decisions.
Quality and Consistency	Improved product quality and batch consistency through predictive models and real-time adjustments.	Potential variability and quality issues due to lack of real-time monitoring and control.
Investment and Costs	Higher initial investment but potential long-term cost savings through efficiency and reduced error rates.	Lower initial costs but potentially higher long-term operational expenses.
Personnel Training and Expertise	Requires staff trained in AI and data science.	Relies more on traditional bioprocessing skills.

2. Automation

The manufacturing of ATMPs involves numerous supplementary operations beside the bioprocess itself, including supply chain management, sample tracking, the changeover between batches, contamination control, equipment maintenance, regulatory compliance adherence, quality control, batch record reviewing, and data management. AI enables automation of many tasks, to enhance precision and consistency, increase efficiency, allow scalability according to production demands, improve cost-effectiveness, and reduce biotechnology personnel workloads. Robotic process automation can replicate and mimic administrative and repetitive activities like data entry, basic data manipulation, and report generation. ML algorithms can schedule planned equipment calibration, preventive maintenance, and cleaning procedures based on the runtime. Cleanroom environmental monitoring data analyzed by AI models can provide alerts on contamination risks and suggest corrective actions, to avoid batch losses or disruptions to production.

Automated microscopy employing computer vision techniques permits precise characterization of cells. The searching of batch records by an AI makes use of natural language processing and provides appropriate results related to valuable data. Augmented reality helps operators carry out critical manual operations, in order to minimize mistakes. Multifaceted AI-driven automation can be used to address the spatial and staffing limitations that academic ATMP facilities often experience. Flexible cell-factories able to manufacture multiple individualized therapies in accordance with the specific requirements of a healthcare center can also be encompassed within ML methods, instead of following large-scale production. With intelligent automation, workforces can become more productive using limited staffing, while facilities can be made more efficient.

3. Dynamic Control

Adaptive bioprocess control systems are critical for ATMP manufacturing, given the inevitable variability between patient cell samples. Traditional bioreactors rely on feedback and control loops tracking setpoint values ^[3]^[4]. However, the optimal parameters and tolerances may differ for each autologous batch’s unique biologic characteristics. AI allows more advanced control schemes that continuously recalculate in real time the ideal setpoint based on multivariate analysis. ML algorithms can discover relationships between sensor data, bioprocess parameters, and critical quality attributes of the products in bioreactors. Through this approach, a prediction model of the final product’s quality can guide dynamic adjustments of process variables in real time, leading to maximized batch consistency and robustness.

Hybrid AI strategies combine mechanistic first-principles models of cell growth and metabolism with data-driven ML models that capture the existing variability in the data. These models are commonly used for research and learning purposes, to gain insights into the intrinsic behavior of biological systems. Thus, adaptive model predictive control integrates these models to continuously re-optimize process parameters based on current state measurements of the process. Thus, as more batches are produced, semi-supervised ML can improve the controller model. AI control systems are particularly well-suited for handling the high variability and lack of expertise that are frequently experienced during early ATMP process development and technology transfer, because of their capacity for self-learning. The capability of AIs in handling complexity allows flexible and sophisticated adaptive control, especially when dealing with the specific needs of tailored batches ^[5].

4. Data Management

ATMP manufacturing fundamentally relies on synthesizing various databases, starting from the clinical status of the incoming patient sample and ending with the final product testing results. The correlation between clinical indicators and production factors can have an impact on product quality and treatment outcomes. Traditional data infrastructure faces challenges due to the vast amount of data produced. This data comes from various sources: sensors in process arrays, omics analyses, and electronic records. Additionally, the need to track and trace this data adds to the complexity, creating stress and pressure on existing systems. In this regard, AI can provide tools suited for managing and handling heterogeneous structured, unstructured, and semi-structured data. ML is efficient for identifying imperceptible patterns and correlations within massive datasets. AI can integrate electronic batch records, processing data, laboratory results, clinical data, and other sources into contextualized knowledge that enhances process comprehension. Well-known techniques such as natural language processing permit mining unstructured textual records. AI can enhance data management through dimensionality reduction, noise filtering, missing data imputation, and real-time data cleansing, to consolidate relevant information from discordant sources and that is apparently unrelated into usable and applicable models, guiding decision making and controlling adjustments to optimize each personalized batch. Thus, the integration of AI’s multifaceted capabilities, including knowledge contextualization, data mining, dimensionality reduction, and real-time data management, highlights its strengths for advancing process understanding and facilitating optimal real-time decision-making for the production of personalized gene therapy ATMPs.

5. Systems Biology and AI

The aim of systems biology is developing complex models that integrate data from multiple disciplines, in order to explain complex biologic relationships or transactions. This is contrary to the reductionist approach of the twentieth century that gave insight into some areas but not in full regarding the comprehension of complexities and interpretation of non-structured elements. In addition to other methods, systems and network biology offer a unique approach to analyze multi-omics data. This method integrates various data types to uncover new patterns and behaviors in complex molecular and cellular networks. It does this by utilizing machine learning (ML) algorithms, which helps in understanding these complex biological systems more effectively. Network biology empowered by AI is important for describing the associations within normal and dysfunctional phenotypes. It aspires to clearer, explicit, and deterministic models, in order to further predictive, preventive, and personalized medicine. The large volume of multi-omics data makes functional integration impracticable manually, necessitating advanced analytics like network analysis, Bayesian methods, and multivariate techniques—now powered by AI.

6. Advanced Control Systems Leveraging AI

Recently, bioreactors have employed automatic proportional-integral-derivative (PID) control for parameters like temperature, while others require manual regulation ^[6]. The progress in control strategies, optimization algorithms, and software systems has been significant. Contemporary bioreactors often incorporate SCADA systems requiring a human–machine interface ^[7]^[8]. Newer systems are starting to implement adaptive model-based controllers, providing two key advantages: (i) optimizing constraints and control signal ranges; (ii) dynamically adjusting actions based on control outcomes and system changes. Their strength is in continuously recalculating the optimal next steps during operation. The development of distributed control systems in communication protocols is gradually progressing. The combination of increased computing power and a wider array of algorithms now allows the implementation of iterative control techniques. This employs an extensive pool of data obtained during monitoring and a deeper understanding of cell culture behavior, facilitating the creation of more precise and sensitive AI-driven models using systems that integrate ML algorithms, computer vision, and the other AI techniques described above. Basically, AI plays a central role in managing the variability and complexity within the ambit of ATMP manufacturing. It is undeniable that AI can greatly benefit academic ATMP production, but previously it required proficient control over different facets. As previously discussed, AI can enhance process monitoring, automation, control, and data integration for personalized therapies. However, implementing AI in this sector clearly demands rigorous oversight of technology capabilities, data accuracy, personnel competencies, and regulatory adherence.

A cross-functional understanding of AI maturity is crucial. Models like the Artificial Intelligence Optimization Team’s ^[5]^[9]^[10] AI maturity level characterization model allow an organization to evaluate its current AI proficiencies across manufacturing categories, identifying gaps ^[11]. This assessment enables strategic roadmaps for developing capabilities required for AI tools. Rather than algorithm development, the focus is on integrating AI into operational functions. Structured maturity evaluation also tracks progress over time.

AI depends on quality data ^[12] and data integrity, which are critical for patient information. Systems that ensuring privacy and effective data governance are essential before launching any AI initiatives. Data security vulnerabilities require specific and exhaustive risk analysis, as well as controls for AI/ML applications. For instance, the authentication of data provenance, quality, and security as just as important a qualifier for infrastructure. AI models continue to stay in line with changing processes by constantly curating and maintaining data. High-quality data and its integrity are crucial in AI implementation in a clinical environment as this entails sensitive patients’ information. As the basis of any health AI-driven initiative, it is essential to establish strong systems, which will also guarantee privacy safety and management procedures. AI and machine learning apps that deal with medical data contain a high level of data security vulnerabilities, which require comprehensive risk assessment and customized controls specific for AI/ML in medicine.

However, within the scope of processes controlled by AI, especially the initial steps such as academic ATMP process development, this may be considered a disadvantage of applying ML techniques. The batches of data mined and collected from industries such as pharmaceuticals manufacturing can train and improve numerous ML models. Nevertheless, when the number of batches is limited, such as in academic production, the quality of training a better model becomes limited.

Therefore, in this scenario, ML algorithms may not be able to take all possible scenarios into consideration, resulting in inaccuracies in application. Under such circumstances, predictions become inaccurate, there is less adaptability to changing conditions, adaptability is low, and the insight into underlining processes is minimal, since this depends on insufficient data. Consequently, in order to improve the effectiveness of ML applications, it is imperative to increase the quantity of manufacturing batches or find alternative means to increase data collection. One available alternative is consolidation of a Healthcare center network established to, not only to improve health outcomes among patients, but also enhance ATMP quality. Within this network of hospitals, various nodes would collaborate by exchanging information and sharing expertise, to advance ATMP manufacturing without compromising on quality and safety.

Focusing on data sharing between the nodes in a network via the implementation and usage of blockchain technology for secure data sharing can enable an encrypted system for data distribution, which can be a tool for information exchange that is suitable for being validated by regulatory agencies. This immutable and private process would ensure that both raw or processed data are available for ML analysis and enhanced cell processing. This innovative application of blockchain’s immutable and private framework could not only enhance data security but also update the process of data sharing, ensuring accuracy and integrity in the handling of sensitive data across the network ^[13]^[14]^[15].

Reskilling biotechnologist staff to implement augmented intelligence, while managing additional risks, is critical. AI success requires personnel fluent in its applications and limitations, especially for regulated manufacturing. Cross-functional collaboration between biotechnologists and AI experts can promote effective adoption of AI and ML. Thus, institutional cybersecurity defense requires a deep bench of talent specialized in AI/ML data risks. Specialized personnel combining computational proficiency with bioprocessing expertise remain scarce yet essential for overseeing responsible AI adoption. Most researchers lack formal data science training, while data analysts are often detached from manufacturing realities. Fostering collaborative groups between bioengineers, computer scientists, and clinicians could promote the development of truly tailored systems. Academic curriculums must also evolve, integrating statistics and bioinformatics skills with traditional knowledge. Some leading educational institutions have already implemented introductory data science courses into cell and gene therapy majors. For current personnel, institutions could offer continuing education programs through partnerships with technical universities (https://www.theattcnetwork.co.uk, accessed on 14 January 2024). Proactive reskilling builds the competencies needed to securely apply AI towards enhancing ATMP translation, rather than simply reducing costs. Therefore, investment in people and expertise is crucial, along with cutting-edge technology.

Regulatory authorities encourage AI when the benefits are demonstrated with proper GMP governance. However, ambiguity exists around emerging applications. Continuous communication, documented risk management, and a focus on enhancing quality help ensure compliance. Ultimately, AI for ATMPs requires synchronized integration across technology, data, people, regulation, and health care objectives—comprehensive management of this convergence would allow a controlled evolution toward more intelligent manufacturing ^[16].

Furthermore, a focus on product quality improvement must be a priority in AI integration. However, if appropriately designed and adjusted, AI systems can make processes more efficient, less prone to error, and consistent, thus improving product quality and safety. This is a proactive approach that guarantees compliance, while ensuring the healthcare system provides safe and quality services to patients.

Achieving the successful implementation of AI in relation to the manufacturing of ATMPs relies on the synchronous integration of various spheres. These areas include the implementation of cutting-edge technological solutions in healthcare, responsible data management, health practitioner cooperation, strict regulation compliance, and the alignment of AI strategies with other academic and public health players. Proper management of this diverse convergence is a challenge, as each requires explicit leadership, operational procedures, and well-defined roles and responsibilities that are coherent in a complex academic–hospital ecosystem ^[17].

References

Kern, S.; Wander, L.; Meyer, K.; Guhl, S.; Mukkula, A.R.G.; Holtkamp, M.; Salge, M.; Fleischer, C.; Weber, N.; King, R.; et al. Flexible automation with compact NMR spectroscopy for continuous production of pharmaceuticals. Anal. Bioanal. Chem. 2019, 411, 3037–3046.
Feng Báez, J.P.; George De la Rosa, M.V.; Alvarado-Hernández, B.B.; Romañach, R.J.; Stelzer, T. Evaluation of a compact composite sensor array for concentration monitoring of solutions and suspensions via multivariate analysis. J. Pharm. Biomed. Anal. 2023, 233, 115451.
Rathore, A.S.; Mishra, S.; Nikita, S.; Priyanka, P. Bioprocess Control: Current Progress and Future Perspectives. Life 2021, 11, 557.
Jabarivelisdeh, B.; Carius, L.; Findeisen, R.; Waldherr, S. Adaptive predictive control of bioprocesses with constraint-based modeling and estimation. Comput. Chem. Eng. 2020, 135, 106744.
Sarker, I.H. AI-Based Modeling: Techniques, Applications and Research Issues Towards Automation, Intelligent and Smart Systems. SN Comput. Sci. 2022, 3, 158.
Chevalier, M.; Gómez-Schiavon, M.; Ng, A.H.; El-Samad, H. Design and Analysis of a Proportional-Integral-Derivative Controller with Biological Molecules. Cell Syst. 2019, 9, 338–353.e10.
Mitra, S.; Murthy, G.S. Bioreactor control systems in the biopharmaceutical industry: A critical perspective. Syst. Microbiol. Biomanuf. 2022, 2, 91–112.
Mujawar, S.; Deshpande, A.; Gherkar, A.; Simon, S.E.; Prajapati, B. Introduction to Human-Machine Interface. In Human-Machine Interface; John Wiley & Sons, Ltd.: Hoboken, NJ, USA, 2023; pp. 1–23. ISBN 9781394200344.
Zheng, Y. Optimization of computer programming based on mathematical models of artificial intelligence algorithms. Comput. Electr. Eng. 2023, 110, 108834.
Heaton, H.; Fung, S.W. Explainable AI via learning to optimize. Sci. Rep. 2023, 13, 10103.
Cheng, Y.; Bi, X.; Xu, Y.; Liu, Y.; Li, J.; Du, G.; Lv, X.; Liu, L. Artificial intelligence technologies in bioprocess: Opportunities and challenges. Bioresour. Technol. 2023, 369, 128451.
Arinez, J.F.; Chang, Q.; Gao, R.X.; Xu, C.; Zhang, J. Artificial Intelligence in Advanced Manufacturing: Current Status and Future Outlook. J. Manuf. Sci. Eng. 2020, 142, 110804.
Lam, C.; van Velthoven, M.H.; Meinert, E. Developing a Blockchain-Based Supply Chain System for Advanced Therapies: Protocol for a Feasibility Study. JMIR Res. Protoc. 2020, 9, e17005.
Singh, R.; Dwivedi, A.D.; Srivastava, G. Internet of Things Based Blockchain for Temperature Monitoring and Counterfeit Pharmaceutical Prevention. Sensors 2020, 20, 3951.
Moosavi, J.; Naeni, L.M.; Fathollahi-Fard, A.M.; Fiore, U. Blockchain in supply chain management: A review, bibliometric, and network analysis. Environ. Sci. Pollut. Res. Int. 2021.
EMA. Multi-annual AI workplan 2023–2028 HMA-EMA Big Data Steering Group. 2023, 1–13. Available online: https://www.ema.europa.eu/en/news/artificial-intelligence-workplan-guide-use-ai-medicines-regulation (accessed on 1 February 2024).
EMA. Reflection paper on the use of artificial intelligence in lifecycle medicines. Eur. Med. Agency 2023, 31, 1–17.

© Text is available under the terms and conditions of the Creative Commons Attribution (CC BY) license; additional terms may apply. By using this site, you agree to the Terms and Conditions and Privacy Policy.

Upload a video for this entry

Information

Subjects: Biotechnology & Applied Microbiology

Contributors MDPI registered users' name will be linked to their SciProfiles pages. To register with us, please refer to https://encyclopedia.pub/register :

Cristobal Aguilar-Gallardo

Ana Bonora-Centelles

View Times: 80

Update Date: 23 Feb 2024

Table of Contents

Video Upload Options

Confirm