AI-Driven Quality by Design (QbD): Advancing Predictive, Data-Driven Pharmaceutical Formulation Development Using Artificial Neural Networks

Paritosh Singh, Senior Scientist – Pharmaceutical R&D

Paritosh Singh is a pharmaceutical R&D leader with extensive expertise in advanced drug delivery and AI-driven formulation development, leveraging Quality by Design (QbD), statistical modeling, and machine learning to develop robust and scalable pharmaceutical products. He is recognized for original and impactful contributions to the application of artificial intelligence in formulation science, with influence across multiple development programs. He has held critical scientific roles within global pharmaceutical organizations, including Jubilant, Johnson & Johnson, and Perrigo. He serves as a reviewer for leading scientific platforms, including AAPS and the Journal of Controlled Release, and as an international award judge, including AOAC scholarship panels, reflecting peer recognition of his expertise and contributions to the field.

Introduction: Moving Beyond Conventional QbD

Quality by Design (QbD) has become a cornerstone of pharmaceutical development by enabling a structured, science- and risk-based approach to formulation and process design. Regulatory frameworks such as ICH Q8 and ICH Q9 emphasize linking critical quality attributes (CQAs) to formulation and process variables to ensure consistent product performance.

However, as formulations become more complex, traditional QbD approaches—largely dependent on Design of Experiments (DoE)-face practical limitations. These include challenges in capturing non-linear interactions, increasing experimental burden, and limited ability to predict outcomes beyond defined experimental spaces.

Artificial intelligence (AI) tools, particularly artificial neural networks (ANN), offer a complementary approach by enabling predictive, data-driven modeling. Integrating ANN within QbD frameworks provides an opportunity to move from empirical development toward a more efficient and predictive paradigm.

“AI-enabled QbD is transforming pharmaceutical development from experimental optimization to predictive, data-driven design.”

Hybrid QbD–AI Framework

A hybrid QbD–AI approach integrates structured experimentation with advanced modeling to improve formulation understanding and optimization. In this framework, initial DoE studies generate experimental data that are used to train ANN models. These models extend the design space, identify complex relationships, and support virtual optimization of formulation variables. Experimental confirmation then validates predictions, enabling robust and scalable formulation design.

“The integration of artificial neural networks within QbD enables the identification of complex, non-linear relationships that traditional models fail to capture.”

Hybrid QbD–AI workflow (practical view)

1. Define QTPP and CQAs
2. Run initial DoE screening to generate structured data
3. Train ANN on curated formulation/process datasets
4. Predict and refine the design space
5. Confirm with targeted validation runs
6. Establish control strategy and monitoring plan (PAT-enabled)

Figure 1. Hybrid QbD–AI framework for formulation development. Integration of traditional QbD elements with AI-driven modeling enables predictive design space expansion, optimized formulation selection, and robust control strategy development.

A comparative summary of traditional and AI-enabled QbD approaches is provided in Table 1

Table 1. Comparison of Traditional QbD vs AI-Enabled Hybrid QbD Approaches

“The integration of AI into QbD enables a shift from experimental optimization to predictive and knowledge-driven formulation development.”

Application Example

In a representative application aligned with previously published work (Singh et al., 2026), artificial neural networks (ANN) were integrated within a QbD-guided formulation development strategy to optimize a multi-component modified release tablet system. Initial DoE studies were conducted to evaluate key formulation variables and their influence on CQAs. The resulting dataset was used to train ANN models capable of capturing non-linear relationships not fully described by conventional statistical approaches.

In the published study, the ANN model showed excellent predictive performance, with generalized R² values of 0.9989 (training), 0.9999 (validation), and 0.9988 (test), supporting reliable optimization of polymer ratios and modified-release performance.

These findings highlight the practical value of integrating AI within QbD frameworks to enhance formulation robustness, scalability, and development efficiency.

“Hybrid QbD–AI approaches significantly reduce experimental burden while improving formulation robustness and scalability.”

“AI models enable expansion of the design space beyond experimental boundaries, enhancing predictive confidence.”

Figure 2. Comparison of Traditional QbD and AI-enabled Hybrid QbD approaches.

Predictive accuracy is one win.

The real test is whether that robustness holds up against the hidden cost of quality failures downstream.

→ Read: The Hidden Tax on Pharma: Mastering the Cost of Quality Before It Masters You

Impact on Manufacturing and Scale-Up

AI-guided models improve scale-up predictability by quantitatively linking critical material attributes (CMAs) and critical process parameters (CPPs) to CQAs across scales. By identifying process sensitivities and non-linear interactions, these models enable more robust design space definition and support control strategy optimization. Integration with process analytical technology (PAT) further allows data-driven monitoring and reduces variability during scale-up, technology transfer, and commercial manufacturing.

AI models should be managed under lifecycle governance (e.g., version control, retraining criteria, and audit trails) to support sustainable deployment within a pharmaceutical quality system.

“AI-guided formulation development improves scale-up predictability and reduces risk during commercialization.”

Practical adoption checklist (industry-ready)

1. Use structured DoE to generate high-quality training data and avoid bias in factor coverage.
2. Define CQAs and acceptance criteria before model training (QTPP-driven).
3. Validate ANN predictions using an independent confirmation set before expanding the design space.
4. Translate model outputs into a control strategy and monitoring plan (PAT where appropriate).
5. Maintain model governance through documentation, change control, and periodic performance review.

Analytical procedures supporting PAT, release, and stability testing should follow science- and risk-based development and validation principles consistent with current ICH guidance (e.g., ICH Q14 and ICH Q2(R2)).

AI models need lifecycle governance to survive scale-up. The same risk-based thinking is rewriting how MES systems get validated.

→ Read: GAMP 5 Second Edition for MES: A Practical Validation Guide

Conclusion

Hybrid QbD–AI approaches represent a significant advancement in pharmaceutical development, enabling more predictive, efficient, and robust formulation design. By integrating AI with established QbD principles, these approaches improve understanding of complex formulation–process relationships, enhance design space definition, and support robust control strategies across development and manufacturing. As pharmaceutical systems become increasingly complex, AI-enabled QbD frameworks offer a practical pathway toward more informed, efficient, and reliable product development.

“The future of pharmaceutical R&D lies in converging QbD principles with AI-driven predictive modeling.”

References

1. Singh, P.K., Kumar, V., Parihar, A., et al. AI-Powered Predictive Modeling to Optimize Pharmaceutical Formulation and Precise Drug Delivery in Modified Release Tablets. Journal of Pharmaceutical Innovation. 2026; 21:52. https://doi.org/10.1007/s12247-025-10283-2
2. International Council for Harmonisation (ICH). ICH Q8(R2): Pharmaceutical Development. Final version (Step 4), August 2009. https://database.ich.org/sites/default/files/Q8_R2_Guideline.pdf
3. International Council for Harmonisation (ICH). ICH Q9(R1): Quality Risk Management. Final version (Step 4), adopted 18 January 2023. https://database.ich.org/sites/default/files/ICH_Q9%28R1%29_Guideline_Step4_2022_1219.pdf
4. International Council for Harmonisation (ICH). ICH Q10: Pharmaceutical Quality System. Final version (Step 4), 4 June 2008. https://database.ich.org/sites/default/files/Q10%20Guideline.pdf
5. International Council for Harmonisation (ICH). ICH Q14: Analytical Procedure Development. Final version, adopted 1 November 2023. https://database.ich.org/sites/default/files/ICH_Q14_Guideline_2023_1116.pdf
6. International Council for Harmonisation (ICH). ICH Q2(R2): Validation of Analytical Procedures. Final version, adopted 1 November 2023. https://database.ich.org/sites/default/files/ICH_Q2%28R2%29_Guideline_2023_1130.pdf