Roadmap: control, PAT, and modeling additions from the FDA freeze-drying trends paper (Rhieu et al. 2025)

[bernalde]

Source

Rhieu SY, Korang-Yeboah M, Anderson DD, Arigo J, O'Connor T, Shah R (2025). Recent trends in pharmaceutical freeze-drying and control strategies observed in human drug applications and manufacturing inspections. AAPS Open 11:27. https://doi.org/10.1186/s41120-025-00132-4

This is an FDA/CDER Office of Pharmaceutical Quality perspective paper. It analyzes the control strategies actually filed in 162 regulatory submissions (24 NDAs, 118 ANDAs, 20 BLAs; 2020–2023), reviews 483 inspection observations from 201 establishment inspections (2015–2019), and surveys the emerging technologies FDA is actively encouraging: controlled ice nucleation (CIN), PAT, process modeling, continuous freeze-drying, and microwave-assisted freeze-drying (MAFD). It cites LyoPRONTO directly (Shivkumar et al. 2019b) among the process-modeling tools whose industrial adoption is increasing.

Findings that frame the opportunities below:

• The dominant filed endpoint methods are product temperature vs. shelf temperature (62.5% of NDAs) and Pirani vs. capacitance manometer comparative pressure (~65% of BLAs). Trial-and-error cycle development is still filed in 17–37% of applications. SMART™/MTM appears in only ~2.5–3% of ANDAs.
• PAT in commercial manufacturing is rare (4.2% of NDAs, 0.9% of ANDAs) despite mature tools (TDLAS, NIR, wireless sensors).
• Scale-up factors of 1.5–10× are filed; lack of scale-up justification leads to deficiencies.
• Top 483 themes include: critical process limits not defined, absence of product-specific development data supporting the cycle, and inadequate deviation investigations — all directly answerable with model-based artifacts.
• Models used as part of a control strategy carry higher decision consequence than development-only models, pulling in the model-credibility frameworks (FDA 2023 draft guidance, ASME V&V 40, EMA 2024 QIG).

Already covered (no action)

• Quasi-steady primary-drying model (Kv/Rp), freezing calculator, design-space generator, SciPy + Pyomo/IPOPT optimizers
• Equipment capability / choked-flow limit (eccurt, based on Shivkumar et al. 2019a)
• Kv/Rp fitting from experimental data (fitting, calc_unknownRp)
• End-of-primary-drying detection from measured Pirani traces (cycle_time.identify_pd_end)
• RF/microwave drying model (rf)

Proposed additions

1. Virtual PAT / instrument soft sensors

LyoPRONTO simulates the physics but not the instrument signals, so predictions can't be compared 1:1 against the endpoint evidence firms actually file (paper Fig. 2).

• Virtual Pirani vs. capacitance manometer: predict the Pirani reading from chamber gas composition (thermal-conductivity gauge reads ~1.6× high in water vapor), and report predicted onset/midpoint/offset endpoint times. Complements the existing measured-trace identify_pd_end. (Nail et al. 2017; Patel & Pikal 2009)
• Pressure-rise test / MTM / SMART™ emulator: simulate the valve-closed pressure-rise transient from current model state; inverse-estimate sublimation-front temperature and Rp; optionally reproduce SMART-style automated cycle design. (Tang, Nail & Pikal 2005; Gieseler et al. 2007a)
• Virtual TDLAS outputs: batch-average sublimation mass flow, batch-average product temperature, Rp estimate, endpoint, and margin to choked flow vs. the eccurt curve. (Gieseler et al. 2007b; Yu et al. 2024)
• Energy-balance sublimation-flow estimator from shelf-fluid inlet/outlet ΔT or chamber–condenser Δp — a "no new hardware" soft sensor recently demonstrated. (Authelin et al. 2024)

2. Closed-loop control and state estimation

The paper points at ICH Q13-style closed-loop drying control as the exemplary direction (Leys et al. 2023), and closed-loop control of MAFD was just demonstrated by the LyoPRONTO group itself (Alexeenko et al. 2025). The existing pyomo_models multi-period trajectory optimization is most of an MPC already.

• Receding-horizon MPC: re-solve the Pyomo trajectory problem from an estimated current state under Tp < Tc and equipment-capability constraints, driven by simulated (or recorded) Pirani/CM or TDLAS feedback. (ICH Q13 2023; Leys et al. 2023; Alexeenko et al. 2025)
• State/parameter estimation layer: moving-horizon or recursive estimation of Rp(l), Kv, and front position from noisy measurements — including the sparse-sensor GMP case (gravimetric + pressure only). (Geremia et al. 2022)
• Robust / uncertainty-aware cycle optimization ("fast and robust" formulations). (Vanbillemont et al. 2023; Mockus et al. 2011)

3. Uncertainty quantification and batch heterogeneity

• Monte Carlo over Kv, Rp, and nucleation-temperature distributions → product-temperature percentile bands, drying-time distributions, worst-case vial tracking. (Pikal et al. 2018)
• Probabilistic design space: probability-of-success contours instead of a single deterministic boundary. (Pikal et al. 2018; Mockus et al. 2011)
• Edge- vs. center-vial classes with positional Kv distributions (extend vials). (Nail et al. 2017)
• Bayesian parameter estimation with credible intervals in fitting. (Mockus et al. 2011)

4. Freezing → primary-drying coupling and controlled ice nucleation

Freezing outputs currently don't inform primary-drying inputs. The paper treats nucleation temperature as a controlling variable for Rp and drying time, and CIN as the flagship emerging technology (approved in one NDA and one BLA since 2020; three ETP evaluations).

• Nucleation temperature → ice-crystal/pore size → Rp correlation, so the freezing calculator feeds the primary-drying model. (Searles et al. 2001; Assegehegn et al. 2019; Juckers et al. 2023)
• CIN scenario modeling: uniform nucleation at a set temperature; post-nucleation shelf ramp-rate and hold-time effects; predicted drying-time savings vs. stochastic nucleation; note the documented inverse cases where higher nucleation temperature hurt efficiency/quality. (Luoma et al. 2019; Strongrich et al. 2021; Korang-Yeboah et al. 2023; Fang et al. 2020; Ganguly 2024)
• Annealing step model (same physics hooks as CIN/freezing coupling).
• Optional: ingest structural characterization data (e.g., OCT freeze-drying microscopy, pore-structure measurements) as Rp priors. (Korang-Yeboah et al. 2018; Ma et al. 2025)

5. Secondary-drying module

Biggest single model gap: docs/technical/physics-reference.md names secondary drying as phase 3, but nothing models it. The paper's submission analysis covers endpoint determination for both drying phases, and residual moisture is the CQA behind closed-loop demonstrations (per-vial in-line NIR moisture in Leys et al. 2023).

• Desorption-kinetics model → residual moisture vs. time, endpoint prediction, ramp optimization; enables full-cycle simulation and connects to dew-point/residual-gas endpoint indication.

6. Scale-up, tech transfer, and deviation analysis

Filed scale-up factors are 1.5–10×, and the top lyo-cycle 483 themes (critical process limits not defined; inadequate deviation investigations) are directly answerable with the models already in this repo.

• "GMP factor" support: empirical lab→plant Rp adjustment plus lab-vs-plant scenario diffing. (Tchessalov et al. 2021; Zhu et al. 2018)
• Deviation replay: feed a recorded Tsh/Pch excursion through the calculator and report impact on product temperature, Tp−Tc margin, sublimation flux, and endpoint shift — a ready-made deviation-investigation artifact.
• Extend equipment characterization beyond the choke-flow curve: minimum controllable pressure, condenser/refrigeration capacity, standardized equipment-performance-qualification inputs. (Ganguly et al. 2023; Shivkumar et al. 2019a)
• Exploratory: chamber-level gradients / digital-twin coupling (3-D equipment CFD + 1-D vial model). (Zadravec et al. 2025)

7. Model credibility / V&V workflow for regulatory use

The paper leans on risk-based model-credibility frameworks and notes that control-strategy models carry higher decision consequence than development-only models.

• Context-of-use + credibility documentation template, validation-dataset registry with error metrics, and an auto-generated validation report for a given model configuration. (FDA 2023; ASME V&V 40 2018; EMA 2024; O'Connor et al. 2024)

8. MAFD/RF extensions (build on rf)

• Closed-loop MAFD control demo (pairs with §2). (Alexeenko et al. 2025)
• Heating-heterogeneity treatment: statistical-electromagnetics-informed variability inputs for field uniformity. (Abdelraheem et al. 2022; Bhambhani et al. 2021; Gitter et al. 2018)
• Formulation dielectric-property data hooks — CDER is funding exactly this pairing (MAFD heat/mass model + formulation dielectric database).

9. Continuous freeze-drying (exploratory, long-term)

No filed applications yet per the paper, but FDA flags continuous manufacturing as the direction with ICH Q13 in force.

• Spin freeze-drying vial model: thin annular product layer with noncontact IR heating. (De Meyer et al. 2015; Van Bockstal et al. 2017; Leys et al. 2023)
• Suspended-vial continuous concept and spray freeze-drying kinetics. (Capozzi et al. 2019; Pisano et al. 2019; Borges Sebastião et al. 2021; Clénet et al. 2019)

Suggested sequencing

1. Near-term, pure-Python, high leverage: virtual Pirani/CM + pressure-rise emulator (§1), deviation replay (§6), Monte Carlo UQ (§3) — all reuse existing calculators.
2. Fork-aligned: MPC/MHE on pyomo_models (§2) — dovetails with the POUNCE/NLP comparison work (Add solver-swappable NLP instances with an IPOPT baseline, motivating a POUNCE migration #140, Surface convergence quality in the pseudosteady limit study, and settle the cross-solver mapping #146).
3. Model gaps: freezing→Rp coupling (§4), secondary drying (§5).
4. Long-term: V&V tooling (§7), MAFD extensions (§8), continuous modes (§9).

References checked (from the paper's bibliography)

Expand full list with DOIs

• Abdelraheem A, Tukra R, Kazarin P, Sinanis MD, Topp EM, Alexeenko A, et al. (2022) Statistical electromagnetics for industrial pharmaceutical lyophilization. PNAS Nexus. https://doi.org/10.1093/pnasnexus/pgac052
• Alexeenko A, Darwish A, Strongrich D, Kazarin P, Patil C, Tower W, et al. (2025) Randomized-field microwave-assisted pharmaceutical lyophilization with closed-loop control. Sci Rep 15:10536. https://doi.org/10.1038/s41598-025-91642-4
• ASME (2018) V&V 40 — Assessing Credibility of Computational Modeling through Verification and Validation: Application to Medical Devices.
• Assegehegn G, Brito-de la Fuente E, Franco JM, Gallegos C (2019) The importance of understanding the freezing step and its impact on freeze-drying process performance. J Pharm Sci 108(4):1378–1395. https://doi.org/10.1016/j.xphs.2018.11.039
• Authelin J-R, et al. (2024) A simple and cost-effective technique to monitor the sublimation flow during primary drying using shelf inlet/outlet temperature difference or chamber-to-condenser pressure drop. J Pharm Sci 113(7):1898–1906. https://doi.org/10.1016/j.xphs.2024.02.015
• Bhambhani A, et al. (2021) Evaluation of microwave vacuum drying as an alternative to freeze-drying of biologics and vaccines: the power of simple modeling. AAPS PharmSciTech 22(1):52. https://doi.org/10.1208/s12249-020-01912-9
• Borges Sebastião I, Bhatnagar B, Tchessalov S (2021) A kinetic model for spray-freezing of pharmaceuticals. J Pharm Sci 110(5):2047–2062. https://doi.org/10.1016/j.xphs.2020.11.032
• Capozzi LC, Trout BL, Pisano R (2019) From batch to continuous: freeze-drying of suspended vials for pharmaceuticals in unit-doses. Ind Eng Chem Res 58(4):1635–1649. https://doi.org/10.1021/acs.iecr.8b02886
• Clénet D, et al. (2019) A spray freeze-dried micropellet based formulation proof-of-concept for a yellow fever vaccine candidate. Eur J Pharm Biopharm 142:334–343. https://doi.org/10.1016/j.ejpb.2019.07.008
• De Meyer L, et al. (2015) Evaluation of spin freezing versus conventional freezing as part of a continuous pharmaceutical freeze-drying concept for unit doses. Int J Pharm 496(1):75–85. https://doi.org/10.1016/j.ijpharm.2015.05.025
• EMA (2024) Preliminary QIG Considerations regarding Pharmaceutical Process Models.
• Fang R, Bogner RH, Nail SL, Pikal MJ (2020) Stability of freeze-dried protein formulations: contributions of ice nucleation temperature and residence time in the freeze-concentrate. J Pharm Sci 109(6):1896–1904. https://doi.org/10.1016/j.xphs.2020.02.014
• FDA (2023) Assessing the Credibility of Computational Modeling and Simulation in Medical Device Submissions.
• FDA (2025) Considerations for the Use of Artificial Intelligence To Support Regulatory Decision-Making for Drug and Biological Products.
• Ganguly A, et al. (2023) Recommended best practices in freeze dryer equipment performance qualification: 2022. AAPS PharmSciTech 24(1):45. https://doi.org/10.1208/s12249-023-02506-x
• Ganguly A (2024) Learnings from implementation of production scale controlled nucleation technology. ISL-FD, Turin.
• Geremia M, Bano G, Tomba E, Barolo M, Bezzo F (2022) Practical use of primary drying models in an industrial environment with limited equipment sensors. Int J Pharm 619:121699. https://doi.org/10.1016/j.ijpharm.2022.121699
• Gieseler H, Kramer T, Pikal MJ (2007a) Use of manometric temperature measurement (MTM) and SMART™ freeze dryer technology for development of an optimized freeze-drying cycle. J Pharm Sci 96(12):3402–3418. https://doi.org/10.1002/jps.20982
• Gieseler H, et al. (2007b) Evaluation of tunable diode laser absorption spectroscopy for in-process water vapor mass flux measurements during freeze drying. J Pharm Sci 96(7):1776–1793. https://doi.org/10.1002/jps.20827
• Gitter JH, Geidobler R, Presser I, Winter G (2018) Significant drying time reduction using microwave-assisted freeze-drying for a monoclonal antibody. J Pharm Sci 107(10):2538–2543. https://doi.org/10.1016/j.xphs.2018.05.023
• ICH (2023) Q13 Continuous Manufacturing of Drug Substances and Drug Products.
• Juckers A, Knerr P, Harms F, Strube J (2023) Effect of the freezing step on primary drying experiments and simulation of lyophilization processes. Processes 11(5):1404. https://doi.org/10.3390/pr11051404
• Korang-Yeboah M, et al. (2018) Application of optical coherence tomography freeze-drying microscopy for designing lyophilization process and its impact on process efficiency and product quality. AAPS PharmSciTech 19(1):448–459. https://doi.org/10.1208/s12249-017-0848-4
• Korang-Yeboah M, et al. (2023) Root cause analysis of an inverse relationship between the ice nucleation temperature, process efficiency and quality of a lyophilized product. J Pharm Sci 112(12):3035–3044. https://doi.org/10.1016/j.xphs.2023.08.019
• Leys L, Nuytten G, Van Bockstal P-J, De Beer T (2023) Evaluation of a PAT-based in-line control system for a continuous spin freeze-drying process. Int J Pharm 641:123062. https://doi.org/10.1016/j.ijpharm.2023.123062
• Luoma J, Magill G, Kumar L, Yusoff Z (2019) Controlled ice nucleation using ControLyo® pressurization-depressurization method. In: Lyophilization of Pharmaceuticals and Biologicals. Springer, pp 57–77. https://doi.org/10.1007/978-1-4939-8928-7_3
• Ma L, et al. (2025) Revealing the pore structure variance of lyophilized cakes from microcollapse, controlled ice nucleation, and protein concentration changes. J Pharm Sci 114(6):103775. https://doi.org/10.1016/j.xphs.2025.103775
• Mockus L, LeBlond D, Basu PK, Shah RB, Khan MA (2011) A QbD case study: Bayesian prediction of lyophilization cycle parameters. AAPS PharmSciTech 12(1):442–448. https://doi.org/10.1208/s12249-011-9598-x
• Nail S, et al. (2017) Recommended best practices for process monitoring instrumentation in pharmaceutical freeze drying—2017. AAPS PharmSciTech 18(7):2379–2393. https://doi.org/10.1208/s12249-017-0733-1
• O'Connor TF, et al. (2024) An examination of process models and model risk frameworks for pharmaceutical manufacturing. Int J Pharm X 8:100274. https://doi.org/10.1016/j.ijpx.2024.100274
• Patel SM, Pikal MJ (2009) Process analytical technologies (PAT) in freeze-drying of parenteral products. Pharm Dev Technol 14(6):567–587. https://doi.org/10.3109/10837450903295116
• Patel SM, et al. (2017) Lyophilized drug product cake appearance: what is acceptable? J Pharm Sci 106(7):1706–1721. https://doi.org/10.1016/j.xphs.2017.03.014
• Pikal MJ, et al. (2018) Impact of natural variations in freeze-drying parameters on product temperature history: application of quasi steady-state heat and mass transfer and simple statistics. AAPS PharmSciTech 19(7):2828–2842. https://doi.org/10.1208/s12249-018-1155-4
• Pisano R, Arsiccio A, Capozzi LC, Trout BL (2019) Achieving continuous manufacturing in lyophilization: technologies and approaches. Eur J Pharm Biopharm 142:265–279. https://doi.org/10.1016/j.ejpb.2019.06.027
• Pisano R (2019) Alternative methods of controlling nucleation in freeze drying. In: Lyophilization of Pharmaceuticals and Biologicals. Springer, pp 79–111. https://doi.org/10.1007/978-1-4939-8928-7_4
• Schneid S, Gieseler H (2008) Evaluation of a new wireless temperature remote interrogation system (TEMPRIS) to measure product temperature during freeze drying. AAPS PharmSciTech 9(3):729–739. https://doi.org/10.1208/s12249-008-9099-8
• Searles JA, Carpenter JF, Randolph TW (2001) The ice nucleation temperature determines the primary drying rate of lyophilization for samples frozen on a temperature-controlled shelf. J Pharm Sci 90(7):860–871. https://doi.org/10.1002/jps.1039
• Shivkumar G, et al. (2019a) Freeze-dryer equipment capability limit: comparison of computational modeling with experiments at laboratory scale. J Pharm Sci 108(9):2972–2981. https://doi.org/10.1016/j.xphs.2019.04.016
• Shivkumar G, Kazarin PS, Strongrich AD, Alexeenko AA (2019b) LyoPRONTO: an open-source lyophilization process optimization tool. AAPS PharmSciTech 20(8):328. https://doi.org/10.1208/s12249-019-1532-7
• Strongrich A, Lim FJ, Kumar L, Alexeenko A (2021) Rapid depressurization based controlled ice nucleation in pharmaceutical freeze-drying: the roles of the ballast gas and the vial. J Pharm Sci 110(11):3639–3647. https://doi.org/10.1016/j.xphs.2021.07.011
• Tang X, Nail SL, Pikal MJ (2005) Freeze-drying process design by manometric temperature measurement: design of a smart freeze-dryer. Pharm Res 22(4):685–700. https://doi.org/10.1007/s11095-005-2501-2
• Tchessalov S, et al. (2021) Application of first principles primary drying model to lyophilization process design and transfer: case studies from the industry. J Pharm Sci 110(2):968–981. https://doi.org/10.1016/j.xphs.2020.11.013
• Tsinontides SC, et al. (2004) Freeze drying—principles and practice for successful scale-up to manufacturing. Int J Pharm 280(1):1–16. https://doi.org/10.1016/j.ijpharm.2004.04.018
• Van Bockstal P-J, De Meyer L, Corver J, Vervaet C, De Beer T (2017) Noncontact infrared-mediated heat transfer during continuous freeze-drying of unit doses. J Pharm Sci 106(1):71–82. https://doi.org/10.1016/j.xphs.2016.05.003
• Vanbillemont B, et al. (2023) A model-based optimization strategy to achieve fast and robust freeze-drying cycles. Int J Pharm X 5:100180. https://doi.org/10.1016/j.ijpx.2023.100180
• Yu T, et al. (2024) Development of a single vial mass flow rate monitor to assess pharmaceutical freeze drying heterogeneity. AAPS PharmSciTech 25(8):245. https://doi.org/10.1208/s12249-024-02961-0
• Zadravec M, et al. (2025) Towards a digital twin of primary drying in lyophilization using coupled 3-D equipment CFD and 1-D vial-scale simulations. Eur J Pharm Biopharm 208:114662. https://doi.org/10.1016/j.ejpb.2025.114662
• Zhu T, et al. (2018) Predictive models of lyophilization process for development, scale-up/tech transfer and manufacturing. Eur J Pharm Biopharm 128:363–378. https://doi.org/10.1016/j.ejpb.2018.05.005

[bernalde]

@parkyr @Ickaser — could you both go through the analysis above and leave your comments before we turn this into a plan of action?

Specifically useful from each of you:

• Which sections are worth prioritizing (and which you'd deprioritize or drop), given effort vs. impact
• Anything the paper's framing misses or gets wrong from your experience with the codebase and the lyo physics
• Additions to the list — related literature or capabilities you think belong on this roadmap

Yirang: §2 (closed-loop MPC/estimation on pyomo_models) and §3 (UQ/probabilistic design space) are closest to the POUNCE/NLP line of work. Isaac: your read on §1 (virtual PAT), §4 (freezing→Rp coupling), and §5 (secondary drying) would be especially valuable given your work on the physics core and typed API — and whether any of it should land upstream in LyoHUB/LyoPRONTO rather than (or in addition to) this fork.

Once your comments are in, we'll sequence this into concrete issues.

[Ickaser]

I want to point out the following repos, which factor into my current mental roadmap:

• My pure-Julia complement to Python LyoPRONTO, with some overlapping functionality and some further additions: https://github.com/LyoHUB/LyoPronto.jl
• The current main consumer of LyoPRONTO: https://github.com/LyoHUB/LyoProntoServer (currently a closed repo but I gave you both read access; I plan to open source it at some point.)

LyoProntoServer is a multi-service set of composed Docker containers, which separately run a frontend, a backend running Python LyoPRONTO, and a backend running Julia LyoPronto.jl. It's currently running on Purdue's Geddes cluster with a public frontend available at https://lyopronto2.geddes.rcac.purdue.edu/ The old frontend is also running at https://lyopronto.geddes.rcac.purdue.edu/ but it's harder for me to maintain, since all I have is the minified JavaScript (not a documented source file).

In theory LyoPRONTO is installable and usable as a Python package, but to my knowledge relatively few users are doing so at this point, so the bulk of my efforts have gone into improvements that contribute to the web interface's functionality and usability (or to my own research progress, which is largely built in Julia).

Also, any industrial adoption is going to be achieved only after significant education of current professionals. I am collaborating

With respect to these specific ideas:

• §1 is a nice idea but I think simulating experimental measurements directly is not necessarily feasible at the level of model complexity in LyoPRONTO; for example, Pirani-CM convergence is widely understood in the field as a sufficient condition for end of primary drying, but experimental and model results differ about where exactly the "necessary" condition falls. I would have to think harder about whether the same applies to delta-T or delta-P methods (which have a new update here.
• §4 is a multi-year research project unto itself, which I have made some progress towards in a manuscript I need to submit; once that manuscript is submitted, my research will be available as an extension to LyoPronto.jl.
• §5 is on my shortlist of things I would like to implement, but there is not a literature-standard model for secondary drying in the same way that there is for primary drying, for a variety of reasons. If I get to this, it will likely be exposed to the web interface through LyoPronto.jl, not the Python backend.

I can provide other thoughts later if helpful.