Machine Learning-Based Prediction of Mononuclear Cell Collection Efficiency for CAR-T Manufacturing

Abstract

Objective

To develop and internally validate an interpretable machine-learning model using routine precollection variables to predict mononuclear cell (MNC) collection efficiency (CE) during leukapheresis for CAR-T manufacturing.

Methods

This retrospective study included 206 consecutive patients who underwent leukapheresis between 2022 and 2025. Candidate variables were screened with Pearson correlation, Boruta, recursive feature elimination, and variance-threshold filters. Six supervised algorithms were trained with nested cross-validation, using an inner fivefold loop for tuning and an outer tenfold loop repeated three times for performance estimation. Bootstrap resampling (1000 iterations) generated confidence intervals, and SHAP analysis was used for interpretation.

Results

Extreme gradient boosting (XGBoost) showed the best continuous prediction performance, with an R² of 0.67 (95% CI: 0.64–0.70), Pearson r = 0.82, RMSE = 12.4%, and MAE = 9.8%. In the secondary triage analysis, observed high CE was defined as CE ≥ 60% (33/206, 16.0%). The predicted CE score discriminated high CE with an AUC of 0.84. A predicted-CE threshold of ≥ 50% yielded sensitivity of 84.8%, specificity of 97.1%, PPV of 84.8%, and NPV of 97.1%. The leading predictors were precollection lymphocyte count, lymphocyte/WBC ratio, age, body surface area, and monocyte count.

Conclusion

Routine precollection variables estimated MNC CE with good internal validity and may support individualized leukapheresis planning before CAR-T manufacturing.

Keywords

CAR-T therapy / Leukapheresis / Collection efficiency / Mononuclear cells / Machine learning / XGBoost

References

[1] June CH, O’Connor RS, Kawalekar OU, et al.. CAR T cell immunotherapy for human cancer. Science., 2018, 359(6382): 1361-1365

[2] Blüm P, Kayser S. Chimeric antigen receptor (CAR) T-cell therapy in hematologic malignancies: clinical implications and limitations. Cancers., 2024, 16(8): 1599

[3] Ayala Ceja M, Khericha M, Harris CM, et al.. CAR-T cell manufacturing: Major process parameters and next-generation strategies. J Exp Med., 2024, 221(2 ArticleID: e20230903

[4] Korell F, Laier S, Sauer S, et al.. Current challenges in providing good leukapheresis products for manufacturing of CAR-T cells for patients with relapsed/refractory NHL or ALL. Cells., 2020, 9(5): 1225

[5] Tuazon SA, Li A, Gooley T, et al.. Factors affecting lymphocyte collection efficiency for the manufacture of chimeric antigen receptor T cells in adults with B-cell malignancies. Transfusion., 2019, 59(5): 1773-1780

[6] Agliardi G, Dias J, Rampotas A, et al.. Accelerating and optimising CAR T-cell manufacture to deliver better patient products. Lancet Haematol., 2025, 12(1): e57-e67

[7] Qayed M, McGuirk JP, Myers GD, et al.. Leukapheresis guidance and best practices for optimal chimeric antigen receptor T-cell manufacturing. Cytotherapy., 2022, 24(9): 869-878

[8] Leone G, Baldini V, Bramanti S, et al.. Managing leukapheresis in adult and pediatric patients eligible for chimeric antigen receptor T-cell therapy: suggestions from an Italian Expert Panel. Blood Transfus., 2023, 21(6): 514-525

[9] O’Reilly MA, Malhi A, Cheok KPL, et al.. A novel predictive algorithm to personalize autologous T-cell harvest for chimeric antigen receptor T-cell manufacture. Cytotherapy., 2023, 25(3): 323-329

[10] Huang X, Gan GPL, Chan EHL, et al.. A proposed predictive mathematical model for efficient T-cell collection by leukapheresis for manufacturing chimeric antigen receptor T cells. Haematologica., 2023, 108(11): 3131-3134

[11] Li M, Yao D, Zeng X, et al.. Age related human T cell subset evolution and senescence. Immun Ageing., 2019, 16(1): 24

[12] Xiang J, Shi M, Fiala MA, et al.. Machine learning-based scoring models to predict hematopoietic stem cell mobilization in allogeneic donors. Blood Adv., 2022, 6(7): 1991-2000

[13] Lewis JE, Cooper LAD, Jaye DL, et al. Automated deep learning-based diagnosis and molecular characterization of acute myeloid leukemia using flow cytometry. bioRxiv. 2023:2023.09.18.558289.

[14] Asteris PG, Gandomi AH, Armaghani DJ, et al.. Pre-transplant and transplant parameters predict long-term survival after hematopoietic cell transplantation using machine learning. Transpl Immunol., 2025, 90 ArticleID: 102211

[15] Rudin C. Stop explaining black box machine learning models for high stakes decisions and use interpretable models instead. Nat Mach Intell., 2019, 1(5): 206-215

[16] Collins GS, Moons KGM, Dhiman P, et al.. TRIPOD+AI statement: updated guidance for reporting clinical prediction models that use regression or machine learning methods. BMJ., 2024, 385 ArticleID: e078378

[17] Kartal O, Laner-Plamberger S, Rohde E, et al.. Evaluating a new photopheresis system: a comparison with two established systems on cell yield and collection efficiency. Diagnostics (Basel)., 2024, 14(20): 2290

[18] Kartal O, Lindlbauer N, Laner-Plamberger S, et al.. Collection efficiency of mononuclear cells in offline extracorporeal photopheresis: can processing time be shortened?. Blood Transfus., 2024, 22(2): 150-156

[19] van Buuren S, Groothuis-Oudshoorn K. mice: multivariate imputation by chained equations in R. J Stat Soft., 2011, 45(3): 1-67

[20] Du Bois D, Du Bois EF. A formula to estimate the approximate surface area if height and weight be known. Nutrition, 1989, 5(5): 303-311

[21] Kursa MB, Rudnicki WR. Feature selection with the Boruta Package. J Stat Soft., 2010, 36(11): 1-13

[22] Guyon I, Weston J, Barnhill S, et al.. Gene selection for cancer classification using support vector machines. Mach Learn., 2002, 46(1): 389-422

[23] Chen T, Guestrin C. XGBoost: a scalable tree boosting system. Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. San Francisco California USA. ACM, 2016: 785–794.

[24] Varma S, Simon R. Bias in error estimation when using cross-validation for model selection. BMC Bioinformatics., 2006, 7: 91

[25] Moons KGM, Damen JAA, Kaul T, et al.. PROBAST+AI: an updated quality, risk of bias, and applicability assessment tool for prediction models using regression or artificial intelligence methods. BMJ., 2025, 388 ArticleID: e082505

[26] Lundberg SM, Lee SI. A unified approach to interpreting model predictions. In: Proceedings of the 31st International Conference on Neural Information Processing Systems. Red Hook, NY, USA: Curran Associates Inc. 2017. pp. 4765–4774.

[27] Cutillo CM, Sharma KR, Foschini L, et al.. Machine intelligence in healthcare-perspectives on trustworthiness, explainability, usability, and transparency. NPJ Digit Med., 2020, 3: 47

[28] Noaks E, Peticone C, Kotsopoulou E, et al.. Enriching leukapheresis improves T cell activation and transduction efficiency during CAR T processing. Mol Ther Meth Clin Dev., 2021, 20: 675-687

[29] Mehta PH, Fiorenza S, Koldej RM, et al.. T cell fitness and autologous CAR T cell therapy in haematologic malignancy. Front Immunol., 2021, 12 ArticleID: 780442

[30] Moog R. Apheresis techniques for collection of peripheral blood progenitor cells. Transfus Apher Sci., 2004, 31(3): 207-220

[31] Saavedra D, Añé-Kourí AL, Barzilai N, et al.. Aging and chronic inflammation: highlights from a multidisciplinary workshop. Immun Ageing., 2023, 20(1): 25

[32] Alswied A, Daniel D, Chen LN, et al.. CD34⁺ cell yield among healthy donors: Large-scale model development and validation. J Clin Apher., 2024, 39(3 ArticleID: e22135

[33] Hutt D, Bielorai B, Baturov B, et al.. Feasibility of leukapheresis for CAR T-cell production in heavily pre-treated pediatric patients. Transfus Apher Sci., 2020, 59(4 ArticleID: 102769

[34] Małachowski R, Grzybowska-Izydorczyk O, Szmigielska-Kapłon A, et al.. Comparison of prediction models for two different peripheral stem cell collection protocols in autologous patients How to avoid errors in calculating total blood volume to process?. Acta Haematol Pol., 2021, 52(1): 38-47

[35] Castillo-Aleman YM, Villegas-Valverde CA, Bencomo-Hernandez AA, et al.. The day two advantage: a retrospective paired analysis of consecutive HPC(A) collections. J Clin Apher., 2025, 40(4 ArticleID: e70046

[36] Vasey B, Nagendran M, Campbell B, et al.. Reporting guideline for the early-stage clinical evaluation of decision support systems driven by artificial intelligence: DECIDE-AI. Nat Med., 2022, 28(5): 924-933

[37] Chen GM, Chen C, Das RK, et al.. Integrative bulk and single-cell profiling of premanufacture T-cell populations reveals factors mediating long-term persistence of CAR T-cell therapy. Cancer Discov., 2021, 11(9): 2186-2199

Funding

National Natural Science Foundation of China(82425003)

Wuhan Science and Technology Achievement Transformation Project(2024030803010193)

RIGHTS & PERMISSIONS

The Author(s)

Just Accepted

This article has successfully passed peer review and final editorial review, and will soon enter typesetting, proofreading and other publishing processes. The currently displayed version is the accepted final manuscript. The officially published version will be updated with format, DOI and citation information upon launch. We recommend that you pay attention to subsequent journal notifications and preferentially cite the officially published version. Thank you for your support and cooperation.