2026-04-23 ノースカロライナ州立大学（NC State）
Photo credit: National Cancer Institute.
＜関連情報＞
- https://news.ncsu.edu/2026/04/ai-counting-viruses/
- https://ieeexplore.ieee.org/abstract/document/11480695
機械学習強化型免疫センサーを用いたラベルフリーのウイルス力価定量 Label-Free Quantification of Virus Titer using Machine Learning-Enhanced Immunosensors
Rajendra P. Shukla; He Sun; Junhyeong Wang; Jack S. Twiddy; Shriarjun Shastry; Mahshid Hosseini,…
IEEE Sensors Journal Published:13 April 2026
DOI:https://doi.org/10.1109/JSEN.2026.3681515
Abstract:
Process analytical technology (PAT) for gene therapy manufactur-ing requires real-time monitoring of critical quality attributes (CQAs), yet current methods remain labor-intensive and incompatible with in-line deployment. We de-veloped a label-free electrochemical impedance spectroscopy (EIS) immunosen-sor integrated with machine learning (ML) algorithms that directly extract features from raw impedance spectra, eliminating the need for equivalent circuit modeling and enabling real-time classification and quantification. We demonstrated this platform via simultaneous classification of buffer pH conditions and quantification of adeno-associated virus titer, using legacy serotype 2 (AAV2) for proof-of-con-cept. Gold microelectrodes functionalized with anti-AAV2 antibodies were tested across four pH conditions (4, 6, 7.4, 9) and AAV2 titers spanning 108 to 1012 cap-sids·mL-1, capturing highly overlapping and complex electrochemical signatures. Task-specific feature selection identified optimal descriptors for classification and regression. Logistic regression and k-nearest neighbors (KNN) achieved high pH classification accuracies (training: 0.99; testing: 0.94 and 0.95, respectively). For AAV2 quantification, ensemble re-gression models XGBoost, Random Forest, and Gradient Boosting outperformed linear models, yielding training set R² values of 0.90, 0.85, and 0.81 and test set R² values of 0.78, 0.72, and 0.68, respectively. The non-Faradaic impedi-metric sensing platform, coupled with physics-informed ML feature engineering, provides a robust platform for auto-mated, label-free monitoring of viral vector CQAs in biomanufacturing environments.
