Comparison of Machine Learning-Enhanced Dynamic Hybrid Models for a Nanobody Scorpion Antivenom Production with Escherichia Coli
Document Type
Conference Paper
Year
2025
Authors
Irene Martínez-Menéndez, Juan C. Acosta-Pavas, David Camilo Corrales, Susana María Alonso Villela, Balkiss Bouhaouala-Zahar, Georgios K. Georgakilas, Konstantinos Mexis, Stefanos Xenios, Theodore Dalamagas, Antonis Kokosis, Michael O’donohue, Luc Fillaudeau, Nadia Boukhelifa, Alberto Tonda & César A. Aceves-Lara
Source
Lecture Notes in Networks and Systems , 2025, 1259, pp. 307–316
Abstract
This study compares several hybrid dynamic models built with offline data to optimize nanobody production in Escherichia coli. These models combine the insights of mechanistic knowledge and Machine Learning algorithms (ML). Four ML algorithms were tested: Support Vector Regressor (SVR), Random Forest (RF), Decision Tree (DT) and K-Nearest Neighbors (KNN). The efficacy of each model was analyzed with three metrics: mean absolute error (MAE), root mean squared error (RMSE), normalized root mean squared error (NRMSE). Results demonstrate that the radial basis function (RBF) SVR model performed best simulations, with a NRMSE between 0.063 and 0.665. The insights gained are pivotal for advancing computational modeling techniques in biomanufacturing, with a specific emphasis on the production of recombinant therapeutic proteins, such as nanobodies, using bacterial hosts.
