Page 14 - IJAMD-1-1

International Journal of AI
for Material and Design ML in 3D bioprinting of cultivated meat

compositions and concentrations, the measurement of the printing parameters for each bioink. A recent study
rheological properties (specifically viscosity and storage utilized a support vector machine (SVM) model to generate
modulus), the actual printing of scaffolds (involving a process map aimed at aiding the selection of optimal
parameters such as printing pressure, nozzle diameter, printing parameters, ensuring a high likelihood (>75%)
nozzle velocity, and nozzle length), and the capture of of producing high-quality prints. The study focused on
45
optical images using a digital microscope with a resolution three crucial parameters: (i) Biomaterial concentration,
of 2592 × 1944 pixels (Figure 3). The relevant data about (ii) nozzle temperature, and (iii) path height. Utilizing a
material concentration, printing pressure, nozzle diameter, uniform design technique, 12 experimental data points
nozzle velocity, nozzle length, and printing resolution were were selected within a three-parameter, four-level data
initially compiled into the printing dataset. Subsequently, space. The SVM process optimization method presented a
a sub-dataset was created for the rheological properties, solution for analyzing the intricate 3D bioprinting process;
including measured viscosity and storage modulus data. the generated optimal combination of parameters resulted
A total of 537 data points pertaining to the printing resolution in the fabrication of high-fidelity prints (Figure 4). 45
were collected and then split into the training set, validation
set, and testing set. These data were utilized to train the 3. Meat flavor characterization
ML model and evaluate its learning performance. The
hierarchical ML model exhibited exceptional performance, The characterization of meat flavor extends beyond the
achieving the lowest training loss of 0.05 and a validation assessment of individual compounds responsible for odor
loss of 0.08. Consequently, the model attained the highest or taste. Although it may be straightforward to directly
level of accuracy in predicting the printing resolution. 44 associate a single compound with a particular flavor or
aroma characteristic, complexity arises when multiple
2.2. Optimization of printing parameters using ML flavor molecules interact, resulting in synergistic effects.
Apart from optimizing the material formulation for For instance, vanillin is known for enhancing perceived
optimal printability, an alternative approach is to optimize sweetness at low concentrations 46,47 despite possessing

Figure 3. Summary of the procedure for predicting printing resolution using a hierarchical M model based on rheological properties. Figure reproduced
from Oh et al. .
44

Volume 1 Issue 1 (2024) 8 https://doi.org/10.36922/ijamd.2279