Page 569 - IJB-10-5

International Journal of Bioprinting ML-generated GelMA compression database

1. Introduction Within the BO algorithm, an acquisition function was used
to select data points for experimentation; the function
3D bioprinting can fabricate cell constructs that closely is reportedly biased toward selecting data points that
mimic the 3D cytoarchitecture of native tissues in vitro. maximize information gain about the model, indicated by
The physicochemical characteristics of the biomaterial and the relatively high levels of uncertainty (standard deviation)
the mechanical properties of the scaffold are paramount in its predictions. To improve the data collection
26
for supporting the targeted cell line, providing a temporary efficiency in the physical experiments, recommendations
environment until the tissue-specific extracellular matrix is for each iteration are provided in batches. These batches
1,2
produced. With the growing number of biomaterials and are conditioned such that all recommendations have the
composite bioinks in tissue engineering, there remains same GelMA concentration but allow for a broad range
3–5
a substantial gap in knowledge to precisely determine of values for other variables. This is a valuable tool for
the optimal mechanical properties for these biomaterials predicting the compression modulus of a selected bioink
to enhance cell-matrix interactions and promote cell to determine its optimal biological requirements, while
maturation and function.
significantly reducing the time needed for testing each
In tissue engineering, the mechanical properties variable individually.
(especially stiffness) of 3D-printed scaffolds have been
highlighted as a crucial element in modulating cell 2. Materials and methods
adhesion, growth, migration, and differentiation for tissue 2.1. Preparation of GelMA bioink
development. Hence, simulating the substrate rigidity Gelatin methacryloyl (GelMA), blended from 300-bloom
6–8
or softness of the native tissue target would constitute gelatin type A (synthesized as described previously in
a promising strategy in 3D biofabrication for tissue O’Connell et al. ), was used to prepare 5, 7.5, and 10%
19
engineering and regenerative medicine. The quantification (w/v) solutions. Appropriate amounts of freeze-dried
of biomaterial stiffness of tissue scaffolds is widely GelMA (degree of functionalization: 57%) were dissolved
performed using elastic, shear, and bulk moduli. Among in cell culture grade phosphate-buffered saline (PBS;
8,9
these, many studies focus primarily on the elastic modulus Sigma-Aldrich, USA), containing 100 U/mL penicillin and
as the preferred technique to analyze substrate stiffness. 10
100 μg/mL streptomycin (Gibco, USA), in a laminar flow
The stiffness of a scaffold can be modulated in the cabinet. The bioinks were heated at 37°C with intermittent
manufacturing and processing stages, 11,12 oftentimes vortexing to expedite dissolution.
by varying the reactant concentrations, 12–16 blending
biomaterials, 13,17 and modifying the crosslinker 2.2. Addition of photoinitiator/crosslinker to
concentration, ultraviolet light (UV) intensity (i.e., for the bioinks
photocrosslinking), and the duration of crosslinking in the Lithium phenyl-2,4,6-trimethylbenzoylphosphinate (LAP;
post-printing stage. 13,14,18,19 However, the optimization of TRICEP, Australia) was selected as the photoinitiator
these parameters to attain a specific scaffold stiffness is an for crosslinking GelMA at 405 nm (UV). An 8.5% (w/v)
extensive and time-consuming process. LAP stock solution was prepared in sterile PBS and
diluted into the GelMA bioink as required. The final LAP
With recent advancements, machine learning methods concentration range used in this study was 0.01–1% (w/v).
have been applied to fast-track and fine-tune the 3D The photoinitiator-containing bioink was then transferred
bioprinting process. 20–25 In this study, we utilized the to 10 cc printer reservoirs (Nordson EFD, USA) for printing
Bayesian optimization (BO) algorithm with a Gaussian using the GeSiM BioScaffolder (GeSiM, Germany).
process (GP) probabilistic model to predict scaffold
stiffness, i.e., compression modulus based on bioink 2.3. GelMA scaffold printing
concentration, crosslinker concentration, UV exposure Extrusion printing was performed using BioScaffolder
time, and the distance from the UV source (Figure 1). 3.2 (GeSiM, Germany). Lattice structures of 10 × 10 × 1.5
Gelatin methacryloyl (GelMA) was used as our model mm with 130 μm slicing were fabricated using 27-gauge
bioink as it is currently the most prevalent bioink for tissue smooth-flow tapered nozzles (200 μm; Nordson EFD,
engineering applications. An active sampling method USA). Reservoir temperature was optimized for 5%
5,15
was applied to iteratively select experimental points for (w/v), 7.5% (w/v), and 10% (w/v) GelMA bioinks. The
system modeling, with BO driving the active sampling printed scaffolds with variable LAP concentrations were
process and GP constructing the system model. Our exposed to 405 nm UV (Omnicure Lx400+; (Excelitas
experimentation and fine-tuning of the model continued Technologies, USA) at the specified distance and time as
until the model achieved a pre-specified degree of certainty. per values predicted by the BO framework (described in

Volume 10 Issue 5 (2024) 561 doi: 10.36922/ijb.3814