Page 69 - IJAMD-1-2

International Journal of AI for
Materials and Design
Machine learning for gel fraction prediction

3D bioprinting process, improving print accuracy, speed, off dialysis tubes and dialyzed against deionized water for
and resolution, while reducing material waste. 45,46 This 7 days at 40°C to remove unreacted methacrylic anhydride
approach can lead to the creation of highly customized and by-products from the initial reaction. After dialysis,
and optimized biomaterials for specific tissue engineering the GelMA solution was lyophilized for 7 days. Lyophilized
applications, ultimately enhancing the effectiveness and GelMA was kept at −20℃ in the refrigerator until further
precision of 3D bioprinting technologies. The large number use. The chemical process for the synthesis of GelMA is
of factors for the gel fraction makes it difficult to determine illustrated in Figure 1A.
a model analytically or with traditional empirical methods.
In such cases, ML is an attractive method to solve the 2.2. Preparation of varying opacity samples
problem. PEDOT:SPSS (Clevios PH1000, Heraeus) was adjusted to
the desired concentration stated in Table 1 using deionized
Therefore, in this work, we used ML to optimize the
curing of GelMA-PEDOT:SPSS hydrogels by predicting water. Lyophilized GelMA was added to the PEDOT:SPSS
solution and held at 37°C for 3h to ensure fully dissolved.
the gel fraction from three different feature groups. The Subsequently, LAP (L0290, TCI) was added to the GelMA-
purpose of the three feature groups is to understand the PEDOT:SPSS solution. The GelMA-PEDOT:SPSS solution
importance of different types of features on the prediction was casted into a cylindrical mold to yield samples with
accuracy and for different real-world applications. Feature 8 mm diameter, and 2 mm height. The sample will be
Group 1 utilizes bioink formulation and crosslinking crosslinked with UV light as shown in Figure 1B.
parameter as input. This is useful to predict the gel fraction
before the experiment as the optimized parameter for the 2.3. Data collection for gel fraction
desired gel fraction can be selected without a wide range of
experiment, aiding in saving material cost and shortening 2.3.1. Experiment setup
experiment time. Feature Groups 2 and 3 predict the gel A high-intensity spot-curing system, Dymax BlueWave
fraction without the prior crosslinking parameter. Feature QX4, equipped with a 405 nm LED wand (VisiCure LED
Group 2 has absorption coefficient only as the input for head, 405 nm), was used as the UV light source. UV light
ML model, while feature Group 3 is a combination of transmission through the hydrogel was measured using a
bioink formulation and absorption coefficient. The usage UV light sensor (S120VC, Thorlabs) covered with a square
of absorption coefficient with no crosslinking parameter pinhole with a size of 1 mm × 1 mm (S1000QK, Thorlabs),
as input in feature Groups 2 and 3 will be significant for connected to an optical power and energy meter console
in situ monitoring for gel fraction. In real application, (PM400, Thorlabs). The LED wand was mounted on a retort
the UV power intensity and the UV exposure duration stand and positioned directly above the UV light sensor.
will most likely be different from the input setting due to The power output is controlled by adjusting the height of
imperfection of machine or human error. By predicting the the LED wand. For the mold setup, a clean glass slide was
gel fraction with the measurement of absorption coefficient used as the base beneath the mold. Two clips were used
instead of crosslinking parameter, the gel fraction can be for the securing of the mold to prevent liquid leakage from
fine-tuned in real time with a non-destructive method to between the mold and the glass slide, as seen in Figure 2A.
improve the precision of the sample, at a relatively low cost Post-crosslinking, the clips were removed to separate the
as only a UV sensor is required for the measurement. mold from the glass slide for hydrogel extraction, as shown
in Figure 2B.
2. Methods
2.3.2. Crosslinking of GelMA-PEDOT:SPSS hydrogel
2.1. Synthesis of GelMA
The prepared and mixed hydrogel solution (120 µL) was
GelMA was synthesized according to previously pipetted into the assembled mold setup. The mold was
procedures described by Loessner et al. Methacrylic placed above the UV light sensor, as shown in Figure 2C,
47
anhydride (Sigma-Aldrich) of 1.4 mL was added dropwise while curing the sample, to record the change in the
into a 10% w/v gelatin type A (bloom strength 300, Sigma- received UV intensity during the crosslinking process.
Aldrich) solution dissolved in 100 mL of 1× phosphate- The UV light was shone on the mold for a specified power
buffered saline (1× PBS, pH 7.2, Vivantis). The solution and duration, allowing for the photopolymerization
was stirred at 400 rpm while maintaining a temperature process to occur. After the crosslinking process, the
at 50°C, and 400 mL of 1× PBS was added to quench hydrogel was removed from the mold and transferred into
the reaction after 3 h. The mixture was transferred into a 35 mm petri dish of known mass, m . The experiment
p
50 mL tubes and centrifuged at 3500× g for 3 min. The was repeated three times for every combination of bioink
clear supernatant was transferred into 12 – 14 kDa cut- formulations, UV power intensity, and UV duration

Volume 1 Issue 2 (2024) 63 doi: 10.36922/ijamd.3807