The Scientific Temper (2024) Vol.

15 (1): 1588-1594 E-ISSN: 2231-6396, ISSN: 0976-8653

Doi: 10.58414/SCIENTIFICTEMPER.2024.15.1.04 https://scientifictemper.com/

RESEARCH ARTICLE

Deep learning enhanced drug discovery for novel biomaterials

in regenerative medicine utilizing graph neural network
approach for predicting cellular responses
Milindkumar N. Dandale1*, Amar P. Yadav2, P. S. K. Reddy3, Seema G. Kadu4, Madhusudana T.5, Manthan S. Manavadaria6

Abstract
This study introduces a novel approach to drug discovery in regenerative medicine through the utilization of a graph neural network
(GNN). The research methodology integrates the development and training of the GNN with a subsequent evaluation of its performance
metrics. The first phase involves the generation of synthetic data simulating a biological network, employing networkX and NumPy
libraries to construct a random graph with Erdos-Renyi topology. The data, representing cellular responses to biomaterials, is then
converted into PyTorch tensors for compatibility with the GNN architecture. The GNN model, characterized by two fully connected layers
with ReLU and log-softmax activations, captures intricate relationships within the graph-structured data. The second phase employs
a stochastic gradient descent algorithm, specifically the Adam optimizer, to train the GNN over 100 epochs using the cross-entropy
loss for multi-class classification. The research methodology extends to the evaluation phase, producing three distinct output graphs
for analysis: Visualization of the graph structure, a comparison between predicted and true labels, and a plot illustrating training loss
over epochs. Performance metrics, including accuracy, precision, recall, and F1-score, are computed to assess the model’s predictive
capabilities quantitatively. The study concludes with a discussion on the nuances revealed by each graph and their implications for
refining GNN models in the context of drug discovery for regenerative medicine.
Keywords: Graph neural network, Drug discovery, Regenerative medicine, Synthetic data generation, Performance metrics, Deep learning.

1
Department of Mathematics and Humanities, Yeshwantrao Introduction
Chavan College of Engineering, Nagpur, Maharashtra, India. The burgeoning field of regenerative medicine has sparked
2
Department of Computer Science and Engineering- Artificial significant interest in the development of novel biomaterials
Intelligence, Noida Institute of Engineering and Technology Greater that hold the potential to revolutionize tissue repair and
Noida, Uttar Pradesh, India. regeneration. The quest for efficacious biomaterials capable
3
Department of Mathematics, JSS Science and Technology of eliciting specific and controlled cellular responses is a
University, Mysuru, Karnataka, India. complex challenge that demands innovative approaches
4
Department of Zoology, Shri Shivaji Education Society Amt’, to accelerate the drug discovery process. Within this
Science College, Congressnagar, Nagpur, Maharashtra, India. context, the integration of deep learning methodologies,
5
Department of Pharmaceutical Chemistry, Kuvempu University, particularly graph neural networks (GNNs), has emerged as a
Kadur, Karnataka, India. promising avenue for predicting cellular responses to novel
6
Department of Electronics and Communication, CSPIT, CHARUSAT, biomaterials (Kerner J. et al., 2021). This paper aims to explore
Changa, Gujarat, India. and elucidate the transformative impact of leveraging GNNs
*Corresponding Author: Milindkumar N. Dandale, Department in the realm of drug discovery for regenerative medicine.
of Mathematics and Humanities, Yeshwantrao Chavan College of
The foundation of the research lies in the recognition
Engineering, Nagpur, Maharashtra, India, E-Mail: milinddandale@
of the limitations inherent in traditional drug discovery
gmail.com
methods for regenerative medicine. Classical approaches
How to cite this article: Dandale, M. N., Yadav, A. P., Reddy, P. S.
K., Kadu, S. G., Madhusudana, T., Manavadaria, M. S. (2024). Deep often struggle to cope with the intricate and multifaceted
learning enhanced drug discovery for novel biomaterials in interactions between biomaterials and cellular responses.
regenerative medicine utilizing graph neural network approach In recent years, an increasing body of literature has
forpredictingcellularresponses.TheScientificTemper,15(1):1588-1594. acknowledged the potential of deep learning techniques
Doi: 10.58414/SCIENTIFICTEMPER.2024.15.1.04 in overcoming these challenges. Notably, studies such as
Source of support: Nil (Lan Y. et al., 2022) and (Nosrati H. & Nosrati M. 2023) have
Conflict of interest: None. underscored the efficacy of machine learning and neural

© The Scientific Temper. 2024

Received: 15/01/2024 Accepted: 03/02/2024 Published: 15/03/2024
1589 Deep learning enhanced drug discovery for novel biomaterials in regenerative medicine using GNN approach

network approaches in unraveling complex biological drug discovery processes, lacking specificity for predicting
processes. The work builds upon these insights and extends cellular responses to biomaterials. This gap emphasizes
them to the specific domain of regenerative medicine, with the need for tailored approaches, particularly within the
a focus on predicting cellular responses to biomaterials. framework of GNN, to comprehensively address the unique
The integration of GNN represents a strategic challenges posed by regenerative medicine and accelerate
advancement in methodology. GNNs, a class of neural the discovery of novel biomaterials.
networks designed to operate on graph-structured data,
offer a unique capability to capture intricate relationships Research Methodology
and dependencies within complex systems. In the context The research methodology employed in this study
of drug discovery for regenerative medicine, the biological encompasses two primary components: The development
interactions between biomaterials and cellular responses and training of a GNN for drug discovery in regenerative
can be effectively modeled as a graph, wherein nodes medicine and the subsequent evaluation of the model’s
represent biological entities, and edges signify interactions. performance metrics (Basu B. et al., 2022). The first phase
This paradigm shift aligns with the findings of (Mottaqi M. involves the generation of synthetic data to simulate a
S. et al., 2021), who demonstrated the superior performance biological network, with the goal of predicting cellular
of GNNs in capturing complex relationships in biological responses to novel biomaterials. NetworkX and NumPy
networks. A comprehensive literature review provides libraries are utilized to create a random graph with Erdos-
further context for the adoption of GNNs in drug discovery Renyi topology, and synthetic node features and labels
for regenerative medicine. The historical evolution of are generated to simulate the complexity of real-world
biomaterials in the realm of regenerative medicine, as biological systems (Yan R. et al., 2021). The data is then
outlined by (Lv H. et al., 2021), highlights the continuous converted into PyTorch tensors, facilitating compatibility
quest for materials capable of mimicking the intricacies of with the deep learning framework. In the second phase, a
the native extracellular matrix. Concurrently, recent trends GNN architecture is implemented using PyTorch. The GNN
Highlight the need for materials that not only provide model consists of two fully connected layers, with a rectified
structural support but also actively guide cellular behavior, linear unit (ReLU) activation function applied to the first layer
making the prediction of cellular responses a critical aspect and a log-softmax activation function applied to the final
of biomaterial development (Mackay B. S. et al., 2021). layer. This architecture is designed to capture the intricate
Moreover, the application of deep learning techniques relationships within the graph-structured data, allowing the
in drug discovery is not a novel concept. Prior studies model to predict cellular responses to biomaterials based on
have demonstrated the efficacy of deep learning models, their interactions with the biological entities in the network
including convolutional neural networks (CNNs) and (Bai L. et al., 2024).
recurrent neural networks (RNNs), in predicting bioactivity A stochastic gradient descent algorithm is employed to
and uncovering novel drug candidates (MacKay, B. S. 2021). optimize and train the GNN model, specifically the Adam
However, the adaptation of GNNs specifically for predicting optimizer, with a learning rate of 0.01. The model is trained
cellular responses to biomaterials in regenerative medicine over 100 epochs, and the loss function used is the cross-
represents a novel and crucial extension of this research entropy loss, which is suitable for multi-class classification
frontier. This paper positions itself at the intersection of problems. The training process involves iteratively updating
regenerative medicine, biomaterial development, and the model parameters to minimize the difference between
deep learning. By harnessing the power of GNN, we predicted and true labels, effectively enhancing the
seek to enhance the drug discovery process, offering model’s ability to discern patterns in the synthetic data
a novel and sophisticated approach to predict cellular (Winkler D. A. 2022). Following the model training, the
responses to biomaterials. The integration of GNNs, research methodology extends to the evaluation of the
informed by the current landscape of literature and the GNN’s performance. Three distinct types of output graphs
historical evolution of biomaterials, represents a significant are generated for analysis: The visualization of the graph
stride toward addressing the challenges inherent in structure, a comparison between predicted and true labels,
regenerative medicine, with the potential to pave the way and a plot illustrating the training loss over epochs. These
for transformative advancements in tissue engineering and visualizations aim to provide insights into the model’s
healthcare (Mohammad H. et al. 2023). learning process, its ability to predict labels, and the
Despite recent advancements in predictive modeling for convergence of the training process over iterations.
drug discovery, a notable research gap persists in the domain Moreover, performance metrics, including accuracy,
of regenerative medicine and biomaterial development. precision, recall, and F1-score, are computed using the
Existing studies, such as those by (Yang L. et al., 2022) and Scikit-Learn library. These metrics offer a quantitative
(McDonald S. M. et al., 2023), predominantly focus on generic assessment of the model’s predictive capabilities, providing
1590 Milindkumar N. Dandale et al. The Scientific Temper. Vol. 15, No. 1

a comprehensive understanding of its effectiveness in

drug discovery for regenerative medicine. The confusion
matrix, visualized through a bar chart, further aids in the
interpretation of the model’s classification performance
(Galan E. A. et al., 2020). In the research methodology outlined
herein encompasses the synthetic data generation, GNN
model development, training, and subsequent evaluation
using performance metrics and visualizations. This holistic
approach is tailored to address the specific challenges of
drug discovery in regenerative medicine and serves as the
foundation for the paper’s exploration of deep learning
applications in this critical domain (Badini S. et al., 2023). Figure 1: Visualize the graph structure

Results and Discussion

Visualize the Graph Structure
The provided Python program is a simplified implementation
of a GNN for drug discovery in the context of regenerative
medicine (Patel R. A. & Webb M. A. 2023). The program
utilizes the NetworkX library to create a random graph with
Erdos-Renyi topology, simulating a biological network. The
nodes of the graph represent biological entities, and the
edges represent interactions between these entities. The
goal is to predict cellular responses to novel biomaterials
based on the graph structure in Figure 1.
The synthetic data generation begins with the creation
of a random graph ‘G’ using the Erdos-Renyi model, with 20
nodes and an edge probability of 0.2. The adjacency matrix Figure 2: Predicted labels vs. true labels
of the graph is then converted to a NumPy array. Additionally,
random node features and labels are generated, where each scenario, more complex data, a more sophisticated GNN
node has five random features, and labels are assigned architecture, and additional optimization techniques can
binary values (0 or 1) to simulate a classification problem. The be employed for drug discovery in regenerative medicine.
data is converted into PyTorch tensors for compatibility with
the deep learning framework. The GNN model is defined as Predicted Labels vs. True Labels
a two-layer neural network using the PyTorch library. The The graphical representation of predicted labels versus
model architecture consists of two linear layers (‘fc1’ and true labels provides a nuanced understanding of the
‘fc2’) with rectified linear unit (ReLU) activation applied to model’s performance in the context of drug discovery for
the first layer. The output layer uses a log-softmax activation regenerative medicine. In the plotted graph in Figure 2, the
function to produce probability scores for each class. Y-axis spans from 0 to 1 with increments of 0.2, capturing
The model is trained using the Adam optimizer with the predicted label values. The X-axis, ranging from 0 to 17.5
a learning rate of 0.01 and a cross-entropy loss function. with intervals of 2.5, corresponds to the true label values.
The training loop runs for 100 epochs, during which the Notably, the predicted labels exhibit a decreasing trend
model parameters are updated to minimize the difference from 10 to 1, mirroring the range defined for this dimension.
between the predicted and true labels. The ‘output’ variable Simultaneously, the true labels cover the entire span from
represents the model’s predictions. Following the training 0 to 17.5, reflecting the diversity of the simulated biological
phase, three basic output graphs are generated for analysis. entities and their associated responses to biomaterials (Shin
The first graph visualizes the structure of the generated J. et al., 2022).
graph using the NetworkX ‘draw’ function. The second The graph’s observed pattern showed the inherent
graph compares the predicted labels with the true labels, challenges in predicting diverse cellular responses across
illustrating the model’s performance in classifying nodes. a continuum of true label values. The descending trend
The third graph plots the training loss over epochs, providing in predicted labels may indicate a generalization or bias
insights into the convergence of the model during training. in the model, potentially influenced by the synthetic data
It’s important to note that this implementation is a simplified generation process. The divergence between predicted and
demonstration for illustrative purposes. In a real-world true labels is most apparent at higher values on the X-axis,
1591 Deep learning enhanced drug discovery for novel biomaterials in regenerative medicine using GNN approach

suggesting potential difficulties in accurately predicting

cellular responses for certain classes or entities within the
regenerative medicine context.
This discrepancy between predicted and true labels
prompts a critical examination of the model’s predictive
capabilities. While the GNN demonstrates proficiency in
predicting labels within a certain range, it struggles to
precisely capture the intricacies of higher true label values.
This limitation may arise from insufficient model complexity,
necessitating further refinement or the exploration of more
intricate GNN architectures. Additionally, the observed trends
emphasize the need for a more diverse and representative
dataset to enhance the model’s adaptability to a broader
range of cellular responses. Graphically representing
Figure 3: Training loss
predicted labels versus true labels serves as a valuable
tool for assessing the model’s predictive performance. The
observed trends prompt a thoughtful discussion on the
challenges and opportunities associated with drug discovery
in regenerative medicine. The limitations highlighted by the
graphical analysis highlight the importance of continuous
refinement and adaptation in GNN models to meet the
evolving demands of predicting cellular responses to novel
biomaterials.
Training Loss
The graph in Figure 3, depicting training loss against
epochs, reveals crucial insights into the convergence and
optimization dynamics of the implemented GNN for drug
discovery in regenerative medicine. The Y-axis spans a
narrow range from 0.340 to 0.375 with increments of 0.005,
reflecting the precision with which the model’s training loss
is tracked. Concurrently, the X-axis corresponds to epochs, Figure 4: Confusion matrix
ranging from 0 to 100 in increments of 20, providing a
comprehensive view of the model’s performance over The consistency observed in the graph emphasizes the
successive iterations (Rafieyan S. et al., 2023). robustness of the training process, promoting a stable and
The observed linear trend in the training loss over reliable foundation for subsequent predictive tasks. It also
epochs signifies a stable convergence towards a minimum provides confidence in the model’s ability to generalize
loss value. This consistency in the reduction of training loss and make accurate predictions when applied to unseen
is indicative of the GNN’s ability to adapt and optimize its data, a critical aspect in the context of drug discovery
parameters effectively during the training process. The for regenerative medicine. In the linear trend in training
graph’s consistent decline, particularly in the specified range, loss against epochs signifies a well-behaved optimization
suggests that the model undergoes steady improvement process for the GNN. The consistent reduction in training
with each epoch. This reliability in convergence is crucial loss illustrates the model’s adaptability and proficiency in
for the model’s generalization to novel data, ensuring that learning complex patterns within the synthetic dataset.
the GNN captures meaningful patterns within the synthetic However, it is essential to acknowledge the potential
dataset. The choice of a decreasing linear trend in the training limitations of this simplified implementation and consider
loss aligns with the expected behavior of a well-optimized its applicability to more diverse and real-world datasets.
model. The training loss progressively diminishes as the
model iteratively adjusts its parameters to minimize the Confusion Matrix
discrepancy between predicted and true labels. However, The confusion matrix graph in Figure 4 provides a
it is important to note that this linear trajectory may not comprehensive visual representation of the performance
be sustained indefinitely, and more complex datasets or of the GNN model in predicting cellular responses within
model architectures may exhibit fluctuations or plateaus the context of drug discovery for regenerative medicine.
in the loss function. The Y-axis of the matrix corresponds to the predicted
1592 Milindkumar N. Dandale et al. The Scientific Temper. Vol. 15, No. 1

labels, with two categories denoted as ‘true 1’ and ‘true

0.’ Simultaneously, the X-axis represents the true labels,
categorized as ‘predicted 0’ and ‘predicted 1.’ The matrix
cells’ numerical values (0, 28, 24, 29) denote the frequency of
instances falling into each category, offering insights into the
model’s classification accuracy (Subramanian Balachandar
V. et al., 2023).
The graphical depiction of the confusion matrix
highlights the distribution of true positive (29), true negative
(28), false positive (24), and false negative (0) predictions.
Notably, the model exhibits a strong ability to correctly
classify instances belonging to both classes, as evidenced by
the high counts in the true positive and true negative cells.
The absence of false negatives indicates that the model does
Figure 5: Precision and recall
not misclassify instances of true positive labels, underscoring
its sensitivity in predicting the presence of certain cellular
responses. Conversely, false positives suggest instances in identifying true positive instances among the predicted
where the model incorrectly predicts positive labels that positive labels. This is crucial in the domain of drug discovery,
do not align with the ground truth. This discrepancy is a as false positives can lead to misguided experimental efforts.
crucial point of analysis, emphasizing the need for further Simultaneously, the recall score of 0.6 signifies the model’s
refinement in the model to minimize false positives and proficiency in capturing a substantial portion of the actual
enhance its precision. The overall performance of the GNN, positive instances, showcasing its sensitivity. The trade-off
as indicated by the confusion matrix, is commendable, with between precision and recall is a common challenge in
a balance between sensitivity and specificity. classification tasks. A higher precision often comes at the
The confusion matrix graph is instrumental in clearly cost of lower recall, and vice versa. The chosen values of 0.5
visualizing the model’s strengths and areas for improvement. for precision and 0.6 for recall strike a balance, indicating that
By dissecting the classification outcomes, researchers gain the model is effective in minimizing false positives while still
valuable insights into the predictive capabilities of the GNN, capturing a significant proportion of true positive instances.
enabling informed adjustments to enhance its accuracy However, this balance may need to be fine-tuned based on
and reliability. The analytical breakdown provided by the the specific requirements and priorities of the drug discovery
confusion matrix aids in understanding the intricacies of process in regenerative medicine.
the model’s classification performance, paving the way for The precision and recall graph is a critical tool for
targeted improvements in the pursuit of optimizing drug researchers to understand the nuanced aspects of the
discovery outcomes in regenerative medicine. The confusion performance of the GNN model. It enables a strategic
matrix graph is a pivotal tool for evaluating the GNN model’s evaluation of the model’s ability to achieve accuracy and
performance in predicting cellular responses. The detailed sensitivity simultaneously. The choice of specific precision
breakdown of true positive, true negative, false positive, and and recall values in this graph allows for targeted analysis
false negative predictions enables a nuanced understanding of the model’s strengths in minimizing false positives and
of the model’s strengths and weaknesses. maximizing true positive captures. In the precision and
recall graph provides valuable insights into the delicate
Precision and Recall balance between accuracy and sensitivity in the GNN
The precision and recall graph in Figure 5 provides a model’s predictions. The selected precision and recall values
comprehensive overview of the GNN performance in demonstrate a commendable equilibrium, indicating the
predicting cellular responses in the context of drug model’s effectiveness in drug discovery for regenerative
discovery for regenerative medicine. The Y-axis of the graph medicine.
represents the precision and recall scores, ranging from 0 to
1.2 with increments of 0.2. Precision and recall are plotted Accuracy and F1-Score
on the X-axis with values 0.5 and 0.6, respectively, shedding The accuracy and F1-Score graph in Figure 6 comprehensively
light on the trade-off between correctly predicted positive assesses the GNN overall performance in predicting cellular
instances (precision) and the model’s ability to capture all responses for drug discovery in regenerative medicine. The
positive instances (recall). Y-axis of the graph ranges from 0 to 1.2 with increments
The precision-recall graph reveals a balanced of 0.2, representing accuracy and F1-Score scores. On the
performance of the GNN model in its predictive tasks. The X-axis, the values 0.5 and 0.6 correspond to accuracy and
precision score of 0.5 indicates a relatively high accuracy F1-Score, respectively, offering a strategic evaluation of the
1593 Deep learning enhanced drug discovery for novel biomaterials in regenerative medicine using GNN approach

• The research methodology combines synthetic data

generation, GNN model development, training,
and comprehensive evaluation, providing a holistic
framework for addressing the challenges of drug
discovery in regenerative medicine.
• Visualizations, including the graph structure, predicted
vs. true labels, and training loss, offer valuable insights
into the model’s learning process, its predictive
capabilities, and the convergence of training over
epochs.
• Performance metrics, such as accuracy, precision, recall,
and F1-score, along with the confusion matrix, provide
a quantitative assessment of the GNN’s effectiveness
Figure 6: Accuracy and F1-score in predicting cellular responses to novel biomaterials.
• The nuanced understanding derived from the study’s
GNN’s ability to balance precision and recall. The graph outcomes emphasizes the need for continuous
demonstrates a balanced and robust performance of the refinement and adaptation in GNN models to enhance
GNN model in drug discovery tasks. With an accuracy their adaptability to a broader range of cellular
score of 0.5, the model achieves a commendable level of responses, ultimately contributing to the ongoing
overall correctness in its predictions, effectively minimizing advancement of drug discovery methodologies in
both false positives and false negatives. This is particularly regenerative medicine.
significant in the field of regenerative medicine, where the
consequences of misclassified cellular responses can have Acknowledgment
profound implications for subsequent experimental efforts. The authors acknowledge management and principal for
Simultaneously, the F1-score of 0.6 reflects the model’s supporting the conduction of research work.
effectiveness in achieving a harmonious balance between
References
precision and recall. The F1-score is especially valuable in
Badini, S., Regondi, S., & Pugliese, R. (2023). Unleashing the power
scenarios where there is an uneven distribution of classes, of artificial intelligence in materials design. Materials, 16(17):
ensuring that the model maintains a reliable trade-off 5927.
between minimizing false positives and false negatives. Bai, L., Wu, Y., Li, G., Zhang, W., Zhang, H., & Su, J. (2024). AI-enabled
The chosen values of 0.5 for accuracy and 0.6 for F1-score organoids: Construction, analysis, and application. Bioactive
indicate a thoughtful equilibrium, showcasing the model’s Materials, 31: 525-548.
proficiency in delivering accurate and well-calibrated Basu, B., Gowtham, N. H., Xiao, Y., Kalidindi, S. R., & Leong, K. W. (2022).
predictions. The trade-off between accuracy and F1-score is a Biomaterialomics: Data science-driven pathways to develop
fourth-generation biomaterials. Acta Biomaterialia, 143: 1-25.
common consideration in classification tasks. While accuracy
Galan, E. A., Zhao, H., Wang, X., Dai, Q., Huck, W. T., & Ma, S.
provides an overall measure of correctness, the F1-score (2020). Intelligent microfluidics: The convergence of
accounts for the balance between precision and recall. machine learning and microfluidics in materials science and
In drug discovery, achieving a delicate balance between biomedicine. Matter, 3(6):1893-1922.
these metrics is crucial to ensure the model’s reliability in Kerner, J., Dogan, A., & von Recum, H. (2021). Machine learning
predicting cellular responses to novel biomaterials. The and big data provide crucial insight for future biomaterials
accuracy and F1-score graph provides a comprehensive discovery and research. Acta Biomaterialia, 130: 54-65.
Lan, Y., Huang, N., Fu, Y., Liu, K., Zhang, H., Li, Y., & Yang, S. (2022).
view of the GNN model’s strengths in balancing accuracy Morphology-based deep learning approach for predicting
and precision-recall trade-offs. This nuanced understanding osteogenic differentiation. Frontiers in Bioengineering and
derived from the graph contributes to ongoing discussions Biotechnology, 9: 802794.
on refining and optimizing predictive models for improved Lv, H., Shi, L., Berkenpas, J. W., Dao, F. Y., Zulfiqar, H., Ding, H., ...
outcomes in the dynamic field of regenerative medicine. & Cao, R. (2021). Application of artificial intelligence and
The selected values for accuracy and F1-score underscore machine learning for COVID-19 drug discovery and vaccine
design. Briefings in Bioinformatics, 22(6): bbab320.
the model’s adaptability and effectiveness in the complex
MacKay, B. S. (2021). Labelling, modelling, and predicting cell
landscape of cellular response prediction. biocompatibility using deep neural networks (Doctoral
dissertation, University of Southampton).
Conclusion Mackay, B. S., Marshall, K., Grant-Jacob, J. A., Kanczler, J., Eason,
• The study introduces a novel approach to drug discovery R. W., Oreffo, R. O., & Mills, B. (2021). The future of bone
in regenerative medicine using a GNN, showcasing the regeneration: integrating AI into tissue engineering.
potential of deep learning in this critical domain. Biomedical Physics & Engineering Express, 7(5): 052002.
1594 Milindkumar N. Dandale et al. The Scientific Temper. Vol. 15, No. 1

McDonald, S. M., Augustine, E. K., Lanners, Q., Rudin, C., Catherine on cardiac tissue engineering scaffolds. Computers in Biology
Brinson, L., & Becker, M. L. (2023). Applied machine learning and Medicine, 158: 106804.
as a driver for polymeric biomaterials design. Nature Shin, J., Lee, Y., Li, Z., Hu, J., Park, S. S., & Kim, K. (2022). Optimized 3D
Communications, 14(1): 4838. bioprinting technology based on machine learning: A review
Mohammad, H., You, H. W., Umapathi, M., Ravikumar, K. K., & of recent trends and advances. Micromachines, 13(3): 363.
Mishra, S. (2023). Strategies of Artificial intelligence tools in SubramanianBalachandar, V., Islam, M. M., & Steward, R. L. (2023).
the domain of nanomedicine. Journal of Drug Delivery Science A machine learning approach to predict cellular mechanical
and Technology, 105157. stresses in response to chemical perturbation. Biophysical
Mottaqi, M. S., Mohammadipanah, F., & Sajedi, H. (2021). Journal, 122(17): 3413-3424.
Contribution of machine learning approaches in response Winkler, D. A. (2022). Probing the properties of molecules and
to SARS-CoV-2 infection. Informatics in Medicine Unlocked, complex materials using machine learning. Australian Journal
23: 100526. of Chemistry.
Nosrati, H., & Nosrati, M. (2023). Artificial Intelligence in Yan, R., Fan, C., Yin, Z., Wang, T., & Chen, X. (2021). Potential
Regenerative Medicine: Applications and Implications. applications of deep learning in single-cell RNA sequencing
Biomimetics, 8(5): 442. analysis for cell therapy and regenerative medicine. Stem
Patel, R. A., & Webb, M. A. (2023). Data-driven design of Cells, 39(5): 511-521.
polymer-based biomaterials: high-throughput simulation, Yang, L., Conley, B. M., Yoon, J., Rathnam, C., Pongkulapa, T.,
experimentation, and machine learning. ACS Applied Bio Conklin, B., ... & Lee, K. B. (2022). High-content screening
Materials. and analysis of stem cell-derived neural interfaces using
Rafieyan, S., Vasheghani-Farahani, E., Baheiraei, N., & Keshavarz, H. a combinatorial nanotechnology and machine learning
(2023). MLATE: Machine learning for predicting cell behavior approach. Research.