DISCUSSION
Machine Learning Analysis of DNA Methylation to Detect Major Inflammatory Changes from Dietary Intervention
Izack Takazawa
Chaminade University
INTRODUCTION
METHODS
RESULTS
CONCLUSIONS
ACKNOWLEDGMENTS
Vasilescu, I.-A. & Esteban, R. (2022, August 16). Methylation Detection with Nanopore Sequencing: Reduced‑Representation Methylation Sequencing (RRMS). Oxford Nanopore Technologies. Retrieved June 17, 2025.
Lewerenz, J., & Maher, P. (2015). Chronic Glutamate Toxicity in Neurodegenerative Diseases-What is the Evidence?. Frontiers in neuroscience, 9, 469. https://doi.org/10.3389/fnins.2015.00469
Jutel, M., Akdis, M. and Akdis, C.A. (2009), Histamine, histamine receptors and their role in immune pathology. Clinical & Experimental Allergy, 39: 1786-1800. https://doi.org/10.1111/j.1365-2222.2009.03374.x
REFERENCES
Chronic inflammation is driven by dysregulated immune signaling and environmental factors, including diet, that contribute to persistent inflammation. Glutamate, also known as glutamic acid, is an amino acid that is the primary excitatory neurotransmitter in the brain. In excess amounts, it contributes to neurotoxicity and activation, promoting the release of pro-inflammatory cytokines such as TNF-α, IL-6, and IL-1β (Lewerenz & Maher, 2015). Similarly, histamine, a signaling chemical, plays a critical role in allergic responses and is released during immune activation, influencing cytokine secretion, and T-cell modulation (Jutel et al., 2009). It has been hypothesized that elevated levels of these molecules, either due to overproduction or dietary intake, can exacerbate inflammatory responses, especially in individuals who suffer from chronic inflammation. This project focuses on understanding how a dietary change, specifically a low-glutamate and low-histamine diet, affect inflammation at the epigenetic level.
Research Question:
What specific differences in DNA methylation exist between separate individuals, and how might these epigenetic changes be linked to chronic inflammatory response to glutamate and histamine?
Hypothesis:
It has been hypothesized that chronic inflammation is driven by altered methylation patterns, either by hypermethylation or hypomethylation. Will a low-glutamate and low-histamine diet shift these patterns toward reduced inflammatory gene activity?
This case comparison study investigated DNA methylation patterns in two individuals, Patient 0 and Patient 1, before and after undergoing a low-glutamate and low-histamine diet designed to reduce the inflammatory responses in Patient 1 who suffers from chronic inflammation.
Blood samples were collected from both participants at two timepoints: before and after a sustained low-glutamate, low-histamine diet. Peripheral blood mononuclear cells (PBMCs) were isolated, and DNA was extracted using the Qiagen Blood & Tissue Kit. Genomic DNA was sheared to ~6,000 base pairs and prepared for sequencing with the Oxford Nanopore Ligation Sequencing Kit. Sequencing was performed on a PromethION flow cell over five days. Basecalling was conducted using Dorado with the high-accuracy HAC_400bps@v4.3.0_5mc_5hmc model to detect 5-methylcytosine (5mC) modifications. DNA reads were demultiplexed, aligned to the hg38 human reference genome, and methylated CpG sites were identified and aggregated using the Modkit toolkit. To enrich for inflammation-related regions, adaptive sampling was applied based on the RRMS target region BED file.
Several pro-inflammatory and anti-inflammatory genes such as, TNF, IL6, IL1B, IL10, NFKB1, STAT3, CRP, PTGS2 (COX-2), were targeted to look at the differences at specific inflammatory regions within the data. These genes were located and verified using the UCSC Genome Browser (hg38 assembly), which allowed for precise genomic coordinate identification and visualization.
All analyses were conducted using Python, with key libraries including pandas, numpy, matplotlib, seaborn, and scikit-learn.
Dimensionality Reduction: Principal Component Analysis (PCA) and t-Distributed Stochastic Neighbor Embedding (t-SNE) were employed to reduce high-dimensional methylation data and reveal latent structure among CpG site profiles.
Unsupervised Learning: K-means clustering was used to group CpG sites and gene regions with similar methylation change trajectories, both within and between subjects and timepoints.
All computations were performed on Stampede3 using TACC’s Analysis Portal running Jupyter notebook and Jupyter Lab.
Methylation percentages at certain CpG regions differ between the two patients. After visualizing the data, the most prominent observation was the overall methylation shift in Patient 1 and overall differences found between the two patients. To highlight some of these differences Figures 5 - 12 show the shifts in methylation percentages in inflammatory gene regions, notably highlighting TNF-alpha, Interleukin-6, and Interleukin-1 beta with the largest differences between the before and after time period. Prior to the diet, Patient 1 reported chronic, full-body inflammation. After the diet, changes in methylation patterns suggest reduced expressions of these inflammatory genes, which may have contributed to her overall health improvement.
Interestingly, both Patient 0 and Patient 1 followed similar trends either both increasing or decreasing in methylation across most inflammatory genes after the diet. However within IL6, IL10, and PTGS2 (COX-2), they showed divergence between the two. Patient 1 showed decreased methylation at these sites, while Patient 0 showed increased methylation, most notably in IL10, a key anti-inflammatory cytokine.
To visualize the overall methylation changes, I applied unsupervised machine learning using k-means, PCA, t-SNE to analyze any similarities and differences within the methylation values of both Patient 0 and 1 prior and post dietary intervention. While high data dimensionality limited separation in PCA and t-SNE scatter plots were created, as I continued to analyze the visualizations, it became evident that due to the sheer amount of data and minute changes between the values it was hard to see any differences present. I then tried filtering sites with >20% difference in methylation values between Patient 0 and 1 and it improved interpretability.
Ultimately, heatmaps (Figures 1–4) using a diverging color map (red = high methylation, white = moderate, blue = low) effectively highlighted differences between each patient. Each vertical stripe represents a single genomic region with visible shifts, prominently shown in Patient 1.
These findings suggest early evidence that a low-glutamate, low-histamine diet may influence DNA methylation patterns associated with chronic inflammation. These shifts were particularly evident in key pro-inflammatory cytokine genes, suggesting epigenetic changes to dietary triggers. Heatmap visualizations revealed that Patient 1 initially exhibited higher methylation levels, which shifted toward a lower profile following the dietary intervention, suggesting that measurable epigenetic changes occurred over the course of the study. While further research with larger sample sizes is needed, these findings highlight the potential use of machine learning and methylation based biomarkers to understand inflammatory responses and their link to one’s diet. In the future, I would like to continue to look into the relationship between diet and inflammation with the goal of developing a diagnostic tool to help identify dietary triggers of chronic inflammatory responses.
Kelly Gaither, Amber Camp, Ethan Hill
Mentors: Logan Lasell, Ashley Sofia Alfaro
FIGURE 1.
FIGURE 2.
FIGURE 3.
FIGURE 4.
FIGURE 5.
FIGURE 6.
FIGURE 7.
FIGURE 8.
FIGURE 9.
FIGURE 10.
FIGURE 11.
FIGURE 12.
Introduction and Background
Glutamate
Histamine
Research Question & Hypothesis
Research Question:
What specific differences in DNA methylation exist between separate individuals, and how might these epigenetic changes be linked to chronic inflammatory response to glutamate and histamine?
Hypothesis:
It has been hypothesized that chronic inflammation is driven by altered methylation patterns, either by hypermethylation or hypomethylation. Will a low-glutamate and low-histamine diet shift these patterns toward reduced inflammatory gene activity?
Methods
PromethION Genome Sequencing
Qiagen DNA Extraction Kit
Results & Visualizations
FIGURE 5.
FIGURE 66.
FIGURE 7.
FIGURE 8.
FIGURE 9.
FIGURE 10.
FIGURE 11.
FIGURE 12.
Results & Visualizations
FIGURE 1.
FIGURE 2.
FIGURE 3.
FIGURE 4.
Conclusions & Next Steps
Conclusions
Next Steps
DISCUSSION
Machine Learning Analysis of DNA Methylation to Detect Major Inflammatory Changes from Dietary Intervention
Izack Takazawa
Chaminade University
INTRODUCTION
METHODS
RESULTS
CONCLUSIONS
ACKNOWLEDGMENTS
Vasilescu, I.-A. & Esteban, R. (2022, August 16). Methylation Detection with Nanopore Sequencing: Reduced‑Representation Methylation Sequencing (RRMS). Oxford Nanopore Technologies. Retrieved June 17, 2025.
Lewerenz, J., & Maher, P. (2015). Chronic Glutamate Toxicity in Neurodegenerative Diseases-What is the Evidence?. Frontiers in neuroscience, 9, 469. https://doi.org/10.3389/fnins.2015.00469
Jutel, M., Akdis, M. and Akdis, C.A. (2009), Histamine, histamine receptors and their role in immune pathology. Clinical & Experimental Allergy, 39: 1786-1800. https://doi.org/10.1111/j.1365-2222.2009.03374.x
REFERENCES
Chronic inflammation is driven by dysregulated immune signaling and environmental factors, including diet, that contribute to persistent inflammation. Glutamate, also known as glutamic acid, is an amino acid that is the primary excitatory neurotransmitter in the brain. In excess amounts, it contributes to neurotoxicity and activation, promoting the release of pro-inflammatory cytokines such as TNF-α, IL-6, and IL-1β (Lewerenz & Maher, 2015). Similarly, histamine, a signaling chemical, plays a critical role in allergic responses and is released during immune activation, influencing cytokine secretion, and T-cell modulation (Jutel et al., 2009). It has been hypothesized that elevated levels of these molecules, either due to overproduction or dietary intake, can exacerbate inflammatory responses, especially in individuals who suffer from chronic inflammation. This project focuses on understanding how a dietary change, specifically a low-glutamate and low-histamine diet, affect inflammation at the epigenetic level.
Research Question:
What specific differences in DNA methylation exist between separate individuals, and how might these epigenetic changes be linked to chronic inflammatory response to glutamate and histamine?
Hypothesis:
It has been hypothesized that chronic inflammation is driven by altered methylation patterns, either by hypermethylation or hypomethylation. Will a low-glutamate and low-histamine diet shift these patterns toward reduced inflammatory gene activity?
This case comparison study investigated DNA methylation patterns in two individuals, Patient 0 and Patient 1, before and after undergoing a low-glutamate and low-histamine diet designed to reduce the inflammatory responses in Patient 1 who suffers from chronic inflammation.
Blood samples were collected from both participants at two timepoints: before and after a sustained low-glutamate, low-histamine diet. Peripheral blood mononuclear cells (PBMCs) were isolated, and DNA was extracted using the Qiagen Blood & Tissue Kit. Genomic DNA was sheared to ~6,000 base pairs and prepared for sequencing with the Oxford Nanopore Ligation Sequencing Kit. Sequencing was performed on a PromethION flow cell over five days. Basecalling was conducted using Dorado with the high-accuracy HAC_400bps@v4.3.0_5mc_5hmc model to detect 5-methylcytosine (5mC) modifications. DNA reads were demultiplexed, aligned to the hg38 human reference genome, and methylated CpG sites were identified and aggregated using the Modkit toolkit. To enrich for inflammation-related regions, adaptive sampling was applied based on the RRMS target region BED file.
Several pro-inflammatory and anti-inflammatory genes such as, TNF, IL6, IL1B, IL10, NFKB1, STAT3, CRP, PTGS2 (COX-2), were targeted to look at the differences at specific inflammatory regions within the data. These genes were located and verified using the UCSC Genome Browser (hg38 assembly), which allowed for precise genomic coordinate identification and visualization.
All analyses were conducted using Python, with key libraries including pandas, numpy, matplotlib, seaborn, and scikit-learn.
Dimensionality Reduction: Principal Component Analysis (PCA) and t-Distributed Stochastic Neighbor Embedding (t-SNE) were employed to reduce high-dimensional methylation data and reveal latent structure among CpG site profiles.
Unsupervised Learning: K-means clustering was used to group CpG sites and gene regions with similar methylation change trajectories, both within and between subjects and timepoints.
All computations were performed on Stampede3 using TACC’s Analysis Portal running Jupyter notebook and Jupyter Lab.
Methylation percentages at certain CpG regions differ between the two patients. After visualizing the data, the most prominent observation was the overall methylation shift in Patient 1 and overall differences found between the two patients. To highlight some of these differences Figures 5 - 12 show the shifts in methylation percentages in inflammatory gene regions, notably highlighting TNF-alpha, Interleukin-6, and Interleukin-1 beta with the largest differences between the before and after time period. Prior to the diet, Patient 1 reported chronic, full-body inflammation. After the diet, changes in methylation patterns suggest reduced expressions of these inflammatory genes, which may have contributed to her overall health improvement.
Interestingly, both Patient 0 and Patient 1 followed similar trends either both increasing or decreasing in methylation across most inflammatory genes after the diet. However within IL6, IL10, and PTGS2 (COX-2), they showed divergence between the two. Patient 1 showed decreased methylation at these sites, while Patient 0 showed increased methylation, most notably in IL10, a key anti-inflammatory cytokine.
To visualize the overall methylation changes, I applied unsupervised machine learning using k-means, PCA, t-SNE to analyze any similarities and differences within the methylation values of both Patient 0 and 1 prior and post dietary intervention. While high data dimensionality limited separation in PCA and t-SNE scatter plots were created, as I continued to analyze the visualizations, it became evident that due to the sheer amount of data and minute changes between the values it was hard to see any differences present. I then tried filtering sites with >20% difference in methylation values between Patient 0 and 1 and it improved interpretability.
Ultimately, heatmaps (Figures 1–4) using a diverging color map (red = high methylation, white = moderate, blue = low) effectively highlighted differences between each patient. Each vertical stripe represents a single genomic region with visible shifts, prominently shown in Patient 1.
These findings suggest early evidence that a low-glutamate, low-histamine diet may influence DNA methylation patterns associated with chronic inflammation. These shifts were particularly evident in key pro-inflammatory cytokine genes, suggesting epigenetic changes to dietary triggers. Heatmap visualizations revealed that Patient 1 initially exhibited higher methylation levels, which shifted toward a lower profile following the dietary intervention, suggesting that measurable epigenetic changes occurred over the course of the study. While further research with larger sample sizes is needed, these findings highlight the potential use of machine learning and methylation based biomarkers to understand inflammatory responses and their link to one’s diet. In the future, I would like to continue to look into the relationship between diet and inflammation with the goal of developing a diagnostic tool to help identify dietary triggers of chronic inflammatory responses.
Kelly Gaither, Amber Camp, Ethan Hill
Mentors: Logan Lasell, Ashley Sofia Alfaro
FIGURE 1.
FIGURE 2.
FIGURE 3.
FIGURE 4.
FIGURE 5.
FIGURE 6.
FIGURE 7.
FIGURE 8.
FIGURE 9.
FIGURE 10.
FIGURE 11.
FIGURE 12.