Modeling Drug Side-Effect Burden on Organ Systems

Project Introduction - Modeling Drug Side-Effect Burden on Organ Systems

Background:

Adverse drug events (ADEs) are a growing public health crisis, driving over 1.5 million U.S. emergency visits and nearly 500,000 hospitalizations each year. Polypharmacy (taking five or more medications simultaneously), sharply increases the risk of dangerous drug interactions and along with it comes the impact of its side-effects. Older adults and those with chronic conditions are most vulnerable to this.

Each individual drug has its own side-effects. When drugs have overlapping side-effect profiles, they can cumulatively target the same organ systems. This can potentially increase physiological stress even if each drug contributes very little individually.

This can lead to ADEs and prescribing cascades, resulting in increased financial and health costs to vulnerable patients and healthcare systems.

Problem Statement:

Current clinical tools generally look at drug interactions in pairs to analyze shared targets or pathways. These methods, which mainly focus on molecular interactions, can miss systemic off-target effects that are commonly known as drug side-effects.

These drug-drug interaction checking tools ignore side-effect-induced stress, which may not stem from direct drug interactions.

In case of multi-drug regimens, the drug side effects can overlap or accumulate. Multiple drugs may indirectly affect the same organ systems even when that is not their intended purpose. This cumulative effect may have a big physiological impact.

These critical gaps inspired me to build a computational tool that re-examines multi-drug regimens from a new perspective, from the angle of medication side effects, to help reveal hidden risks on organ systems previously not analyzed.

Engineering Goal:

Build a statistical model or tool that leverages weighted side-effect data to quantify and visualize the systemic burden of drug side effects on physiological systems in multi-drug regimens, additionally using a size- and ATC class-matched null model as a statistical baseline for comparison.

Databases/Libraries Used:

SIDER (Side Effect Resource), version 4.1 (http://sideeffects.embl.de/)

DrugBank Academic License XML, ver. 5.1.14 (https://go.drugbank.com/releases/latest)

MedDRA Hierarchy (https://www.meddra.org/)

Python libraries: pandas, scikit-learn, matplotlib, numpy, networkx, lxml, etc.

Methodology:

I started by integrating side-effect data from SIDER with drug information from DrugBank, mapping each adverse effect to organ systems using MedDRA SOC domains. To reduce noise, I developed a TF-IDF – based scoring approach and incorporated real-world side-effect frequencies.

I then built a statistical model to estimate the side-effect burden for any drug combination, accounting for overlapping drug targets and shared mechanisms of action within a regimen.

To benchmark these scores against a strong baseline, I used Monte Carlo sampling to generate hundreds of randomized drug combinations that preserve the original regimen’s both regimen size and ATC therapeutic class composition. I then compared observed results for each regimen to this baseline using Z-scores, allowing me to identify drug regimens with unusually high or low organ-system burden relative to expectations.

For any user-entered drug combination, my tool displays four graphs as output, visualizing this data in an intuitive format.

Expected Outcomes:

My tool takes a user-entered list of drugs and identifies the organ systems most affected by overlapping side effects. It outputs four graphs:

A radar chart shows how side-effect burden is distributed across the eight most affected organ systems for a set of drugs. Compared to the baseline null model, it highlights unusually high or low burden, with Z-scores indicating how far results deviate from the expected average.

A stacked bar chart breaks down the top five most affected organ systems and shows which drugs contribute the most to the overall burden in each one.

Another stacked-bar chart of the top five SOCs compares the additive burden versus the extra synergistic stress contribution from overlapping drug targets, displaying how shared mechanisms are computed to amplify total physiological burden compared to the additive model. This can allow comparison of predicted stress across different regimens.

A network graph maps connections between drugs based on shared biological targets. Drugs are shown as blue nodes, shared targets as red nodes, with edges linking them, making it easier to see how drugs in a combination may be related at a mechanistic level.

Overall, these graphs show how drug combinations may distribute and potentially intensify side effects across different organ systems in ways that are not visible when looking at each drug individually.

Conclusion:

By analyzing how the cumulative stress of drug side effects may affect the body, this tool reveals risks in drug combinations that are not obvious from how the drugs are designed or what they are intended to treat. Since medications can act in unexpected ways inside the body, side effects can oftentimes provide a more realistic picture of their true impact.

Sometimes different drugs in the same regimen can end up affecting the same organ systems without it being obvious. When this happens, their effects may accumulate and add extra strain on those systems. This kind of hidden overlap is easy to miss but can be crucial for patient safety, especially for people who already have existing vulnerabilities in that area.

As populations age, polypharmacy is rapidly becoming the norm rather than the exception. Millions of patients now take five or more medications daily, increasing the risk of dangerous side effects and hospitalizations.

My model helps reveal hidden risks within these complex regimens by showing how multiple drugs can collectively stress the same organ systems. By visualizing this cumulative burden, it gives clinicians an early warning signal before those risks become fatal.

Reflections:

I hope to improve this project by increasing the dataset size and incorporating real-time side-effect reports to better reflect real-world patient populations.

I would also like to expand my tool’s capabilities by adding features such as cheminformatics-based analysis of drug chemical structure. This could help identify shared mechanisms of action and better explain why two medications from different therapeutic classes have similar side effects.

This project has a lot of potential to grow because the problem it addresses is crucial and increasingly relevant to today’s aging population.

Acknowledgements:

Special thanks goes out to DrugBank, SIDER and MedDRA for allowing me to use their databases for my project:

Knox, C., Wilson, M., Klinger, C. M., et al. (2024). DrugBank 6.0: the DrugBank Knowledgebase for 2024. Nucleic Acids Research, 52(D1), D1265–D1275. https://doi.org/10.1093/nar/gkad976

Brown, E. G., Wood, L., & Wood, S. (1999). The Medical Dictionary for Regulatory Activities (MedDRA). Drug Safety, 20(2), 109-117. https://doi.org/10.2165/00002018-199920020-00002

Kuhn, M., Letunic, I., Jensen, L. J., & Bork, P. (2016). The SIDER database of drugs and side effects. Nucleic Acids Research, 44(D1), D1075-D1079. https://doi.org/10.1093/nar/gkv1075

I would like to thank my parents for supporting me throughout this journey. Also, I’d especially like to thank WRSTEF and the sponsors for providing me this invaluable opportunity to attend CWSF 2026.

References:

Berger, S. I., & Iyengar, R. (2011). Role of systems pharmacology in understanding drug adverse events. Wiley interdisciplinary reviews. Systems biology and medicine, 3(2), 129–135. https://doi.org/10.1002/wsbm.114

Srivastava S. B. (2023). Polypharmacy, Unintended Consequences, and Impact of Lifestyle Medicine. American journal of lifestyle medicine, 18(1), 54–57. https://doi.org/10.1177/15598276231207302

Pazan, F., & Wehling, M. (2021). Polypharmacy in older adults: a narrative review of definitions, epidemiology and consequences. European geriatric medicine, 12(3), 443–452. https://doi.org/10.1007/s41999-021-00479-3

Brouwers, L., Iskar, M., Zeller, G., van Noort, V., & Bork, P. (2011). Network neighbors of drug targets contribute to drug side-effect similarity. PloS one, 6(7), e22187. https://doi.org/10.1371/journal.pone.0022187

World Health Organization. (n.d.). Anatomical therapeutic chemical (ATC) classification. World Health Organization. Anatomical Therapeutic Chemical (ATC) Classification

Technical Scripter 2020. (2025, December 17). Understanding TF-IDF (Term Frequency-Inverse Document Frequency). GeeksforGeeks. Understanding TF-IDF (Term Frequency-Inverse Document Frequency) - GeeksforGeeks

Banack, H. R., Hayes-Larson, E., & Mayeda, E. R. (2022). Monte Carlo Simulation Approaches for Quantitative Bias Analysis: A Tutorial. Epidemiologic reviews, 43(1), 106–117. https://doi.org/10.1093/epirev/mxab012

Bushardt, R. L., Massey, E. B., Simpson, T. W., Ariail, A. L., & Simpson, K. N. (2008). Polypharmacy: Misleading, but manageable. Clinical Interventions in Aging, 3(2), 383-389. https://doi.org/10.2147/cia.s2468

Cascorbi I. (2012). Drug interactions--principles, examples and clinical consequences. Deutsches Arzteblatt international, 109(33-34), 546–556. https://doi.org/10.3238/arztebl.2012.0546

Elmore AR, Sadik A, Paternoster L, Khandaker GM, Gaunt TR, Hemani G (2025) Genetic inference of on-target and off-target side-effects of antipsychotic medications. PLoS Genet 21(7): e1011793. https://doi.org/10.1371/journal.pgen.1011793

Wang, L., Jiang, G., Li, D., & Liu, H. (2014). Standardizing adverse drug event reporting data. Journal of Biomedical Semantics, 5(1), 36. https://doi.org/10.1186/2041-1480-5-36