The Resistance Paradox

Project Summary

The objective of this project was to identify and visualize collateral sensitivity in antibiotic-resistant bacteria using a given dataset. The final artifact is a self-contained, interactive Svelte component that renders a D3.js force-directed graph, enabling users to explore non-obvious antimicrobial relationships.

Phase 1: Data Preparation & Correlation Analysis

This phase focused on transforming raw categorical data into a quantitative format for statistical analysis. Source for dataset: Kaggle

• Data Preprocessing

  The dataset, containing patient metadata and categorical resistance phenotypes (R for Resistant, S for Susceptible), was loaded into a pandas DataFrame. Personally Identifiable Information (PII) was removed, and the analysis focused on antibiotic columns.

• Binarization

  Categorical data was binarized: Resistance (R) was mapped to 1, and all other states (S, I, nulls) were mapped to 0.

# Select only the antibiotic columns
antibiotic_columns = ['AMX/AMP', 'AMC', 'CIP', 'GEN', ...]
df_amr = df[antibiotic_columns].copy()

# Convert R to 1 and everything else to 0
for col in df_amr.columns:
    df_amr[col] = df_amr[col].apply(lambda x: 1 if x == 'R' else 0)


<div class="not-prose my-12 grid grid-cols-1 md:grid-cols-2 gap-8 items-center">
	<div>
		<img src="/data/Resistance_paradox/Resistance_paradox_result.jpg" alt="Resistance paradox" class="rounded-lg shadow-md" />
	</div>
	<div>
		<h4 class="text-xl font-bold">### Correlation Analysis</h4>
		<p class="text-muted-foreground">
A Pearson correlation matrix was calculated to quantify linear relationships between antibiotic resistance profiles, with coefficients ranging from -1 to +1.

```python

# Calculate the pairwise correlation of all antibiotic columns
correlation_matrix = df_amr.corr()

		</p>
	</div>
</div>


* ### Data Export
The correlation matrix was exported to a JSON file, serving as a static data source for the frontend component.


```markdown
correlation_matrix.to_json('resistance_correlations.json', orient='index')

Phase 2: Backend Logic (Client-Side Implementation)

To ensure portability and performance for a portfolio context, a client-side data lookup strategy replaced a traditional server-side predictive model. The resistance_correlations.json file acts as a static “backend,” eliminating the need for server maintenance or API calls, making the visualization self-contained.

Phase 3: Interactive Frontend Component (Svelte & D3.js)

This phase involved building a reusable Svelte component for data visualization.

Visualization Engine

D3.js was used to render a force-directed graph:

• Initialized d3.forceSimulation for physics-based node layout (antibiotics).
• Bound SVG <circle> elements to antibiotic data.
• Attached DOM events (.on('click', ...)) to capture user interactions.

Reactivity

User clicks trigger an updateSelection function that looks up correlation data. Svelte’s class: directive dynamically applies CSS classes to D3 nodes based on correlation coefficients.

<button
  class:selected={antibiotic.id === selectedAntibiotic}
  class:susceptible={predictions[antibiotic.id] < -0.5}
>
  {antibiotic.name}
</button>

Find the Superbug’s Secret Weakness

This interactive chart visualizes a fascinating phenomenon in microbiology called collateral sensitivity. Sometimes, when a bacteria becomes resistant to one drug, it develops a surprise weakness to another. Your goal is to find these hidden weaknesses.

How to Interact

Click any circle (an antibiotic) to see how bacteria resistant to it are affected by other drugs.

• Watch the colors change. Each color tells a part of the story.

• Drag the circles to rearrange the graph for a better view.

• Use the “Reset View” button to start a new discovery.

What the Colors Mean

• Red (Selected): This is the antibiotic the bacteria is already resistant to.
• Green (Rival): This is the jackpot! You've found a hidden weakness. The bacteria is now likely susceptible to this antibiotic.
• Orange (Teammate): The bacteria is likely resistant to these antibiotics as well.
• Grey (Neutral): These have no strong relationship with your selection.

Understanding the Controls

• Use Demo Data: Real-world data can be subtle. Check this box to load a perfect “textbook example” that is guaranteed to show all the color interactions.

• Enable Animation: This turns on the bacterium in the corner. Watch it grow stronger when it finds teammates (orange) or become vulnerable when you find a rival (green)!

Results & User Interaction

Results

The analysis identified multiple antibiotic pairs with strong negative correlations (r < -0.5), validating the collateral sensitivity hypothesis, as well as positive correlations indicating cross-resistance.

User Interaction

Users interact by clicking an antibiotic node in the graph, triggering an instant update. The color-coded feedback indicates:

• Selected (Red): The reference antibiotic to which a bacterium is resistant.
• Susceptible (Blue): Antibiotics likely effective, showing collateral sensitivity.
• Cross-Resistant (Orange): Antibiotics likely ineffective.
• Neutral (Faded): Uncorrelated antibiotics.