Drug-Target Network Analysis Dashboard

Executive Summary

This dashboard presents comprehensive results from applying the DMY shortest-path algorithm to drug-target networks, including both single-objective optimization and multi-objective Pareto front analysis.

Key Findings:

1. Single-objective: Celecoxib remains the most COX-2 selective option (~3.7× vs COX-1) while all sample drugs reach every target
2. Multi-objective: Seven Pareto-optimal drug pathways span efficacy 40–98%, toxicity 3–70%, cost $5–$200, and onset 1.0–7.5 h
3. Performance: DMY achieves ≈4.8× speedup over Dijkstra at n=5000 for sparse graphs (k = ⌈n^{1/3}⌉)

Reproducibility

All scripts accept a deterministic seed via either OPTIM_SP_SEED or a --seed= flag. Example:

OPTIM_SP_SEED=2024 julia --project=. drug_target_network.jl

If no seed is provided, the default (42) is used. Use the same flag when generating figures to ensure the plots and tables align with the narrative below.

Part 1: Single-Objective Analysis

Figure 1: Drug-Target Binding Affinity Matrix

Interpretation:

• Matrix shows normalized binding affinities (0=no binding, 1=perfect binding)
• Celecoxib: Strong COX-2 (0.95), weak COX-1 (0.05) → Selective inhibitor
• Aspirin: Strong COX-1 (0.85), moderate COX-2 (0.45) → Non-selective

Figure 2: COX-2/COX-1 Selectivity Profile

Clinical Significance:

Part 2: Multi-Objective Pareto Front Analysis

The Challenge

Real-world drug selection involves multiple competing objectives:

• Efficacy: How well does it work?
• Toxicity: What are the side effects?
• Cost: Can patients afford it?
• Time: How quickly does it act?

Figure 3: 2D Pareto Front Projections

Four critical trade-offs visualized:

1. Efficacy vs Toxicity: Higher efficacy drugs have more side effects
2. Efficacy vs Cost: Better drugs cost more
3. Toxicity vs Cost: Safer drugs are expensive
4. Time vs Efficacy: Fast-acting drugs may be less effective

Figure 4: 3D Pareto Front Visualization

3D Trade-off Space: This plot shows the three most critical objectives simultaneously:

• X-axis (Efficacy): Treatment effectiveness (0–100 %)
• Y-axis (Toxicity): Side effect severity (0–100 %)
• Z-axis (Cost): Price in dollars ($5–$200 across the frontier)

Legend highlights:

• Red sphere — “Max Efficacy”: Morphine-like → MOR (Solution 5) delivers ~98 % efficacy in 1 h at the cost of high toxicity (70 %) and $50.
• Green sphere — “Min Toxicity”: Ibuprofen-like → COX-2 (Solution 3) keeps toxicity at 10 % with moderate efficacy (~60 %) and $15 cost.
• Orange sphere — “Min Cost”: Aspirin-like → COX-1 (Solution 1) is the $5 budget choice with 85 % efficacy and 30 % toxicity.
• Purple hexagon — “Knee Point”: Morphine-like → COX-1 (Solution 4) marks the steepest trade-off change (95 % efficacy, 60 % toxicity, $50).

The remaining Pareto solutions (grey) illustrate the continuous trade-offs between these extremes.

Representative Pareto-Optimal Solutions

High-cost biologic options (Solutions 6–7) reduce toxicity to ≤5% while maintaining moderate efficacy (40–45%), but require $200 outlay and a 6.5–7.5 h onset.

How to Select from Pareto Front

Method 1: Weighted Sum Approach

Because this problem mixes maximize (efficacy) and minimize (toxicity/cost/time) objectives, a direct weighted sum requires transforming the maximize objectives into costs (e.g., use 1 - efficacy). The example scripts keep this method disabled by default to avoid misleading scoring—convert objectives first if you need a scalar ranking.

Method 2: Constraint-Based Selection

Set hard limits on certain objectives:

• Toxicity ≤ 30% → Aspirin-like → COX-1 (Solution 1) is the lone candidate
• Cost ≤ $20 → Aspirin-like (Solution 1), Ibuprofen-like COX-1/COX-2 (Solutions 2–3)
• Both constraints → Ibuprofen-like COX-1/COX-2 trade a little efficacy for safety

Method 3: Knee Point Selection

The current knee point is the Morphine-like → MOR pathway (Solution 5):

• Maximum efficacy (≈98%) with 1 h onset
• Accept high toxicity (70%) and higher cost ($50)
• Suitable when rapid, potent analgesia outweighs side-effect risk

Part 3: Algorithm Performance

Figure 5: Algorithm Performance Benchmark

Benchmark results from benchmark_results.txt with k = ⌈n^{1/3}⌉:

Key Insights:

• Small graphs (n < 1,000): Dijkstra is faster (DMY at 0.3×–0.4× speed)
• Crossover around n ≈ 2,000 vertices for sparse graphs
• Large sparse graphs (n ≥ 2,000): DMY delivers ≈1.8×–4.8× speedups
• Results match the canonical benchmark_results.txt file

Key Takeaways

Single vs Multi-Objective

• Single-objective: One "best" path (e.g., Celecoxib for COX-2 selectivity)
• Multi-objective: Seven non-dominated solutions on the Pareto front
• Real-world: Multi-objective reflects clinical reality better

Algorithm Performance

• Small graphs (n < 1,000): Dijkstra is faster (DMY at 0.3×–0.4× speed)
• Large sparse graphs (n ≥ 2,000): DMY delivers ≈1.8×–4.8× speedups
• Sparse networks: DMY's sweet spot (k = ⌈n^{1/3}⌉)

Clinical Impact

• No universal "best" drug: Context determines optimal choice
• Trade-offs are explicit: Pareto front visualizes all options
• Personalized medicine enabled: Match solution to patient

Reproducibility

Generate all figures:

julia --project=. generate_figures.jl

Run complete analysis:

julia --project=. drug_target_network.jl

References

1. Duan, R., Mao, J., & Yin, Q. (2025). "Breaking the Sorting Barrier for Directed SSSP". STOC 2025.
2. Multi-objective optimization: Ehrgott, M. (2005). "Multicriteria Optimization". Springer.
3. Drug data: ChEMBL and DrugBank databases.

Dashboard generated using DMYShortestPath.jl - Implementing the breakthrough DMY algorithm with multi-objective extensions