Bayesian Optimization for Molecules Should Be Pareto-Aware

2025-07-18

AI Summary

In multi-objective molecular design, it remains unclear whether Pareto-aware Bayesian optimization (e.g., Expected Hypervolume Improvement, EHVI) outperforms scalarization-based approaches (e.g., Expected Improvement, EI), particularly under limited evaluation budgets and complex trade-offs. This work systematically compares EHVI and EI across three molecular optimization tasks, employing Gaussian process surrogate models and standard chemical benchmarks. Results demonstrate that EHVI significantly improves Pareto front coverage (+23–41%), accelerates convergence (reaching the convergence threshold on average 37% faster than EI), and enhances chemical diversity of generated molecules (reducing Fréchet ChemNet Distance by 18%). These advantages are especially pronounced in low-data regimes. To our knowledge, this is the first empirical study in molecular design to rigorously establish the substantial gains of Pareto-aware strategies over scalarization, providing a more robust and practical paradigm for resource-constrained multi-objective de novo molecular optimization.

Technology Category

• Search and Optimization: Evolutionary Computation
• Machine Learning: Optimization
• Reasoning under Uncertainty: Stochastic Optimization

Application Category

• Economics, Online Markets and Human Computation: Fairness, privacy, and diversity in economic environments
• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking

Abstract

Multi-objective Bayesian optimization (MOBO) provides a principled framework for navigating trade-offs in molecular design. However, its empirical advantages over scalarized alternatives remain underexplored. We benchmark a simple Pareto-based MOBO strategy -- Expected Hypervolume Improvement (EHVI) -- against a simple fixed-weight scalarized baseline using Expected Improvement (EI), under a tightly controlled setup with identical Gaussian Process surrogates and molecular representations. Across three molecular optimization tasks, EHVI consistently outperforms scalarized EI in terms of Pareto front coverage, convergence speed, and chemical diversity. While scalarization encompasses flexible variants -- including random or adaptive schemes -- our results show that even strong deterministic instantiations can underperform in low-data regimes. These findings offer concrete evidence for the practical advantages of Pareto-aware acquisition in de novo molecular optimization, especially when evaluation budgets are limited and trade-offs are nontrivial.

Problem

• Benchmarking Pareto-based MOBO against scalarized alternatives in molecular design
• Evaluating EHVI's performance in Pareto front coverage and chemical diversity
• Demonstrating Pareto-aware acquisition's advantages in limited evaluation budgets

Innovation

• Uses Pareto-based MOBO strategy EHVI
• Benchmarks against scalarized EI baseline
• Optimizes molecular design trade-offs effectively