Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Surrogate-Based Black-Box Optimization Method for Costly Molecular Properties

Jules Leguy, Thomas Cauchy, Beatrice Duval, Benoit Da Mota

TL;DR

This work tackles the challenge of optimizing costly molecular properties by integrating surrogate-based black-box optimization with Gaussian Process Regression and an evolutionary search guided by the Expected Improvement. The framework efficiently explores molecular graphs by learning a surrogate from a minimal initial dataset and iteratively proposing candidates via EI optimization, re-training as expensive evaluations are performed. Two descriptors are studied: MBTR, a geometry-based representation, and a fast graph-based shingles descriptor, with GPR models evaluated for data efficiency and runtime. Empirical results on HOMO energy optimization demonstrate that the surrogate-based approach can outperform a baseline evolutionary algorithm by up to several-fold in both objective evaluations and runtime, suggesting practical scalability to larger chemical spaces and broader property targets.

Abstract

AI-assisted molecular optimization is a very active research field as it is expected to provide the next-generation drugs and molecular materials. An important difficulty is that the properties to be optimized rely on costly evaluations. Machine learning methods are investigated with success to predict these properties, but show generalization issues on less known areas of the chemical space. We propose here a surrogate-based black box optimization method, to tackle jointly the optimization and machine learning problems. It consists in optimizing the expected improvement of the surrogate of a molecular property using an evolutionary algorithm. The surrogate is defined as a Gaussian Process Regression (GPR) model, learned on a relevant area of the search space with respect to the property to be optimized. We show that our approach can successfully optimize a costly property of interest much faster than a purely metaheuristic approach.

Surrogate-Based Black-Box Optimization Method for Costly Molecular Properties

TL;DR

This work tackles the challenge of optimizing costly molecular properties by integrating surrogate-based black-box optimization with Gaussian Process Regression and an evolutionary search guided by the Expected Improvement. The framework efficiently explores molecular graphs by learning a surrogate from a minimal initial dataset and iteratively proposing candidates via EI optimization, re-training as expensive evaluations are performed. Two descriptors are studied: MBTR, a geometry-based representation, and a fast graph-based shingles descriptor, with GPR models evaluated for data efficiency and runtime. Empirical results on HOMO energy optimization demonstrate that the surrogate-based approach can outperform a baseline evolutionary algorithm by up to several-fold in both objective evaluations and runtime, suggesting practical scalability to larger chemical spaces and broader property targets.

Abstract

AI-assisted molecular optimization is a very active research field as it is expected to provide the next-generation drugs and molecular materials. An important difficulty is that the properties to be optimized rely on costly evaluations. Machine learning methods are investigated with success to predict these properties, but show generalization issues on less known areas of the chemical space. We propose here a surrogate-based black box optimization method, to tackle jointly the optimization and machine learning problems. It consists in optimizing the expected improvement of the surrogate of a molecular property using an evolutionary algorithm. The surrogate is defined as a Gaussian Process Regression (GPR) model, learned on a relevant area of the search space with respect to the property to be optimized. We show that our approach can successfully optimize a costly property of interest much faster than a purely metaheuristic approach.

Paper Structure

This paper contains 24 sections, 5 equations, 3 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Preliminaries
  3. Surrogate-based black-box optimization
  4. Design of experiments
  5. Surrogate learning
  6. Merit function
  7. Obtaining candidate solutions
  8. Representing molecules
  9. Geometrical descriptors
  10. Graph-based descriptors
  11. Methods
  12. Black-box optimization of molecular graphs
  13. Learning a surrogate model
  14. Model
  15. Kernels
  16. ...and 9 more sections

Figures (3)

  • Figure 1: Surrogate-based black-box optimization of molecular graphs
  • Figure 2: Learning curves of HOMO energies, showing out-of-sample prediction error (mean absolute error) depending on the number of training molecules drawn at random from QM9 dataset for the two proposed surrogate models.
  • Figure 3: Empirical cumulative distribution functions of the number of DFT calls (top) and CPU time (bottom) for the two proposed surrogate model and the evolutionary algorithm used as a baseline.