Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Retrieval-Retro: Retrieval-based Inorganic Retrosynthesis with Expert Knowledge

Heewoong Noh, Namkyeong Lee, Gyoung S. Na, Chanyoung Park

TL;DR

Retrieval-Retro addresses the challenge of inorganic retrosynthesis by retrieving reference materials with MPC and NRE retrievers and then implicitly extracting precursor information through attention-based fusion, guided by thermodynamic domain knowledge. The method combines composition-graph representations with self- and cross-attention to fuse target and retrieved references, enabling novel synthesis recipes beyond known precedents. Empirical results show improved precursor prediction, especially under realistic year-split evaluations, with ablations highlighting the complementary roles of MPC and NRE and the value of implicit extraction. This approach advances automated planning for inorganic material synthesis and has potential to accelerate materials discovery by broadening feasible synthesis pathways.

Abstract

While inorganic retrosynthesis planning is essential in the field of chemical science, the application of machine learning in this area has been notably less explored compared to organic retrosynthesis planning. In this paper, we propose Retrieval-Retro for inorganic retrosynthesis planning, which implicitly extracts the precursor information of reference materials that are retrieved from the knowledge base regarding domain expertise in the field. Specifically, instead of directly employing the precursor information of reference materials, we propose implicitly extracting it with various attention layers, which enables the model to learn novel synthesis recipes more effectively. Moreover, during retrieval, we consider the thermodynamic relationship between target material and precursors, which is essential domain expertise in identifying the most probable precursor set among various options. Extensive experiments demonstrate the superiority of Retrieval-Retro in retrosynthesis planning, especially in discovering novel synthesis recipes, which is crucial for materials discovery. The source code for Retrieval-Retro is available at https://github.com/HeewoongNoh/Retrieval-Retro.

Retrieval-Retro: Retrieval-based Inorganic Retrosynthesis with Expert Knowledge

TL;DR

Retrieval-Retro addresses the challenge of inorganic retrosynthesis by retrieving reference materials with MPC and NRE retrievers and then implicitly extracting precursor information through attention-based fusion, guided by thermodynamic domain knowledge. The method combines composition-graph representations with self- and cross-attention to fuse target and retrieved references, enabling novel synthesis recipes beyond known precedents. Empirical results show improved precursor prediction, especially under realistic year-split evaluations, with ablations highlighting the complementary roles of MPC and NRE and the value of implicit extraction. This approach advances automated planning for inorganic material synthesis and has potential to accelerate materials discovery by broadening feasible synthesis pathways.

Abstract

While inorganic retrosynthesis planning is essential in the field of chemical science, the application of machine learning in this area has been notably less explored compared to organic retrosynthesis planning. In this paper, we propose Retrieval-Retro for inorganic retrosynthesis planning, which implicitly extracts the precursor information of reference materials that are retrieved from the knowledge base regarding domain expertise in the field. Specifically, instead of directly employing the precursor information of reference materials, we propose implicitly extracting it with various attention layers, which enables the model to learn novel synthesis recipes more effectively. Moreover, during retrieval, we consider the thermodynamic relationship between target material and precursors, which is essential domain expertise in identifying the most probable precursor set among various options. Extensive experiments demonstrate the superiority of Retrieval-Retro in retrosynthesis planning, especially in discovering novel synthesis recipes, which is crucial for materials discovery. The source code for Retrieval-Retro is available at https://github.com/HeewoongNoh/Retrieval-Retro.

Paper Structure

This paper contains 31 sections, 6 equations, 5 figures, 9 tables, 3 algorithms.

Table of Contents

  1. Introduction
  2. Preliminaries
  3. Problem setup
  4. Composition Graph Encoder
  5. Proposed Method: Retrieval-Retro
  6. Reference Material Retrieval
  7. Implicit Precursor Extraction
  8. Model Training
  9. Experiments
  10. Experimental Setup
  11. Empirical Results
  12. Model Analysis
  13. Related Works
  14. Machine Learning for Inorganic Retrosynthesis
  15. Composition-based representation learning for Inorganic materials
  16. ...and 16 more sections

Figures (5)

  • Figure 1: An example case where (a) the target material shares a subset of precursors with reference material, and (b) the target material has an entirely new set of precursors, without sharing any subset of precursors with reference material. (c) The proportion of subset cases and new cases among the materials newly synthesized from 2017 to 2020. (d) The target material is more likely to be synthesized using the precursor set that exhibits a more negative driving force.
  • Figure 2: (a) Training process of the Masked Precursor Completion (MPC) retriever. (b) Training process of the Neural Reaction Energy (NRE) retriever.
  • Figure 3: The overall framework of Retrieval-Retro.
  • Figure 4: Sensitivity Analysis results. KB refers to the knowledge base.
  • Figure 5: Performance Improvements across Various GNN Backbones