[Paper Review] Molecular Inverse-Design Platform for Material Industries

Seiji Takeda, Toshiyuki Hama+18 more|arXiv (Cornell University)|Apr 24, 2020

Machine Learning in Materials Science2 references4 citations

TL;DR

This paper presents an AI-driven molecular inverse-design platform tailored for industrial material development, combining substructure-based feature encoding with molecular graph generation to enable rapid, interpretable, and customizable design of novel molecules. Deployed on cloud infrastructure and used by five partner companies, the system accelerated molecular design speed by over 10× compared to expert chemists, with significantly increased structural diversity while maintaining chemical realism, as demonstrated in sugar and dye molecule design.

ABSTRACT

The discovery of new materials has been the essential force which brings a discontinuous improvement to industrial products' performance. However, the extra-vast combinatorial design space of material structures exceeds human experts' capability to explore all, thereby hampering material development. In this paper, we present a material industry-oriented web platform of an AI-driven molecular inverse-design system, which automatically designs brand new molecular structures rapidly and diversely. Different from existing inverse-design solutions, in this system, the combination of substructure-based feature encoding and molecular graph generation algorithms allows a user to gain high-speed, interpretable, and customizable design process. Also, a hierarchical data structure and user-oriented UI provide a flexible and intuitive workflow. The system is deployed on IBM's and our client's cloud servers and has been used by 5 partner companies. To illustrate actual industrial use cases, we exhibit inverse-design of sugar and dye molecules, that were carried out by experimental chemists in those client companies. Compared to general human chemist's standard performance, the molecular design speed was accelerated more than 10 times, and greatly increased variety was observed in the inverse-designed molecules without loss of chemical realism.

Motivation & Objective

• To address the challenge of exploring vast combinatorial molecular design spaces beyond human experts' capacity in material industries.
• To develop a scalable, user-friendly web platform that enables industrial chemists to rapidly generate novel, diverse, and chemically valid molecular structures.
• To integrate interpretable AI techniques with practical industrial workflows for real-world material discovery.
• To demonstrate the platform’s effectiveness through real-world use cases in dye and sugar molecule design with industrial partners.

Proposed method

• The platform employs substructure-based feature encoding to represent molecular structures in a way that preserves chemical interpretability.
• It uses a molecular graph generation algorithm to synthesize new molecular structures based on user-defined target properties.
• A hierarchical data structure organizes molecular data and design workflows to support flexible and scalable processing.
• The system features a user-oriented graphical interface to enable intuitive interaction and customization by non-expert users.
• The platform is deployed on IBM and client cloud servers, ensuring secure, scalable, and accessible industrial use.
• The design process is guided by a combination of deep generative models and chemical rules to ensure synthetic feasibility and realism.

Experimental results

Research questions

• RQ1 Can an AI-driven inverse-design platform significantly accelerate molecular design in industrial settings while maintaining chemical validity?
• RQ2 How can interpretability and customization be preserved in high-throughput molecular generation for industrial users?
• RQ3 To what extent can such a system increase structural diversity compared to traditional expert-driven approaches?
• RQ4 Can the platform be effectively deployed and adopted in real industrial R&D workflows?

Key findings

• The platform achieved a molecular design speed more than 10 times faster than standard human chemist performance in industrial use cases.
• The inverse-designed molecules exhibited significantly greater structural diversity compared to conventional approaches.
• All generated molecules maintained high chemical realism, as validated by expert chemists and structural analysis.
• The system was successfully deployed and used by five industrial partner companies in real-world R&D workflows.
• The integration of substructure encoding and graph generation enabled both speed and interpretability in molecular design.
• User feedback confirmed the platform’s intuitive interface and practical value in accelerating industrial material development.