The arrival of AI technologies is disrupting the pace of R&D and giving rise to new challenges.
Generative and deep-learning techniques enable the prediction, design, and optimization of materials and molecules faster, more efficiently, and often more accurately than traditional methods.
These technologies open promising horizons in numerous fields, including materials science, pharmaceuticals, energy, industrial chemistry, and more.
They bring increased efficiency, estimated at around 13 to 15% (1), which is significant. However, even now—and this will only increase in the future—the question of data influx and sorting arises. The risk of missing essential information is real for research teams and centers facing a flood of data: in industrial chemistry alone (excluding pharmaceuticals), over 500,000 articles are published each year, and between 75,000 and 120,000 patents are filed (2). Google’s DeepMind team has already produced 2.2 million new molecular structures, 400,000 of which are immediately usable (3).
The conclusion is simple: regardless of its size or the number of research projects, a company without powerful analytical tools is clearly at risk.
To better understand this observation, let’s first summarize the three stages of “classic” industrial R&D and the operations they involve. We will then examine the upheavals brought by AI and the risks they pose to unprepared actors before concluding with the essential defensive strategies to address them.
MAJOR STAGES OF “CLASSIC” INDUSTRIAL R&D:
- Idea formulation and innovation pathways:
This stage aims to generate new and relevant ideas in response to identified problems and/or technological challenges. It encompasses several approaches:
Analyzing the context and understanding scientific or industrial needs.
Exploring the state of the art to highlight gaps to be filled and innovation opportunities. In this sense, originality can be seen as the foundation of literature reviews.
Stimulating creativity and generating concepts through new hypotheses.
Precisely defining the research subject and scope.
- Idea validation through a rigorous process:
Once research ideas and directions are formulated, a rigorous evaluation and refinement process is necessary. This stage includes:
Reviewing ideas to determine their feasibility and scientific or technical relevance.
Designing experimental methodologies to test idea viability, particularly through comparative experimental plans.
Collecting structured data to assess hypothesis robustness and identify promising research avenues.
Iterating concepts based on results obtained by the researcher or team.
- Realization of selected ideas and concepts:
This stage involves:
Designing prototypes and experimental models.
Optimizing solutions to ensure optimal scalability.
Collecting detailed data during tests to evaluate and adjust prototypes.
Assessing results against initial objectives.
These three stages are often cyclical and can be repeated with adjustments and improvements based on discoveries and obtained results. Each stage may also involve interdisciplinary collaborations or inter-research team partnerships within the same organization, along with continuous reflection to ensure logical and efficient research progress.
THE IMPACT OF AI: A REVOLUTION UNDERWAY?
In recent years, AI-assisted prediction of physicochemical properties of materials and AI-driven molecular design have made significant strides. These are two constantly evolving and fascinating fields that combine artificial intelligence, chemistry, and materials science to solve complex problems. They can be summarized as follows:
- AI-assisted prediction of physicochemical properties of materials:
Broadly speaking, this involves using AI models (neural networks, support vector machines, etc.) to predict material properties based on their chemical structures. Traditionally, this required complex chemical calculations or lengthy experiments. AI reduces this need by leveraging extensive databases and machine learning algorithms. Applications include the prediction of:
Electrical or thermal conductivity of specific materials.
Thermal stability.
Mechanical properties (strength, ductility).
Optical properties (absorption, refraction).
Chemical interactions with other materials or substances.
This not only saves time but also helps identify more efficient or safer materials for specific applications, such as in electronics, energy, or biotechnology.
- AI-driven molecular design:
Generative AI, particularly models like generative adversarial networks (GANs), is used to design new molecules with predefined properties. This process involves training AI on existing molecular structures and their properties, enabling it to create innovative molecules for fields such as:
More effective drugs (especially for difficult-to-treat diseases).
Materials for more powerful and durable batteries.
Catalysts to improve specific chemical reactions.
Generative AI can explore vastly larger design spaces than humanly possible, accelerating discovery. Its role in AI-assisted R&D applies to various aspects of the process:
Computational simulations: These verify the stability and feasibility of predicted molecular structures, including energy minimization calculations and molecular dynamics tests.
Experimental validation: Predicted structures are synthesized in laboratories, and their properties are measured experimentally through techniques such as X-ray crystallography, NMR spectroscopy, and infrared spectroscopy.
Comparison with existing data: Predicted structures are compared with validated known structures to verify accuracy, using international molecular structure databases.
Expert assessment: Researchers in the field evaluate results to ensure they align with current knowledge, chemical and biological theories, and present innovation and interest.
Reproducibility and scalability: Experiments are often repeated in other laboratories to confirm results, ensure reproducibility and scalability, and verify practical applicability.
OPTIMIZE YOUR TECHNOLOGY MONITORING AND STAY COMPETITIVE!
When combining the data flood mentioned earlier, the increasing automation of many industrial R&D tasks, AI-driven property prediction, and generative molecular design, one conclusion is clear: more than ever, human intervention is necessary to distinguish real innovations, assess their relevance, and measure their value for the applicable domain.
In this context, the first two steps of the R&D process are crucial. To significantly reduce the risk of overlooking a developing innovation in your field, having an AI-powered tool for analyzing, structuring, and synthesizing scientific literature data represents a strategic asset essential to your company.
Natively designed to leverage the latest AI techniques, the CHEMY LANE® application, developed by the MAYFAIR VILLAGE® teams, was created to meet this need. Its ADVANCED version, an intuitive and user-friendly SaaS solution for industrial R&D, can serve as:
A HIGH-PERFORMANCE TECHNOLOGY MONITORING TOOL for rapid and efficient data analysis.
An INTELLIGENT RESEARCH ASSISTANT that enhances efficiency and saves time.
An “IDEA LABORATORY” accelerating your R&D by opening new research avenues and considerations from the early stages of projects.
Sign up for a free trial on our website! www.chemylane.ai
1. “Artificial Intelligence, Scientific Discovery, and Product Innovation” Aidan Toner-Rodgers† /MIT November 6, 2024 Aghion et al., 2019; Cockburn et al., 2019).
2. Our World in Data – Annual articles published in scientific and technical journals in 2024.
3. “World Intellectual Property Indicators (WIPI)” – WIPO / “Patent Landscape Reports” / WIPO – Global Patent Trends in Chemistry and Pharmaceuticals” – WIPO.