How AI is quietly transforming everyday chemistry labs

Saturday 07 - 10:20

By: Dakir Madiha

How AI is quietly transforming everyday chemistry labs

In chemistry, artificial intelligence is increasingly used as a practical tool to sift through huge troves of experimental and simulation data, predict the properties of thousands of candidate molecules or materials in one shot, and point chemists toward the few that are worth testing in the lab. François‑Xavier Coudert, a CNRS theoretical chemist who works on nanoporous materials, explains that AI is already helping researchers move beyond slow, trial‑and‑error methods and even beyond their own “chemical intuition” when they search for new compounds or re‑use old ones in unexpected ways.

One of the main goals is to identify better molecules or materials for a specific application, such as capturing carbon dioxide from exhaust gases, by scanning databases of known compounds to see whether any might work well in conditions they were never originally designed for. Traditionally, a team would develop a material with one purpose in mind, publish the result if it worked, and move on if it failed, leaving many promising candidates effectively “in a drawer,” whereas AI allows researchers to revisit those archives systematically and rank which structures deserve a second look, in a process similar to drug repurposing where an existing medicine is reassigned to a new disease.

Coudert’s group and others apply machine learning to predict how porous materials such as metal–organic frameworks will adsorb gases or separate difficult mixtures like CO₂ and acetylene, using structural descriptors, pore geometry and chemical information to train models that can screen thousands of structures far faster than classical simulations alone. In recent studies, combining molecular simulations with machine learning has uncovered metal–organic frameworks that outperform previous best‑in‑class materials for CO₂ capture or gas separation, demonstrating that AI‑guided searches can navigate vast chemical spaces and bring genuinely new candidates to light that might never have been chosen by intuition alone.

Beyond discovery, AI also helps improve the computational tools chemists already rely on: machine‑learning‑based interatomic potentials, for example, can reproduce quantum‑level interactions in complex frameworks at a fraction of the cost, making it possible to simulate flexible or disordered materials in more realistic conditions. Coudert notes that many of these models are designed to be explainable rather than black boxes, so that they highlight which structural features or chemical motifs drive a property and thus feed back into human understanding instead of just spitting out numbers.

A growing frontier lies in mining the “dark data” hidden in laboratory notebooks, legacy files and poorly standardized records, where decades of synthesis attempts and partial results remain difficult to exploit because chemists rarely follow strict naming conventions or metadata rules. AI‑driven text and data extraction could eventually turn that noisy archive into structured datasets for training new models, but Coudert warns that this will require serious work on data curation and reproducibility so that future algorithms do not amplify errors or biases buried in past experiments.

For now, he argues, the real impact of AI in chemistry is not about replacing researchers with robots, but about giving human teams faster, more systematic ways to explore ideas, test hypotheses and spot overlooked materials in a landscape where the number of possible molecules is effectively infinite. An AI model might flag a handful of promising structures for CO₂ capture, yet it remains up to chemists to judge whether those candidates can actually be synthesized, whether they are stable, and whether they make sense in industrial processes, keeping human expertise in the loop even as algorithms become central to everyday lab work.