How AI Could Transform the Extraction of Valuable Plant Compounds

The earliest medicine known to man was plants with active compounds that help alleviate pain, reduce fever, fight diseases, etc. To this day, pharmacobotany is a very important field of modern medicine, and rare molecules are regularly found in common and less common plants to develop new therapies.

One issue is that these compounds can be medically powerful but are present in small amounts in a plant, or the plant is difficult to grow. In both cases, this means that the resulting medicine can be very expensive, especially when there is no viable way to chemically synthesize a compound of interest.

As such, any improvement that optimizes the extraction of active molecules from plants can have massive economic and medical implications.

A team of Iranian researchers at the Islamic Azad University, the Tarbiat Modares University, and the Isfahan University of Technology has used AI to optimize the extraction of valuable compounds with supercritical CO2. They published their findings in CO2 Utilization¹, under the title “Sustainable CO₂-based process intensification for the extraction of Nepeta crispa essential oil using ultrasound-assisted supercritical CO₂: Optimization and machine learning modeling”.

Extracting Plant Pharmacopeia

A wide range of herbs and aromatic plants, owing to their valuable properties and benefits, which are largely attributed to their essential oils, have found extensive applications across pharmaceutical, food, cosmetic, paint, and chemical industries. Essential oils are volatile constituents composed of complex chemical mixtures containing various types of molecules.

Traditional techniques for the extraction of essential oils from plant materials, such as hydrodistillation, steam distillation, and organic solvent–based extraction, suffer from several inherent limitations:

• Potential contamination of the extracted products resulting from the use of organic solvents.
• The requirement for subsequent solvent removal.
• Solvent toxicity and flammability.
• High operational costs and significant energy input to achieve acceptable yields

In recent years, ultrasound-assisted extraction (UAE) has caught a lot of attention for its superior results (yields, solvent usage, speed) through acoustic cavitation, which disrupts plant cell walls and facilitates the release of bioactive compounds and essential oils.

Meanwhile, supercritical fluids (SCFs), with particular attention to supercritical carbon dioxide (SC-CO₂), have been increasingly recognized as an effective alternative to conventional extraction media.

A supercritical fluid is a chemical that is pressurized above a point where distinct liquid and gas phases disappear. So it is technically a gas, but mostly acts as a liquid.

Among the advantages of CO2 is its non-explosiveness, non-toxicity, low cost, non-flammability, and non-corrosiveness. It also preserves the natural integrity of bioactive compounds and can provide high selectivity for extraction through precise control of pressure and temperature.

Optimizing Supercritical CO2-Based Extraction

The actual efficiency of extraction with supercritical CO2 depends on a wide array of factors, including pressure, extraction time, cosolvent volume, temperature, particle size, and the flow rate of CO₂. As such, finding the ideal configuration for a given compound in a particular plant can be rather difficult and has, for now, been as much an art as a science.

Statistical and mathematical models can help, but they are complex to design in practice.

To optimize such a method, the researchers used a machine learning algorithm called ANFIS (Adaptive Neuro-Fuzzy Inference System) and the Particle Swarm Optimization (PSO) algorithm.

Perfecting Extraction with CO2 & AI

Plant Selection & Tests

Nepeta crispa was the plant chosen as a test by the researcher. Various Nepeta species have a broad spectrum of biological activities, such as antibacterial, antiviral, and antifungal effects. In traditional medicine, N. crispa has been utilized as an antitussive, antispasmodic, diuretic, diaphoretic, anti-asthmatic, emmenagogue, and sedative remedy.

Source: WorldPlant

The researchers compare the extraction of active compounds from this plant by 3 different methods:

• SC-CO₂ under optimal conditions
• Hydrodistillation
• Ultrasonic-assisted extraction

19 chemical compounds were identified, constituting 97-99% of the total essential oils extracted, with supercritical CO2 the most productive method, with an almost perfect extraction yield. The overall composition and qualitative profile of the essential oils were comparable across all techniques.

Source: CO2 Utilization

Building & Testing The AI Model

Various configurations involving different types and numbers of input variables (temperature, pressure, mass flow rate, and co-solvent volume), with ANFIS employing backpropagation and hybrid algorithms to optimize neuro-fuzzy systems.

Source: CO2 Utilization

Overall, temperature did not impact extraction much, while pressure and CO2 flow rate were more important, and the addition of a co-solvent (ethanol) was also important.

To check if the model predicting supercritical CO2 extraction was accurate, the scientists tested it against multiple experiments. The yield extraction prediction was, in each case, almost perfectly matching the experimental results.

The optimal operating conditions for the process were found to be the pressure of 34 MPa, the temperature of 112 °F / 44.8 °C, the mass flow rate of 1.1 g/min, and the co-solvent volume of 370 µL.

The Future Of Resource Utilization

This study is one among many examples of how even relatively simple AI systems can be used for niche applications with important applications. For example, in this case, the optimization of the extraction of medical molecules from plants, potentially leading to new or cheaper therapies, while also reducing the ecological impact of resource extractions by utilizing renewable and non-polluting resources.

As these systems scale, the economic upside could extend beyond botanical extraction itself, creating opportunities for industrial gas suppliers, process-engineering firms, automation providers, and AI companies that help manufacturers cut costs and improve throughput.

Supercritical CO2 is also likely to be used more and more, for example, the groundbreaking “Chaotan One” supercritical CO2 turbine testing in China, replacing traditional supercritical steam with CO2 to radically improve the energy production of the turbine.

Investing in CO2 Industrial Applications

Linde Plc

Linde is one of the world’s largest industrial gas companies, alongside competitors like Air Products and Chemicals (APD )  (follow the link for our report on that company), with as much as $34B in sales in 2025. It was founded in Germany in 1879 and merged with another giant of the sector, Praxair, in 2018.

Today, Linde is active in the production of both pure gases extracted from the atmosphere, including rare gases, as well as gases produced by industries like LNG (helium), hydrocarbons (ethylene, propane, CO, CO2), as well as hydrogen production.

Source: Linde

Sales are dominated geographically by the Americas (45% of sales) and the EMEA region (25%). As almost every industrial process requires these gases, the customer base of Linde is very diversified, with the largest industries served being chemicals & energy (21% of sales), manufacturing (21%), healthcare (17%), and mining (13%).

Depending on the gas and the customer, Linde produces either directly on the industrial site using these gases or in centralized facilities and then transports them to the final user.

Source: Linde

Industrial gases is a sector where scale is a durable competitive advantage, as it helps the company accumulate a reputation of reliability (quick deliveries, reactive supply) and reduce capital cost and R&D expenses across a larger production volume and extensive distribution network.

Both green hydrogen and carbon capture are opportunities for more growth for the company than “just” the global industrial production growth, as they give it new venues where it can expand in the energy sector.

Source: Linde

Latest Linde (LIN) Stock News and Developments

Study Referenced

^{1. Mitra Amani, Nedasadat Saadati Ardestani, Seyyed Mohammad Ghoreishi. Sustainable CO₂-based process intensification for the extraction of Nepeta crispa essential oil using ultrasound-assisted supercritical CO₂: Optimization and machine learning modeling. CO2 Utilization. 10 July 2026. Article: 103519. Volume: Volume 110. 10.1016/j.jcou.2026.103519}