Nature provides a wealth of bioactive compounds with a wide range of potential applications. However, only about 1% of these compounds have been identified.

LINNA is our Artificial Intelligence platform, which combines multiple predictive models with a primary goal: to uncover novel, untapped bioactive compounds hidden in nature and to identify potential applications for known compounds at a speed and precision beyond human capacity.

Learn more about LINNA®'s analytical capabilities.

In recent years, new and emerging learning methodologies call for algorithms to focus on non-obvious aspects. For instance, in language or image models, techniques like hiding a portion of the picture or sentence allow the model to continue making predictions even when the input is erroneous or missing, forcing it to focus on more subtle aspects that may otherwise go unnoticed. Reinforcement learning is used in the video game industry as well. In this type of learning, the algorithm explores strategies and receives positive or negative feedback based on their success or failure. It discovers strategies that a human would never be able to uncover by thoroughly evaluating the breadth of motions and plays.

LINNA employs several of these advanced learning methods to better understand the natural world. It can, for example, predict the intricate responses of protein interaction networks, as well as the regulatory mechanisms and the impact of minor disturbances on these very sensitive networks. Thanks to the vast amount of data, LINNA learns to draw conclusions based on metabolite properties or genomic context, rather than relying solely on the stricter and more descriptive methods employed by human scientists. Due to its capacity to draw conclusions, LINNA is able to analyze millions of candidates and compare them to multiple interactome regions of the model organism. It reveals less obvious interactions and responses that may be challenging for humans to grasp.

Natural Intelligence

We begin by tapping into Natural Intelligence, leveraging the wisdom accumulated over millennia of evolution. We meticulously study the intricacies of nature’s mechanisms, unveiling its hidden compounds and harnessing its richesthrough biotechnological processes such as pyrolysis, fermentation, and extraction.

Botany, Microbiology, and Microalgae provide an infinite number of potential compounds, which we utilise to develop productive solutions for the future.



LINNA was conceived to address the demands of the agricultural sector. However, given its objectives and immense potential, its application may grow beyond agriculture and into areas such as agribusiness, dermo-cosmetics, and the pharmaceutical industry.

LINNA, proposes to address all of these challenges by discovering new applications for the 1% of known bioactive compounds as well as searching for novel bioactive compounds among the unknown 99%.

Learn more about LINNA®'s fundamental purpose.

With its expansive potential, LINNA is empowered to explore synergies within bioactive compounds, unveil novel mechanisms of action, predict potential toxicity, and uncover non-evident realities—those intricate patterns in nature that ordinary human observation fails to reveal.  

Our unwavering efforts, driven by ambition, propel us forward. Our goal is to comprehensively analyze between 3 and 5 million natural compounds by 2030. Our goal is to obtain innovative solutions, ranging from a novel herbicide to replace the use of glyphosate to identifying viable alternatives for the chemical preservatives used within the food industry.

How Does LINNA® work?

In order to predict the activity of natural compounds and, consequently, prescribe new ones to address a specific need, LINNA® comprises multiple prediction models that are seamlessly integrated with one another.

To achieve this, we employ a variety of extraction and transformation techniques to access the complete chemical spectrum found within various natural raw materials. Through this approach, we can pinpoint more than 10 complex candidates for each analyzed raw material. Each of these candidates, consisting of hundreds of unidentified compounds, triggers a specific response when administered to an organism. By understanding these responses, through diverse methodologies based in systems biology and Big Data, LINNA® has the capacity to predict an organism’s behavior concerning various candidates, all without the need for conventional testing.

Learn more about how LINNA works.

1. DataDiscovery

2. DataShine

3. LINNABrain

4. LINNALearn

1. DataDiscovery.
Accesses internal and external data.

  1. We analyze 3 to 5 million natural compounds that originate from our own candidate amplification and generation node. This collaborative approach with science makes us significantly more competitive. Would you like to know how we do it?
  2. Over 500 TB of transcriptome data
  3. Over 130 million images taken with RGB, PAM sensors and hyperspectral cameras.

All of this information, together with the external data, will allow us to discover previously unknown compounds and mechanisms of action while minimizing the requirement for experimental testing by up to 95%.

Internal Data

A proprietary universe of data that makes our platform unique

To achieve high efficacy, we use data generated internally in our value-search ecosystem, such as molecular profiles, cause-effect relationships between natural molecules and mechanisms of action against a target, and the knowhow on relationships and synergies between natural molecules that we have been developing for 15 years

External Data

External data sources that help us provide value and identify relevant patterns

For instance, we use data on historical temperature data and its correlation with crop performance, crop-specific farm monitoring using IoT devices, external databases for metabarcoding, and metagenomics to explore chemical compound structure-activity relationships.

2. DataShine.

It highlights data through a range of machine learning models, including regression, dimension reduction, and unsupervised clustering, to execute diverse tasks. These tasks encompass everything from exploring a candidate’s chemical space to characterizing the transcriptome response to its application. Additionally, it employs deep learning models such as graph neural networks and encoder-decoder models to improve biological activity predictions.

All these models are integrated to create three modules, transforming LINNA into an unrivaled, potent, and adaptable tool. Utilizing these modules has enabled us to uncover 30 candidates possessing formidable herbicidal potential—discoveries that would have eluded us relying solely on human reasoning.

LINNA Modules

LINNA comprises three core modules – expression, dimensionality reduction, and phenotype – seamlessly integrating multiple models into a unified pipeline which aims not only to forecast a phenotype from a complex candidate but also to propose novel bioactive compounds necessary for attaining a desired phenotype or biological activity:

1. Expression

LINNA®’s “expression” module predicts changes in expression in an organism of interest based on candidate information acquired through high-resolution spectrometry techniques. All of this information is cross-referenced with the LINNA® databases to identify the known compounds for which a fingerprint is calculated. This module also includes neural networks for estimating the structure and fingerprint of unknown compounds.
The candidate’s data is utilized to predict the expression profile that the candidate as a whole and each of its constituent compounds would generate in the model organism.

2. Dimensional Reduction

The expression profile of an organism in response to a candidate, typically consisting of hundreds or even thousands of compounds, necessitates the identification of numerous differentially expressed genes. Predicting a phenotype would become virtually impossible if we treated each gene as an independent entity. Given this intricate challenge, LINNA®’s second module examines the anticipated expression profile generated by the first module. It identifies the primary biological processes that have undergone significant changes and assesses the relevance of the numerous genes involved in these processes. This module streamlines the vast array of differentially expressed genes into a concise set of several dozen altered biological processes, encompassing all related genes and their interactions within the biological context. This approach significantly reduces the computational complexity of the third module, which is responsible for predicting a given phenotype.


Dimensional reduction obtained by LINNA® module 2 following the prediction of over 5000 differentially expressed genes. This representation highlights the various modified biological processes, their constituent genes, associated gene ontology, and the interactions between the proteins encoded by these genes.

3. Phenotype

The third module predicts a wide variety of phenotypic characteristics using the biological processes affected by the research candidate as input. The mathematical models included in this module can not only predict these parameters, but also assign the relative importance that each of them will have in the final phenotype.

Relative importance allocation for phenotypic characteristics predicted by LINNA® module 3. 

All of these parameters, together with their relative importance, allow us to categorize the research candidate (or any of its bioactive compounds) not only by biological activity or predicted phenotype, but also by mechanism of action within each biological activity. This grouping is compared to the prior module’s expression data, allowing for complex correlations between transcriptome and phenotypic response in a reduced dimensional space.

Grouping of five candidates examined in biological activities and mechanisms of action by LINNA® module 3 based on phenotypic parameter predictions and relative importance.

3. LINNABrain.

Our rationale is based on business intelligence, commercial strategies, the real specific needs of the industry, market demand, and other factors, all while providing a 360-degree solution that addresses a real need in the industry.

4. LINNALearn.

It gains knowledge from LINNABrain’s decisions and improves results by increasing its prescription capacity.

LINNA®’s transversality

LINNA was created to address agricultural challenges by developing bioherbicides, biostimulants, probiotics, biofungicides, bioinsecticides, and seed treatment; however, its immense potential makes it a transversal platform capable of addressing challenges in animal production, fish farming, and the cosmetic and pharmaceutical industries.


Anticipating the future

Our target for 2030

 is to analyze 3-5 million natural compounds with a prediction accuracy surpassing 95%.

To benefit from non-evident reality, which is defined as AI’s capacity to find hidden relationships that are undetectable through mere observation.

To manage and analyze a huge amount of data thanks to our extensive experience with natural compounds acquired over the last 15 years.

To explore transversality, as the laws of biology are universally applicable to all organisms.

To study the 99% of unknown natural compounds.

To find new mechanisms of action associated with natural molecules.

To predict possible toxicities in our products.

To look for synergies between natural molecules.

To find new uses for the 1% of already known natural compounds.

They have seen the future

Kimon Theophanopoulus

CEO & Founder Teofert

Alexandre Macedo

Vice President of Biostimulants at Yara Europe

Ernesto Baltodano

General Manager at CISA Agro