The article, led by Jose Arturo Molina-Mora of the University of Costa Rica and colleagues spanning more than a dozen institutions across the region, argues that Latin America’s problem is not a lack of talent or ambition. Researchers across the continent have already applied machine learning to an impressive range of bioscience problems. In Argentina, classification algorithms have been used for gene annotation and immune-oncology analyses; in Brazil, deep learning has detected individual Amazonian palm trees from drone imagery and machine learning has accelerated drug discovery; in Uruguay, transformer-based models have been applied to genomic prediction; in Costa Rica, clustering methods have profiled COVID-19 patients and identified core genes in the pathogen Pseudomonas aeruginosa. These studies, often built on locally generated data addressing regional needs, demonstrate that the scientific groundwork exists.

What holds the region back, the authors contend, are structural barriers identified in assessments such as the International Monetary Fund’s AI Preparedness Index. Gaps in infrastructure, legislation, human resources, and ethical frameworks persist across most Latin American countries. Computing power, cloud platforms, and reliable digital connectivity remain scarce, and temporary or short-term funding mechanisms prevent institutions from making the sustained investments that AI research requires. At the same time, brain drain continues to pull skilled professionals abroad, and most AI tools in use are trained on data from the Global North, undermining their accuracy and relevance for Latin American populations. The result is a research ecosystem that produces pockets of excellence but lacks the coordinated networks, shared infrastructure, and policy engagement needed to scale them.

One response to this challenge is BiotrAIn, a partnership between EMBL’s European Bioinformatics Institute, the University of Costa Rica, and the regional network CABANAnet, with funding from the Chan Zuckerberg Initiative. Building on the CABANA model, which successfully strengthened bioinformatics capacity for pathogen research and genomic surveillance across the region, BiotrAIn set out to develop a foundational, sustainable AI curriculum for the biosciences. The project ran from 2024 to 2026 in three phases: a knowledge exchange meeting, curriculum development, and a cascade of regional training sessions designed to multiply impact far beyond the initial cohort of learners.

The first phase, a virtual knowledge exchange meeting held in December 2024, brought together more than 40 professionals working on machine learning, AlphaFold-based structural biology, bioinformatics, and data science. Ninety percent of participants came from Latin America, with additional observers from the United States, Africa, and Asia. The event served a dual purpose: it showcased real-world applications of AI in pathogen research, human health, microbiomes, agriculture, bioethics, and epidemiology, and it convened a curriculum-development taskforce of eight regional scientific leaders who would go on to design and deliver the pilot course. This taskforce became the backbone of the project’s Community of Practice, an enduring network intended to outlive the funding period.

The curriculum itself was co-created over roughly four months through an iterative process that combined needs identified at the exchange meeting, a regional survey of 118 bioscientists, a review of existing AI and bioinformatics training programs, and the expertise of the taskforce. The designers also incorporated international guidance on responsible AI, including UNESCO recommendations and the FAIR data principles promoted through EMBL-EBI. The resulting course covers general AI concepts, data preprocessing, ethics and responsible AI use, classification, clustering, deep learning, and AlphaFold for structural biology, capped by a train-the-trainer session to amplify dissemination. Crucially, the training datasets are drawn from Latin American projects, grounding the exercises in science that participants recognize as their own. The complete curriculum, in three versions, is openly available on GitHub under a permissive license, allowing reuse and adaptation well beyond the original project.

The pilot course ran for five days in June 2025 at the University of Costa Rica, drawing 25 participants selected from 118 applicants on criteria including programming familiarity, English proficiency, and biological background, with explicit attention to geographic representation, gender equity, and the inclusion of underrepresented groups. The cohort spanned cities from Buenos Aires to Cochabamba to Belém. In April 2026, an updated version of the course scaled up dramatically through a hub-and-spoke model: Costa Rica acted as the central coordinating hub supporting six local training nodes in Colombia, Mexico, Chile, Argentina, Brazil, and Uruguay, creating seven simultaneous classrooms. From 235 applications, 169 participants were selected, representing 12 countries, supported by 57 trainers and organizers. The model combined local classroom teaching with live cross-site question-and-answer sessions, a shared handbook, and a Slack workspace, keeping the classrooms connected across borders. A subsequent five-week asynchronous virtual course in August and September 2026 attracted 325 applications and enrolled 260 participants from 17 countries—more than the pilot and hub-and-spoke courses combined—demonstrating the depth of unmet demand.

Alongside the training, the project conducted a systematic assessment of AI in the life sciences across the region, analyzing survey responses from 118 bioscientists using cause-and-effect diagrams and Pareto charts to prioritize challenges. The most frequently cited barrier was infrastructure: access to computing power, cloud platforms, and connectivity, tightly linked to chronic funding shortfalls. Human capital came next, with respondents calling for training tailored to local languages, societal challenges, and resource constraints, and for train-the-trainer approaches that scale teaching through educators. Regional integration emerged as a third priority, with the authors emphasizing that AI models built on Global North data are often poorly calibrated for Latin American populations and that local stakeholders must be engaged in data collection, governance, and application. Data access, standardization, and the expansion of ethically valid, AI-ready local datasets round out the priority list, followed by emerging concerns about ethics frameworks, private-sector integration, and the environmental footprint of AI.

The authors are candid about BiotrAIn’s limitations. The curriculum’s breadth—spanning pathogen genomics, human and animal health, biodiversity, agriculture, and drug discovery—made it relevant to many communities but prevented deep dives into any single discipline’s specific methodological needs. Some modules proved too dense for the time available in face-to-face delivery, prompting adjustments in later online iterations. Language also remains an issue: the core materials are in English to enable international connection, even though Spanish and Portuguese dominate the region, a gap only partially addressed by delivering hub-and-spoke sessions in local languages. And AI itself evolves faster than any curriculum, with agent-based AI expanding substantially after the project was designed and left out of the original content. Training, the authors stress, cannot by itself resolve structural inequities in infrastructure, data availability, or governance.

Nevertheless, the lessons drawn from BiotrAIn carry weight well beyond Latin America. A distributed training model that develops local trainers and facilitators leaves capacity embedded in institutions after a project ends. Openly licensed core materials provide a foundation that local educators can adapt to their own disciplines, languages, and datasets. Regional relevance matters not only in scientific examples but in the data used during training. And AI education should be treated as an iterative, continuously updated process rather than a one-time intervention. With coordinated investment, inclusive policies, and collaborative networks, the authors conclude, Latin America can position itself as a significant contributor to global AI-driven bioscience—and BiotrAIn offers a replicable blueprint for any region seeking to build a more equitable, ethical, and sustainable AI ecosystem from the ground up.

Subject of Research: AI capacity building and training for biosciences in Latin America through the BiotrAIn project

Article Title: Landscape and training opportunities in artificial intelligence for biosciences in Latin America: the BiotrAIn project

Article References: Landscape and training opportunities in artificial intelligence for biosciences in Latin America: the BiotrAIn project. (n.d.). https://doi.org/10.3389/frai.2026.1931300

Image Credits: AI Generated

DOI: 10.3389/frai.2026.1931300

Keywords: artificial intelligence, biosciences, BiotrAIn, capacity building, Latin America, bioinformatics, machine learning, curriculum development, AlphaFold, FAIR data, AI ethics, EMBL-EBI