PDF Basket
From drug discovery and development to personalised medicine, artificial intelligence (AI) has immense potential in life sciences. But much of this potential remains untapped. AI integration is often hampered by technical complexities and a lack of standardised data.
The AI4LIFE project, managed by Dorothea Dörr, scientific project manager at Euro-BioImaging in Finland, is tackling these challenges head-on. “Deploying and using AI models often requires advanced expertise in programming, machine learning methods and even computational infrastructures, which many life scientists do not have,” she says. “This expertise gap leaves many researchers unable to fully make use of AI’s capabilities.”
AI4LIFE is creating user-friendly platforms that make AI tools and data sets accessible even to those with no expertise in computational sciences. These platforms will facilitate the analysis of biological images such as microscopy scans of tissue samples. The systems allow researchers to leverage AI for complex tasks such as segmentation, object detection and feature extraction.
The project particularly focuses on FAIR (Findable, Accessible, Interoperable and Reusable) principles for AI-ready image data sets and models. “For AI to be effective in life sciences, data sets that are used to train AI models must be not only large and high quality but also FAIR and AI-ready,” explain Anna Kreshuk and Florian Jug, the scientific coordinators of the project.
This means that data is formatted, cleaned and labelled to the extent that it can easily be used to train a new AI model or feed an existing one. For this purpose, AI4LIFE has developed a standard for AI-ready image data set annotation. It is hoped that this will encourage collaboration and reuse while reducing the need for individual researchers to reprocess data, thus cutting computational costs.
Collaboration and open science
The AI4LIFE consortium is committed to developing a community-driven repository of FAIR-compliant AI models for biological image analysis, known as the BioImage Model Zoo. “AI4LIFE’s AI models are developed collaboratively by life scientists and AI experts,” explain Kreshuk and Jug. “This ensures that they are cutting edge and tailored to meet real-world needs in life science research.”
The repository not only allows scientists to access pre-trained models that can be deployed using popular image analysis software tools, but also encourages them to contribute their own, creating a dynamic ecosystem of tools that can be reused and adapted across various research domains.
The initiative’s open and collaborative nature is further demonstrated through its innovative use cases. One example is the ‘Open Calls’ programme, where AI4LIFE offers deep-learning-based image analysis support designed to address specific research needs. This programme has already helped automate complex tasks that would otherwise be highly labour-intensive, allowing researchers to focus more on interpreting biological data than handling technical details. “By leveraging deep learning, researchers can analyse large and complex data sets with higher accuracy and efficiency,” Kreshuk and Jug note, highlighting the efficiency gains made possible by AI4LIFE.
In a notable case, the project’s ‘Denoising Challenge’ launched an open competition to develop the best AI methods for improving the quality of microscope images. Competitions allow developers to refine their methods and provide life scientists with a clear overview of the most effective tools available. This not only drives innovation, but also enhances the visibility of leading solutions, promoting their adoption across the scientific community.
Ensuring long-term impact
Beyond the development of AI models, the AI4LIFE project is heavily involved in community engagement and outreach to maximise the impact of its solutions. “The project actively engages in extensive dissemination efforts, offering training sessions and presentations for life scientists,” says Dörr. This helps ensure that life scientists across all career stages gain the skills needed to integrate AI into their research.
Additionally, AI4LIFE promotes transparency by regularly publishing its research outcomes in open access journals and presenting its findings at international conferences. This open approach allows other scientists to build upon AI4LIFE’s models, expanding the project’s reach.
Finally, the project’s collaboration with the AI4EOSC project is helping to support novel applications for European infrastructures. Integrating AI models within European Open Science Cloud frameworks will make them readily accessible and scalable, while partnerships with industry ensure that AI4LIFE’s solutions benefit not only academic research but also practical applications in various sectors.
By making AI accessible, fostering collaboration and adhering to open science principles, AI4LIFE is setting the stage for transformative advances in life sciences research across Europe. As Dörr remarks: “The adherence to FAIR principles ensures that AI models can be easily reused and adapted by other researchers for their own data, expanding the impact of the project.” Through such efforts, AI4LIFE is driving forward a new era where AI tools are not only cutting edge but also accessible to all, regardless of technical expertise.