Recipient organisationUniversity of ExeterSource-published name: University of Exeter
Funding£161K
PeriodNov 2025 — May 2027
In plain English
AI plain-English summary
Artificial intelligence models can now predict how DNA sequences control cell behaviour, but they cannot explain their own reasoning. This project will build an open-source, web-based tool called Explainome that lets biologists with no coding skills ask an AI agent why it made a particular prediction and get answers grounded in real biological knowledge. The problem is that today’s most powerful AI models in biology are black boxes. A deep neural network might accurately predict which genetic mutations cause disease, but it cannot tell a researcher *why*. Existing explainable AI methods exist, but they lack standard guidelines for biological use, and their outputs are difficult to connect to actual mechanisms inside cells. Explainome will coordinate three technologies—machine learning explainers, biological knowledge graphs, and large language models—to bridge that gap. If successful, a biologist could type a question like “Why does this model predict that this DNA change disrupts gene regulation?” and receive a conversational answer backed by known molecular pathways. This is fundamental, curiosity-driven research. It will not immediately change a patient’s treatment or a farmer’s crop. But by making AI transparent and trustworthy, it could accelerate discovery in areas from sustainable agriculture to new antibiotics—much as open-source tools once transformed genomics.
View original technical description
Artificial intelligence (AI) and machine learning (ML) have greatly shaped the landscape of computational biology. The integration of high-throughput data acquisition and burgeoning computational power has led to the creation of powerful predictive models. More recently, the advancements inspired by large language models (LLMs)—transformer-based, large-scale pretrained foundation models—have further enhanced our ability to model and analyse complex biological tasks such as DNA sequence analysis, cell-type annotation, and network biology prediction. By decoding massive datasets encoded in the chemical languages of life—DNA base pairs, amino acid sequences, and protein structures—AI is driving innovations in engineering biology, drug discovery, and beyond. Despite the rapid development of powerful AI models, there is a longstanding challenge in interpreting and understanding why these models make certain predictions and whether to trust them. The complexity of these black-box models (e.g., deep neural networks with hundreds of millions of parameters) makes them difficult to comprehend. To address this challenge, explainable AI (XAI) methods have emerged to elucidate the predictions of trained ML models. Whilst XAI methods have gained prominence in computational biology, two significant barriers fundamentally limit their effective use in bioscience discovery: 1) the absence of clear guidelines for implementing and evaluating XAI methods in complex biological contexts makes it difficult to determine which explanations to use, and how to interpret them; and 2) a significant gap in interpreting these explanations and aligning them with the mechanistic insights of complex biological systems. This ambitious project aims to create the world’s first open-source, web-based conversational XAI agent called Explainome, specifically tailored for the bioscience community. By coordinating complex tools like ML and XAI methods, biological knowledge graphs (KGs), and LLMs, Explainome will help bioscientists with minimal technical skills translate XAI-generated explanations from ML model predictions into mechanistic insights of complex biological systems. We will focus on three objectives (Os). O1: Develop an open-source modular software framework of XAI toolkits, enabling easy integration and scalability. O2: Develop application programming interfaces that utilise diverse biological KGs as grounded knowledge bases to validate and rationalise the explanations generated by the XAI toolkits (O1), ensuring alignment with biological mechanisms. O3: Develop a web-based automated XAI agent powered by LLMs that coordinates tools developed in O1 and O2 to allow users to understand ML models with biological underpinnings through interactive dialogues. If this early-stage exploratory project is successful, users will be able to engage in interactive conversations with Explainome, gaining insights grounded in validated biological reasoning. They can explore why predictions occur, how changes in data might influence outcomes, and how modifying inputs could achieve desired results. In summary, Explainome is a multi-disciplinary, transformative project that will hold exciting promise for securing the UK’s world-leading position at the forefront of bioscience and AI. Looking ahead, Explainome will be further developed to continuously learn from new data and scientific findings, enhancing its accuracy and depth of understanding. This adaptivity will enable Explainome to remain relevant in the face of rapidly evolving biological data. By staying at the forefront of bioscience research, Explainome will unlock unprecedented flexibility and possibilities in catalysing AI-driven technologies. It will serve as an ‘AI co-scientist’ working alongside bioscientists to drive innovation and accelerate scientific discovery. This will supercharge scientific advances in understanding biology with deep implications for agriculture, energy, sustainability, healthcare, and beyond.
Plain English summaries and category classifications on this site are generated by AI and may not perfectly reflect the original research.
Is something wrong? Let us know