Principles of AI-driven Neuroscience and Translational Biomedicine

Aims

The module Principles of AI-Driven Neuroscience and Translational Biomedicine aims to provide a coherent, multi-scale understanding of neuroscience through the lens of artificial intelligence, integrating concepts that span from brain structure and imaging (Lecture 1) to multi-omics biology (Lecture 2) and advanced AI methodologies (Lectures 3–7), culminating in real-world clinical and translational applications (Lecture 8). At its core, the module explores how modern AI architectures—including convolutional neural networks (CNNs), transformers, graph neural networks (GNNs), and emerging agentic AI systems—can be used to model, analyse, and interpret complex biological and neurological data. Particular emphasis is placed on understanding how these models interact with brain anatomy, connectomics, and neuroimaging modalities (such as structural and functional MRI), multi-omics layers (including genomics, transcriptomics, proteomics, metabolomics, and microbiome data), and large-scale clinical datasets that enable population-level neuroscience. A central aim of the course is to move beyond the use of AI as a purely predictive tool and instead critically examine the origins of its predictive power. Students will engage with both attributional interpretability (e.g., feature importance, saliency, and explanation methods) and mechanistic interpretability (e.g., internal representations, superposition, and circuit-level understanding) to validate learned patterns, identify biologically meaningful signals, and assess model reliability in high-stakes biomedical contexts.

The course includes the following eight lectures (2h) covering foundational topics:

• Lecture 1: Introduction to Basic Brain Anatomy, Connectomes and Neuroimaging
• Lecture 2: Introduction in Multi-Omics and Neurobiology
o Lecture 3: Fundamental Deep Learning and Geometric Deep Learning in Neuroscience
o Lecture 4: Clinical BIG-DATA Neuroscience and Beyond  
o Lecture 5: Large Language Models, Catastrophic Forgetting and Neuroscience  
o Lecture 6: Attributional Interpretability Approaches in Medical Large Language Models (m-LLM), Neuroimaging and Neurobiology
o Lecture 7: Principles of Mechanistic Interpretability Approaches in Medical Large Language Models (m-LLM) and Neurobiology
o Lecture 8: Neuroscience Applications in Multi-Omics, Clinical Reports and Neuro-Imaging

Syllabus

Lecture 1: Introduction to Basic Brain Anatomy, Connectomes and Neuroimaging
The aim of this lecture is to introduce basic brain anatomy, focusing on different lobes, sulci regions, and the brain’s folding patterns. Additionally, it will introduce the fundamental characteristics of various medical imaging modalities, with an emphasis on anatomical and functional MRI. The lecture will also present methods for extracting connectome information [25] using imaging techniques such as functional and structural MRI. [1,2]  

Lecture 2: Introduction in Multi-Omics and Neurobiology
The aim of this lecture is to introduce the conceptual and technological foundations of multi-omics and their application to neurobiology, with a particular focus on the emergence of artificial intelligence (AI)–driven integrative models. Modern biology has transitioned from single-layer analyses to multi-dimensional characterization of biological systems, enabling a more comprehensive understanding of brain function, disease mechanisms, and cellular heterogeneity.  [40,41,42,43] ((invited biology expert like: Dr. Elizabeth Cooper, Gilbertson group CRUK-CI  )

Lecture 3: Fundamental Deep Learning and Geometric Deep Learning in Neuroscience
This lecture will briefly recap the fundamentals of deep learning architectures and layers, such as MLPs, transformers, attention layers, and CNN blocks. It will also present state-of-the-art classifier architectures like ViT and introduce classification problems in neuroscience [3,4]. Moreover, in this lecture, we will introduce the fundamental concepts of graph representation in neuroscience, focusing on how connectomics [21] can be used to model brain imaging data by using nodes as brain regions and edges as connections. Additionally, we will discuss Graph Neural Networks (GNNs), covering essential architectures such as GCNs [22] and GATs [23].  

Lecture 4: Clinical BIG-DATA Neuroscience and Beyond  
Building foundations in anatomical imaging and connectomes; the lecture progresses toward understanding how knowledge can be systematically extracted at scale. Students are introduced to the computational pipelines and analytical frameworks required to work with large clinical datasets, such as the UK Biobank. Within these extensive data resources, meaningful scientific questions emerge that link brain structure, function, and inter-individual variability, enabling population-level analyses and supporting clinically grounded discoveries in neuroscience. (invited medical expert  like: Associate Professor Richard Bethlehem)

Lecture 5: Large Language Models, Catastrophic Forgetting and Neuroscience
This lecture introduces catastrophic forgetting and model collapse in artificial neural networks, focusing on challenges arising from sequential and large-scale learning. It covers core concepts in continual learning, including replay mechanisms, parameter and functional regularization (e.g., Elastic Weight Consolidation), and optimization-based strategies to preserve prior knowledge . The course also examines state-of-the-art approaches such as Hard Attention to the Task (HAT), gradient-based memory methods, and Low-Rank Adaptation (LoRA) in large models. Emphasis is placed on connections to cognitive neuroscience, highlighting parallels between machine and human forgetting, and on applications in neuroimaging and AI-driven biomedical systems [35,36,37,38,39].

Lecture 6:  Attributional Interpretability Approaches in Medical Large Language Models (m-LLM), Neuroimaging and Neurobiology
This session will highlight various Explainable AI (XAI) techniques in attributional interpretability, such as GradCam, SHAP and GNNExplainer. [9-11, 24]. This lecture will highlight how we can use the XAI to verify patterns and potentially identify new biomarkers [16,17,26] in Large Language Models, Transformers and GNNs in medical applications, neurobiology and neuroimaging.

Lecture 7: Principles of Mechanistic Interpretability Approaches in Medical Large Language Models (m-LLM) and Neurobiology
This lecture will cover key terminology related to superposition, polysemantic representations, and the privileged basis. Additionally, it will address the problem of mechanistic interpretability and explore how the sparse autoencoder attempts to provide explanations for various deep learning applications in the medical domain, such as clinical assignment (e.g., medical large language models, m-LLMs) and neurobiology (e.g., multi-omics data, protein language models, PLM) [12-14,18, 33, 34].  

Lecture 8: Neuroscience Applications in Multi-Omics, Clinical Reports and Neuro-Imaging
This Lecture advances to the intersection of theory and real-world practice. In clinical and applied settings, concepts from brain anatomy, machine learning, reasoning, and interpretability are integrated to form cohesive analytical frameworks that support practical decision-making and translational neuroscience applications.  (invited industrial and medical companies speakers (like Professor Richard Gilbertson's Group, CRUK-CI, Astex )

Learning Outcomes

Students will develop practical skills in applying advanced AI methods, including deep learning and interpretability techniques, to real-world neuroscience problems. They will learn to analyse brain data across scales, evaluate models in clinical contexts, and critically assess and refine AI methodologies.

By the end of the module, students will be able to:

•  Develop an integrated understanding of neuroscience across scales, from molecular (multi-omics) to systems-level (connectomes and imaging)
• Gain practical and conceptual knowledge of deep learning and geometric deep learning approaches applied to brain data
• Understand how AI models are trained, adapted, and challenged in real-world biomedical settings, including issues such as catastrophic forgetting and continual learning
• Apply interpretability techniques to understand and evaluate AI models, enabling the discovery and validation of biomarkers and disease-relevant patterns
•  Analyse real-world clinical case studies, including psychosis, brain cancer, and neurodegenerative disorders, using AI-driven frameworks
•  Critically evaluate the role of AI in advancing neuroscience, particularly in multi-omics integration, neuroimaging analysis, and clinical decision support
• Through a combination of lectures and hands-on activities, students will gain the skills needed to contribute meaningfully to the advancement of AI in neuroscience research.

Assessment

The assessment structure will be as follows:

• Theory Test (20%): To evaluate students' understanding of the topics covered through Lecture 1-5, including the fundamental concepts introduced therein, students will be required to complete a 30-minute in-class, closed-book theory test. The test will be administered after Lecture 5 and before Lecture 7 (similar to the assessment format used in Reinforcement Learning, Lent Term).

• Group Mini-Project Report (65%): Students will complete a mini-project in pairs and submit a written report at the end of the course. Students may either propose their own project ideas or express a preference for one of the topics provided at the beginning of the term. The report will be limited to 4,000 words, in line with other modules. Each group will consist of two students.

To ensure a fair assessment of individual contributions, each student will be required to submit a brief statement at the end of the project describing their teammate's contributions. This will provide additional insight into individual effort, team dynamics, and the overall quality of collaboration. Furthermore, each project report must include a contribution statement outlining the role of each team member, which can be cross-referenced with the individual submissions.

• Presentation and Viva (15%): Students will deliver a short presentation outlining their individual contributions to the mini-project, followed by a brief viva examination.

Recommended Reading Material and Resources

Theory:  
[1]https://www.sciencedirect.com/science/article/pii/S1053811913002656
[2] https://pmc.ncbi.nlm.nih.gov/articles/PMC1239902/  
[3] https://arxiv.org/abs/2305.09880  
[4] https://arxiv.org/abs/2010.11929  
[5] https://arxiv.org/abs/1801.10130  
[6] https://arxiv.org/abs/2105.13926  
[7] https://arxiv.org/abs/1910.12892  
[8] https://arxiv.org/abs/2303.15919  
[9] https://arxiv.org/abs/1602.04938  
[10] https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0130140  
[11] https://arxiv.org/abs/1705.07874  
[12] https://arxiv.org/abs/1610.02391  
[13] https://arxiv.org/abs/2209.10652  
[14] https://arxiv.org/abs/2309.08600
[15] https://www.nature.com/articles/s43856-023-00313-w  
[16] https://arxiv.org/abs/2309.00903  
[17] https://arxiv.org/abs/2405.10008  
[18] https://www.biorxiv.org/content/10.1101/2024.11.14.623630v1  
[19] https://www.sciencedirect.com/science/article/pii/S1361841524000471  
[20] https://www.sciencedirect.com/science/article/pii/S1361841522001189
[21] https://www.sciencedirect.com/science/article/pii/S105381190901074X  
[22] https://arxiv.org/abs/1609.02907  
[23] https://arxiv.org/abs/1710.10903  
[24] https://arxiv.org/abs/1903.03894  
[25] 10.1016/j.neuroimage.2011.05.025
[26] https://doi.org/10.1109/CITRExCompanion65208.2025.10981502  
[27] S. Khemlani, P. N. Johnson-Laird (2012), Theories of the syllogism: A meta-analysis, Psychological Bulletin 138 (3) 427–457.  
[28] S. Ferrigno, Y. Huang, J. F. Cantlon (2021), Reasoning Through the Disjunctive Syllogism in Monkeys, Psychological Science 32 (2) 1–9.  
[29] J. L. S. Bellmund, P. Gaerdenfors, E. I. Moser, C. F. Doeller (2018), Navigating cognition: Spatial codes for human thinking, Science 362 (6415).  
[30] K. L. Alfred, A. C. Connolly, J. S. Cetron, D. J. M. Kraemer (2020), Mental models use common neural spatial structure for spatial and abstract content, Communications Biology 3 (1).  
[31] M. Ragni, M. Knauff (2013), A theory and a computational model of spatial reasoning with preferred mental models, Psychological review 120:561–588.  
[32] T. Dong, M. Jamnik, P. Liò. (2025). Neural Reasoning for Sure Through Constructing Explainable Models. Proceedings of the AAAI Conference on Artificial Intelligence, 39(11), 11598-11606.  
[33] https://doi.org/10.1109/BIBM62325.2024.10822695  
[34] https://doi.org/10.1109/BIBM62325.2024.10821894
[35]  https://doi.org/10.1016/B978-0-443-15754-7.00073-0
[36]  https://doi.org/10.48550/arXiv.2406.04836
[37] https://doi.org/10.48550/arXiv.2509.01213
[38] https://doi.org/10.48550/arXiv.2402.01348
[39]  https://doi.org/10.1088/0256-307X/39/5/050303
[40]  https://www.nature.com/articles/s41746-026-02553-1
[41] https://www.nature.com/articles/s41587-022-01284-4
[42]  https://www.nature.com/articles/s41592-025-02808-x
[43]  https://www.nature.com/articles/s41592-025-02899-6

Books: 

[1b] "Deep Learning", by Ian Goodfellow, Yoshua Bengio and Aaron Courville.  
[2b] “Graph Representation Learning Book", by William L. Hamilton.  
[3b] “Graph Neural Networks: Foundations, Frontiers, and Applications”, by Lingfei Wu, Peng Cui, Jian Pei, and Liang Zhao.  
[4b] “Changing Connectomes: Evolution, Development, and Dynamics in Network Neuroscience.” by Kaiser, M. (2020). MIT Press. ISBN: 978-0262044615  
[5b] F. Rosenblatt (1962), Principles of Neurodynamics: Perceptrons and the Theory of Brain Mechanisms, Spartan Books, Washington, USA.  
[6b] B. Tversky (2019), Mind in Motion, Basic Books, New York, USA.  
[7b] "Cognitive Foundations of Agentic AI - From Theory to Practice" by Anand Vemula  
[8b] Murphy, K.P., 2012. Machine Learning: A Probabilistic Perspective. Cambridge, MA: MIT Press.
Coding:  
[1] https://pytorch-geometric.readthedocs.io/en/latest/  
[2] https://captum.ai/  
[3] https://pytorch.org/hub/huggingface_pytorch-transformers/  
[4] https://huggingface.co/  
[5] https://transformerlensorg.github.io/TransformerLens/  
[6] https://sites.google.com/site/bctnet/
[7] https://www.sc-best-practices.org/preamble.html