Representations of Structured Biological Data and Prototype-based Machine Learning - dissertation propositions

Aug 5, 2026

The propositions present a research direction for computational biology and machine learning focused on structured biological data. They argue that structured data is inherently complex and varies in form and dimensionality, undermining standard ML assumptions. The propositions emphasize task-aware modular representations, and highlight the importance of interpretability: information-theoretic representations capture dependency structure, reference-based embeddings avoid expensive pairwise comparisons, and sensory response principles clarify embedding learning. Prototype-based classifiers are framed as interpretable yet flexible, while deep models are described as accurate but less transparent.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages2

Reading Time5 min

AccessFree to download

Propositions

Challenges of structured biological data

Representation strategies and interpretability

University of Groningen Representations of structured biological data and prototype-based machine learning Bohnsack, Katrin DOI: 10.33612/diss.1436507459 IMPORTANT NOTE: You are advised to consult the publisher's version (publisher's PDF) if you wish to cite from it. Please check the document version below. Document Version Publisher's PDF, also known as Version of record Publication date: 2025 Link to publication in University of Groningen/UMCG research database Citation for published version (APA): Bohnsack, K. (2025) . Representations of structured biological data and prototype-based machine learning.[Thesis fully internal (DIV), University of Groningen] . University of Groningen. [https://doi.org/10.33612/diss.1436507459](https://doi.org/10.33612/diss.1436507459) Copyright Other than for strictly personal use, it is not permitted to download or to forward/distribute the text or part of it without the consent of the author(s) and/or copyright holder(s), unless the work is under an open content license (like Creative Commons) . The publication may also be distributed here under the terms of Article 25fa of the Dutch Copyright Act, indicated by the “Taverne” license. More information can be found on the University of Groningen website: [https://www.rug.nl/library/open-access/self-archiving-pure/taverne](https://www.rug.nl/library/open-access/self-archiving-pure/taverne)amendment. Take-down policy If you believe that this document breaches copyright please contact us providing details, and we will remove access to the work immediately and investigate your claim. Downloaded from the University of Groningen/U MCG research database (Pure): [http://www.rug. nl/research/portal. For technical reasons the](http://www.rug. nl/research/portal. For technical reasons the)[ ](http://www.rug. nl/research/portal. For technical reasons the)[number of authors shown on this cover page is limited to 10 maximum.](number of authors shown on this cover page is limited to 10 maximum.) Download date: 30-12-2025 Propositions accompanying the dissertation REPRESENTATIONS OF STRUCTURED BIOLOGICAL DATA AND PROTOTYPE-BASED MACHINE LEARNING by Katrin Sophie Bohnsack 1. Structured biological data is inherently complex and varies widely in form and dimensionality, which challenges conventional machine learning assumptions. 2. No single similarity measure or representation suits all purposes: modular, taskaware design is key to effective structured data analysis. 3. In computational biology, clarity about what a model learns is as important as what it predicts. 4. Information-theoretic representations express the dependency structure inherent in data, offering a principled account of its internal organization. 5. Reference-based embeddings provide a scalable strategy for representing structured data without costly pairwise comparisons. 6. The sensory response principle is akin to mathematical embedding via independent basis functions, but it provides a clearer, more illustrative explanation of embedding learning that is accessible to practitioners in diverse quantitative fields. 7. Deep models are powerful tools in computational domains but suffer a black-box drawback; shallow approaches incorporating expert knowledge often match predictive performance while offering superior simplicity and insights. 8. Prototype-based classifiers, such as Learning Vector Quantization, balance structural flexibility with direct interpretability, making them well-suited for domainspecific applications. 9. Classification prediction without interpretation is like explanations without any objective beyond. 10. A full-limb orthopedic cast is the most effective biomedical focus mechanism, maximizing thesis productivity by enforcing the elimination of distractions (validated by a highly flawed self-study, n = 1) .

FAQs

Why do conventional machine learning assumptions struggle with structured biological data?

Structured biological data is described as inherently complex and varying widely in form and dimensionality, which challenges conventional ML assumptions.

What representation approaches are proposed for understanding or learning from structured data?

The propositions include information-theoretic representations for dependency structure, reference-based embeddings for scalable representation without costly pairwise comparisons, and sensory response principles as an accessible embedding explanation.

How do prototype-based classifiers fit the goals of flexibility and interpretability?

Prototype-based classifiers, including Learning Vector Quantization, are presented as balancing structural flexibility with direct interpretability, making them suitable for domain-specific applications.

Aug 5, 2026

Upload by Genevieve