Network Biology Dissertation Topics

Network Biology Dissertation Topics

Info: Network Biology Dissertation Topics
Published: 19th July 2025 in Network Biology Dissertation Topics

Share this:

Facebook
Twitter
Whatsapp
Linkedin

Related Services

Our academic writing and marking services can help you

Dissertation Topic 1:

Connecting Methodological Paradigms in Network Biology: A Hybrid Approach of Graphlets, GNNs, and Geometric Deep Learning

Description & Research Gap

While network biology has become a mature field, methodological silos have persisted between graph-theoretic based approaches (e.g. graphlets), deep learning based models (e.g. GNNs), and geometric deep learning approaches. Instead, each has developed independently with evaluation made independently of one another often with limited feature comparisons across these models. Consequently, there remains uncertainty about the relative capabilities of each method and associated advantages for application purpose. Moreover, these methods are deficient in their ability to quantify uncertainty reliably and mostly miss consideration of meaningful established biological hierarchies.

This research addresses the gap for the integration and comparative understanding between:
  • Traditional graph-based models,
  • Deep learning-based embeddings (e.g. GNNs), and
  • More recent geometric deep learning approaches that have been built specifically for 3D biological structures.

    • Research Objectives

    1. To compare and evaluate standard graphlet based methods, GNNs and geometric deep learning under the same network biology tasks.
    2. To produce a model that combines graphlet statistics, message passing GNNs, and 3D protein-oriented structure embeddings taking advantage of their complementary strengths.
    3. To present uncertainty quantification as an end product in all predictive models to quantify model confidence, for example in the aleatoric and epistemic contexts.
    4. To combine biological ontologies (e.g. GO, Disease Ontology) and obtained knowledge from known hierarchical bases into model architectures with the aim of improving relevance and interpretation of biological meaning.

    Proposed Methods

    Benchmarking tasks: protein function prediction, PPI prediction, drug–target interactions prediction, and graph classification using publicly available data sets.
    Hybrid architecture: a modular framework that integrates graphlet-based topological features, standard and higher-order GNNs, geometric deep learning layers to provide 3D structure embeddings, and has transformer-based graph components to accommodate the encoding of longer-distance dependencies.
    Estimate uncertainty: Use Bayesian GNNs and Monte Carlo dropout to explore, and distinguish, between aleatoric and epistemic uncertainty.
    Modeling with ontological guidance: using ontology-based inductive biases within the model through visible neural networks or structured attention.
    Evaluative framework: like any model, we will use cross-validation for model selection and once we select a robust model we will benchmark against standard datasets (e.g., CAFA, DREAM, PDB) to evaluate on accuracy, scalable architecture, and interpretability.

    Expected Outcomes

  • A hybrid model that can outperform existing bench-marked single paradigm models in one or more key tasks (across accuracy, scalability, and interpretability).
  • A comparative analysis detailing the performance trade-offs between graphlets, GNNs, and geometric deep learning as these premium machine learning paradigms in the industry.
  • A suite of tools for uncertainty-aware network analyses in biology to assist with ‘robust’ decision-making in biomedical settings.
  • A set of guidelines, and framework for benchmarking new models, this contribution would be pivotal for improved standardization in network biology evaluation.

    • Key References

    1. Chen et al. (2020) – On limitations of message-passing GNNs and expressive power:
    Chen, H., et al. “Can GNNs count substructures?”
    2. Nelson et al. (2019) – Comparative study of embeddings vs graph-based methods:
    Nelson, W., et al. “To embed or not to embed: Assessing the effectiveness of graph embeddings.”
    3. Gligorijević et al. (2021) – GNNs for 3D protein structure modeling:
    Gligorijević, V., et al. “Structure-based function prediction using GNNs.”
    4. Tahmasebi et al. (2020, 2023) – Higher-order GNNs for subgraph enumeration:
    Tahmasebi, A., et al. “Subgraph-aware GNNs for network comparison.”
    5. Ma et al. (2018), Gaudelet et al. (2020) – Visible neural networks and biological ontologies:
    Ma, J., et al. “Using biological ontologies in neural network architectures.”

    Dissertation Topic 2:

    Towards a domain-independent network alignment technique: Cross-disciplinary evaluation of biological and social network alignment techniques

    Description & Research Gap

    Much of the current research on network alignment is domain-specific and methods for biological and social networks have not been compared despite both being aimed at solving a necessarily similar problem (i.e. identifying functionally similar nodes across networks). The lack of comparison of biological and social network alignment methods is in part attributable to differences in data types, noise levels and the silos between communities.This project seeks to address this gap by comparatively enabling cross-disciplinary evaluation, adaptation of alignment methods, and assessment of methodological transferability and generalizability between biological and non-biological networks.

    Research Objectives

    1. Compare and evaluate network alignment methods within the field of biology, and within the field of social sciences on common datasets.
    2. Recognize data-related biases (i.e. network type) and adapt models to operate robustly across domains.
    3. Construct an unifying benchmarking framework for equitable comparison across domains.
    4. Propose composite alignment algorithms for fully utilizing strengths from each domain.

    Proposed Methods

  • Assemble a benchmark dataset of biological (e.g., protein interaction, gene regulatory) and social (e.g., collaboration, citation) networks, with synthetic ground-truth alignment.
  • Examine alignment accuracy in relation to precision and recall on matched nodes, and structural integrity.
  • Compare global vs. local, pairwise vs. multiple, and static vs. dynamic alignment approaches.
  • Extend effective methods to address cross-domain heterogeneity and levels of noise.
  • Suggest algorithmic extensions from lessons learned from the project comparative study.

    • Expected Outcomes

  • An extensive comparative study on network alignment across domains.
  • Guidance for how social network approaches should be adapted to biological networks, and vice-versa.
  • A unified benchmarking and evaluation suite that is usable by biological and non-biological network researchers.
  • The potential to publish a generalizable hybrid alignment tool.

    • Key References

    1. Zitnik, M., Li, M.M., Wells, A., Glass, K., Morselli Gysi, D., Krishnan, A., Murali, T.M., Radivojac, P., Roy, S., Baudot, A., Bozdag, S., Chen, D.Z., Cowen, L., Devkota, K., Gitter, A., Gosline, S.J.C., Gu, P., Guzzi, P.H., Huang, H., Jiang, M., Kesimoglu, Z.N., Koyuturk, M., Ma, J., Pico, A.R., Przulj, N., Przytycka, T.M., Raphael, B.J., Ritz, A., Sharan, R., Shen, Y., Singh, M., Slonim, D.K., Tong, H., Yang, X.H., Yoon, B.-J., Yu, H. and Milenković, T., Current and future directions in network biology, Bioinformatics Advances, 2024, 00, vbae099. Available at: https://doi.org/10.1093/bioadv/vbae099 (Accessed: 14 May 2025).
    2. Eyuboglu et al. (2023) – Challenges of comparing biological vs social networks.
    3. Meng et al. (2016) – Global vs local network alignment methods.
    4. Guzzi & Milenković (2017) – Survey and comparison of biological network alignment approaches.
    5. Vijayan et al. (2017, 2020) – Static/dynamic, pairwise/multiple network alignment.
    6. Section 3 of your content – Overview of network-of-networks and alignment challenges.

    Dissertation Topic 3:

    Measuring and Reducing Uncertainty in Graph-Based Biological Predictions

    Description & Research gap

    Although deep learning and graph-based methods are gaining traction in biomedical prediction problems, uncertainty is generally less explored. Distinguishing between “aleatoric” uncertainty (from data) and “epistemic” uncertainty (from models) is especially important in high-stakes areas (e.g. drug discovery, disease diagnosis). Mischaracterizing uncertainty could lead to a wrong prediction, with potentially dangerous consequences. This project aims to systematically investigate and implement uncertainty-aware models for biological network tasks, including using probabilistic GNNs and rigorous testing/benchmarking.

    Research Objectives

    1. Build a framework for quantifying aleatoric/population and epistemic/model uncertainty from graph-based models.
    2. Apply and evaluate uncertainty-aware models in protein function prediction, PPI prediction, and drug–target interactions.
    3. Identify sources of uncertainty between network structure, data quality, and model design.
    4. Offer interpretability methods for evaluating confidence in model predictions.

    Proposed methods

  • Use Bayesian GNNs, Monte Carlo Dropout, and Deep Ensembles for various uncertainty models.
  • Use graph datasets that are missing/noisy labels, to mirror real-world experiences.
  • Evaluate predictive accuracy with calibration plots, entropy scores, and Brier scores.
  • Graph uncertainty across node/edge network to identify key areas.
  • Investigate the influence of biological variability and experimental noise on uncertainty.

    • Expected Outcomes

  • An open-source toolkit for uncertainty quantification in biological networks
  • Better predictive performance and model robustness in the presence of data noise
  • Clear insights regarding when and why models fail, and how confident we are in their predictions
  • Practical guidelines for considering uncertainty in biomedical AI pipelines.

    • Key References

    1. Zitnik, M., Li, M.M., Wells, A., Glass, K., Morselli Gysi, D., Krishnan, A., Murali, T.M., Radivojac, P., Roy, S., Baudot, A., Bozdag, S., Chen, D.Z., Cowen, L., Devkota, K., Gitter, A., Gosline, S.J.C., Gu, P., Guzzi, P.H., Huang, H., Jiang, M., Kesimoglu, Z.N., Koyuturk, M., Ma, J., Pico, A.R., Przulj, N., Przytycka, T.M., Raphael, B.J., Ritz, A., Sharan, R., Shen, Y., Singh, M., Slonim, D.K., Tong, H., Yang, X.H., Yoon, B.-J., Yu, H. and Milenković, T., Current and future directions in network biology, Bioinformatics Advances, 2024, 00, vbae099. Available at: https://doi.org/10.1093/bioadv/vbae099 (Accessed: 14 May 2025).Zhao et al. (2020b) – Overview of uncertainty types in ML.
    2. Hüllermeier & Waegeman (2021) – Principles of uncertainty quantification in structured data.
    3. Cao et al. (2013), Ding et al. (2006) – Use of distances with theoretical guarantees.
    4. Beyer et al. (1999), Radovanović et al. (2010) – Theoretical challenges in high-dimensional distance metrics.

    Share this:

    Facebook
    Twitter
    Whatsapp
    Linkedin

    Cite this work

    Study Resources

    Free resources to assist you with your university studies!

    This will close in 0 seconds