Robust Prediction and Biomarker Discovery in Rare Cancers Using Interpretable Machine Learning - thesis

Aug 4, 2026

Rare cancers such as glioblastoma multiforme (GBM) create persistent challenges in computational oncology because limited data, biological noise, and the difficulty of finding disease-specific molecular signatures undermine standard modeling. Contrary to the expectation of poor rare-cancer performance, machine learning on genomic data achieves strong accuracy, raising the need to distinguish genuine biology from artifacts. This thesis builds an interpretable framework to test predictive robustness and extract biologically meaningful biomarkers under extreme class imbalance, using cascade learning, SHAP interpretability, and Tab2Image visualization, with emphasis on transparency and accountability in scarce-data settings.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages91

Reading Time3 h 49 min

AccessFree to download

Abstract

Introduction

Context and Motivation

Problem Statement

Research Question, Motivation, and Approach

Thesis Contributions

Related Work

Machine Learning in Rare Cancer Prediction

Interpretable Machine Learning for Transparency

Multi-Stage and Cascade Learning Approaches

Methodology

Overview

Architecture and Technologies

Data Sources and Provenance

Dataset Size Summary

Exploratory Data Analysis (EDA)

Data Preprocessing

Robustness Evaluation Design

Classification Models

Cascade Learning for Biomarker Isolation

Interpretability and Feature Analysis

Ethical Considerations

Results and Discussion

Predictive Robustness of the GBM Gene Signature

PCA Analysis: Visual Evidence of Global Separability

SHAP Interpretability Confirms Biological Consistency

Biomarker Discovery Using SHAP and Feature Importance

Biological Validation of Consensus Biomarkers

Cascade Learning and Biomarker Discovery

Tab-to-Image Visualization

Discussion

Conclusion

Bibliography

Appendices

©Copyright 2025 Dhurka Rohini Madasamy Robust Prediction and Biomarker Discovery in Rare Cancers Using Interpretable Machine Learning Dhurka Rohini Madasamy A thesis submitted in partial fulfillment of the requirements for the degree of Master of Science University of Washington 2025 Committee: Wooyoung Kim Munehiro Fukuda Bill Erdly Program Authorized to Offer Degree: Computer Science & Software Engineering University of Washington Abstract Robust Prediction and Biomarker Discovery in Rare Cancers Using Interpretable Machine Learning Dhurka Rohini Madasamy Chair of the Supervisory Committee: Wooyoung Kim Department of Computing and Software Systems Rare cancers such as Glioblastoma Multiforme (GBM, a rare brain cancer) pose persistent challenges in computational oncology due to limited data, biological noise, and difficulty in isolating disease-specific molecular signatures. Based on these constraints, this work began with the expectation that rare-cancer models would perform poorly. However, machine learning approaches on genomic data achieved unexpectedly strong accuracy, motivating investigation into whether this separability reflected genuine biology or artifactual signal. This thesis develops an interpretable machine learning framework that evaluates predictive robustness and isolates biologically meaningful biomarkers under extreme imbalance. Cascade Learning systematically removes broad cancer pathways and reveals biomarkers uniquely associated with the rare cancer, while SHAP-based interpretability aligns these genes with experimentally reported glioma biology. Complementary Tab2Image visualizations provide spatial confirmation of class separability, strengthening biological trust in the learned signal. Overall, this work provides a robust, biologically grounded, and ethically aligned pathway for rare-cancer biomarker discovery that emphasizes transparency, fairness, and accountability in scarce-data environments. TABLE OF CONTENTS Page Abstract ........................................... i Table of Contents ...................................... ii List of Figures ....................................... iii List [of Tables ........................................ vi](of Tables ........................................ vi) [Glossary ........................................... vii](Glossary ........................................... vii) [Acknowledgements ..................................... x](Acknowledgements ..................................... x) [Dedication .......................................... xi](Dedication .......................................... xi) [Chapter 1: Introduction ................................ 1](Chapter 1: Introduction ................................ 1) [1.1 Context and Motivation ............................. 1](1.1 Context and Motivation ............................. 1) [1.2 Problem Statement ................................ 3](1.2 Problem Statement ................................ 3) [1.3 Research Question](1.3 Research Question), [Motivation](Motivation), [and Approach](and Approach) ................. 4 1.4 Thesis Contributions ............................... 4 Chapter 2: Related Work ................................ 5 2.1 Machine Learning in Rare Cancer Prediction .................. 5 2.2 Interpretable Machine Learning for Transparency ............... 6 2.3 Multi-Stage and Cascade Learning Approaches ................. 7 2.4 Summary ..................................... 7 Chapter 3: Methodology ................................ 8 3.1 Overview ...................................... 8 3.2 Architecture and Technologies .......................... 9 3.3 Gene Expression Data Sources and Provenance ................ 9 3.4 Dataset Size Summary .............................. 10 3.5 Exploratory Data Analysis (EDA) ........................ 12 3.6 Data Preprocessing ................................

FAQs

What problem does the thesis address in rare cancer modeling?

Rare cancers like GBM are difficult to model due to limited data, biological noise, and challenges isolating disease-specific molecular signatures. The thesis focuses on making predictions robust and extracting biomarkers despite extreme imbalance.

How does the work improve interpretability of the learned biomarkers?

The framework uses SHAP-based interpretability to link selected genes to experimentally reported glioma biology. Cascade learning further isolates biomarkers by removing broad cancer pathways in stages.

What methods validate whether the model signal reflects meaningful biology?

The thesis combines predictive robustness evaluations with biological validation against known glioma/GBM signatures. Tab2Image visualizations provide spatial confirmation of class separability, supporting trust in the learned signal.

Aug 4, 2026

Upload by Valentina