Drawing Biological Understanding From Machine Learning - Doctoral Dissertation

Aug 3, 2026

Uploaded by Maeve

Large biological datasets, including medical imaging and single-cell genomic measurements, contain rich signals that machine learning can transform into actionable predictions or interpretive insights. This dissertation studies three ML applications in biology. It uses deep learning to predict lesion-level risks for metastatic cancer prognosis from CT images and shows the model reflects clinically known risk indicators. It then applies DeepLIFT and TF-MoDISco to analyze how DNA shape affects transcription factor binding, and finally evaluates CellOracle and SCENIC+ for gene regulatory network inference during reprogramming of fibroblasts.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages105

Reading Time4 h 25 min

AccessFree to download

Introduction

Lesion Prioritization for Cancer Prognosis

Background

Outcome-aware Object Detection on Synthetic Lesions

Metastatic Lung Cancer Prognosis via Deep Image-Based Lesion Prioritization

Related work

Interpreting DNA shape in a deep TF binding model

Background

Overview of the DeepShape model

Attribution-based DeepShape analyses

Discussion

Benchmarking Gene Regulatory Networks

Background

Recovering reprogramming factors

Bibliography

UC Berkeley UC Berkeley Electronic Theses and Dissertations Title Drawing Biological Understanding From Machine Learning Permalink [https://escholarship.org/uc/item/9353r08n](https://escholarship.org/uc/item/9353r08n) Author Yang, Forest Publication Date 2024 Peer reviewed|Thesis/dissertation [eScholarship.org](eScholarship.org) Powered by the California Digital Library University of California Drawing Biological Understanding From Machine Learning by Forest Yang A dissertation submitted in partial satisfaction of the requirements for the degree of Doctor of Philosophy in Computer Science in the Graduate Division of the University of California, Berkeley Committee in charge: Professor Laurent El Ghaoui, Chair Assistant Professor Nilah Ioannidis Associate Professor Adam Yala Summer 2024 Drawing Biological Understanding From Machine Learning Copyright 2024 by Forest Yang 1 Abstract Drawing Biological Understanding From Machine Learning by Forest Yang Doctor of Philosophy in Computer Science University of California, Berkeley Professor Laurent El Ghaoui, Chair Large biological data, such as medical imaging and single-cell level genomic data, are rich sources of biological information. Machine learning is a tool to extract that information into a usable form, whether it be predictions for some prediction task or insights drawn from the model. We explore three applications of machine learning to biology. One is on using deep learning to perform metastatic cancer prognosis from CT images by predicting lesion-level risks. We use the lesion-level risks to show that the model captures clinically known indicators of risk. Next, we utilize the DeepLIFT and TF-MoDISco neural network interpretation techniques to understand how DNA shape affects transcription factor binding. Overall, we find that sequence features are more important for distinguishing bound sites, but that shape features can modulate binding affinity. Finally, we the test the CellOracle and SCENIC+ gene regulatory network inference frameworks in the context of reprogramming fibroblasts to pluripotent cells, to prioritize key factors in reprogramming and recover their effects on differentiation. i To my friends and family ii Contents Contents ii List of Figures iii List of Tables ix 1 Introduction 1 2 Lesion Prioritization for Cancer Prognosis 3 2.1 Background .................................... 3 2.2 Outcome-aware Object Detection on Synthetic Lesions ............ 9 2.3 Metastatic Lung Cancer Prognosis via Deep Image-Based Lesion Prioritization 18 2.4 Related work ................................... 29 3 Interpreting DNA shape in a deep TF binding model 34 3.1 Background .................................... 34 3.2 Overview of the DeepShape model ........................ 43 3.3 Attribution-based DeepShape analyses ..................... 47 3.4 Discussion ..................................... 59 4 Benchmarking Gene Regulatory Networks 63 4.1 Background .................................... 63 4.2 Recovering reprogramming factors ........................ 70 Bibliography 77 iii List of Figures 2.1 Kaplan-Meier curves from a recent study (2021) comparing immunotherapy and chemotherapy applied to patients with non-small-cell metastatic lung cancer [93] . As shown by the estimated survival curves, outcomes are better with the immunotherapy...................................... 6 2.2 Example clean image of 7 randomly generated lesions............... 11 2.3 Demonstration of noise-adding procedure to images of synthetic lesions. The process is done for each color, red and green, individually, and the results are summed together. The process is shown for the green lesions in the above image. 13 2.4 Schematic of outcome aware object detection network used on synthetic lesion data. In our implementation, the convolutional encoder, which maps I → U , co

FAQs

What kinds of biological data does the dissertation focus on?

It focuses on large biological datasets such as medical imaging and single-cell level genomic data, using machine learning to extract useful information.

How is machine learning used for metastatic cancer prognosis?

The work uses deep learning to predict lesion-level risks from CT images, and uses these risks to demonstrate alignment with clinically known indicators of risk.

How are DNA shape effects analyzed in the transcription factor binding model?

The dissertation applies attribution-based interpretation methods, including DeepLIFT and TF-MoDISco, to assess how sequence and shape features modulate transcription factor binding.

Aug 3, 2026

Uploaded by Maeve