Decoding evolutionary and ecological information from pollen morphology: deep learning for phylogenetic and environmental reconstructions

Adaime, Marc-Elie

Permalink

https://hdl.handle.net/2142/127277 Copy

Description

Title

Decoding evolutionary and ecological information from pollen morphology: deep learning for phylogenetic and environmental reconstructions

Author(s)

Adaime, Marc-Elie

Issue Date

2024-12-06

Director of Research (if dissertation) or Advisor (if thesis)

Punyasena, Surangi W

Doctoral Committee Chair(s)

Heath, Katy D

Committee Member(s)

• Kong, Shu
• Tan, Milton
• Leslie, Andrew B

Department of Study

Plant Biology

Discipline

Plant Biology

Degree Granting Institution

University of Illinois at Urbana-Champaign

Degree Name

Ph.D.

Degree Level

Dissertation

Keyword(s)

• Pollen
• Evolution
• Deep Learning
• Phylogenetics
• Paleontology
• Image Processing
• Computational Biology
• Paleoecology

Abstract

The fossil pollen record provides a valuable archive for exploring plant evolutionary history and ecological dynamics. Traditional palynology is limited by conserved pollen across several plant lineages, which restricts identification to higher taxonomic levels. Recent advances in imaging and machine learning have provided new insights, yet most focus on species identification rather than broader evolutionary and ecological questions. This dissertation introduces new methodologies integrating deep learning, image processing, and traditional analyses to advance palynology. Chapter 2 uses phylogenetically-informed neural networks to identify potentially extinct pollen types and place them within a reference phylogeny, revealing the wealth of evolutionary information encoded in pollen morphology. Chapter 3 quantifies Poaceae community diversity and distinguishes C3 from C4 grass pollen using CNN-derived features, revealing physiological adaptations and providing, for the first time, the ability to accurately estimate grass diversity and C3:C4 ratios in paleo-grasslands using pollen morphology alone. Chapter 4 reconstructs the evolutionary history of Podocarpus pollen morphology in relation to environmental variability, showing adaptive responses to temperature and solar radiation. Moreover, it emphasizes the importance of incorporating fossil data in reconstructing shifts in morphospace and ancestral states. Together, these chapters illustrate how abstract pollen morphological features derived from neural networks can be used to resolve evolutionary relationships, quantify community diversity, and reveal physiological and environmental adaptations in plants. This dissertation shows that deep learning, combined with traditional approaches, can decode complex morphological features embedded in pollen grains, addressing unexplored questions in ecology and evolution.

Graduation Semester

2024-12

Type of Resource

Thesis

Handle URL

https://hdl.handle.net/2142/127277

Copyright and License Information

Copyright 2024 Marc-Elie Adaime

Owning Collections