Recovering evolutionary and ecological history from incomplete records of past and present life

My research combines computational methods (including phylogenetics and image analysis), evolutionary theory, and paleoecology to reconstruct biodiversity dynamics, trait evolution, and environmental change from the incomplete evidence organisms leave behind, from molecular sequences to the morphology of fossilized plants.

Training & research questions

  • 2019 · Université Laval, Québec

    M.Sc. Quaternary aquatic paleoecology in the Canadian Arctic

    Laboratoire de paléoécologie aquatique · Centre d'études nordiques · Département de géographie

    I began my training in physical geography and Quaternary paleoecology through research on the freshwater ecosystems of Nunavut and Nunavik. For my M.Sc. at the Centre d'études nordiques, I reconstructed the postglacial history of the Fury and Hecla Strait region in Nunavut, in the Canadian Arctic, using lake sediments. We analyzed marine and freshwater diatoms together with geochemical records preserved in the sediments, which allowed us to establish the first chronology of regional deglaciation. This work also helped clarify the history of marine inundation, the onset of glacial isostatic adjustment, and associated changes in relative sea level. It also established the timing of the postglacial marine reconnection between Atlantic and Pacific waters, with important implications for the exchange of marine flora and fauna between the two oceans. I have remained active in Arctic research through field expeditions and collaborative projects with teams at the University of Tokyo and the National Institute of Polar Research in Japan, as well as colleagues at Université Laval and the Centre d'études nordiques, with further work and collaborations continuing today.

  • 2024 · University of Illinois Urbana-Champaign

    Ph.D. Deep learning for pollen morphology and phylogenetic inference

    Punyasena Lab · Department of Plant Biology · School of Integrative Biology

    My doctoral research developed evolutionarily aware neural networks that extract phylogenetically and ecologically meaningful information from pollen morphology. Combining superresolution imaging with deep learning, I built methods that bridge molecular phylogenetics and morphological analysis, making it possible to place unknown or extinct plant morphotypes onto reference phylogenies from morphology alone, detect cryptic speciation and extinction, and trace adaptations that arose in plant populations over evolutionary time. The same imaging and image-processing methods allowed me to reconstruct changes in grassland diversity and photosynthetic composition over tens of thousands of years in East Africa. A major focus of this work was the evolution of the Podocarpaceae, one of the most remarkable and diverse conifer families. This includes the diversification of the clade, its migrations as continents drifted and the planet's geography was reshaped, and the adaptations that emerged under changing environmental conditions. Reconstructing that history depends on robust molecular phylogenies, a well-resolved fossil record, and, most importantly, tools that can connect the two and rigorously analyze very subtle phenotypic variation. This work grew from a close collaboration with the Smithsonian Tropical Research Institute (STRI), which continues today.

  • 2025 – present · Smithsonian Institution

    Postdoctoral Fellow: Machine learning and biodiversity informatics for macroecology

    Chief Data and AI Office, Office of Digital & Innovation · Washington, D.C.

    Currently, at the Smithsonian, I work at the interface of ecology, evolution, and biodiversity informatics. A few questions sparked many of the ideas behind this project: why are conifers so much more widespread in North America than in South America, under what conditions do South American conifers succeed beyond expectation, and where along this range are conifer distributions and their traits driven by stochastic versus deterministic processes? To address these, I develop new methods, principally using large language models, to extract anatomical, physiological, and genomic characteristics from conifer species worldwide, and integrate them with forest-plot data to model trait-environment relationships. This work builds new ways to extract trait and diversity data and reveals hidden correlations among anatomical and physiological traits, genomic architecture, and the environment. Most importantly, it lets us address open questions in ecology and evolution: the role of stochasticity versus determinism in shaping where species and their traits occur, and how much particular traits matter for a species' ability to establish in a given environment. The project focuses largely on conifers worldwide, but the methods generalize across the tree of life, with ongoing collaborations extending this work to other clades.

  • From January 2027 · American University of Beirut

    Assistant Professor, Bioinformatics & Computational Biology

    Department of Biology · Faculty of Arts and Sciences

    In my incoming faculty position, I will build a research program that extends many of the methods I have developed to the biodiversity of Lebanon and the eastern Mediterranean. The program will focus primarily on plant diversity, while expanding to fungi and animal groups through collaborations. I aim to understand why endemic and regional species occur under particular environmental conditions, how their traits have evolved, what evolutionary processes have shaped those traits, and how lineages are related, using molecular data to resolve their phylogenetic relationships. More broadly, I plan to combine molecular, phenotypic, and environmental data to reconstruct trait evolution and test how environmental pressures, genomic change, and lineage history have shaped biodiversity across the region. I also aim to develop new methods, including deep-learning approaches, for estimating phylogenies, quantifying phylogenetic uncertainty, and reconstructing ancestral sequences and traits.

Field photo Field photo Field photo Field photo Field photo Field photo Field photo

Evolution and adaptation across deep time

I use trait analysis, molecular phylogenetics, and simulations of character evolution to understand how particular traits emerged and why they persisted. Conifers have been a recurring focus of this work, from testing whether pollen-wall thickness in living podocarps is associated with solar radiation, raising the hypothesis that thicker walls may have evolved gradually under higher UV-B exposure to better protect the DNA within the pollen grain, to the puzzling question of why fleshy cones appear to have evolved repeatedly in the group. By studying traits such as these, I can ask whether they arose primarily through adaptation to changing environments, were shaped by inherited developmental or phylogenetic constraints, or reflect a more complex interplay among selection, lineage history, and chance events. I address these questions by combining phylogenetic comparative methods with anatomical, physiological, and genomic data.

Fleshy cones in Podocarpaceae
Conifer seed cones vary widely in form, and some lineages have independently evolved a fleshy structure that makes the cone resemble a fruit. Improved seed dispersal likely plays a role, perhaps linked to the diversification of birds and other vertebrate dispersers, but the full picture is more complex. Answering it means examining cone development and architecture, natural selection, the fossil record of conifer families, and the underlying genomic changes that may have facilitated the diversification of the trait, as well as running character evolution simulations that test how it could have arisen.
Reconstructed grassland diversity and C4 composition at Lake Rutundu over 25,000 years
Reconstructed changes in Poaceae diversity and C4 composition around Lake Rutundu on Mount Kenya over the past 25,000 years. Grass pollen is abundant in sediment records but notoriously difficult to distinguish because grains from different species often look nearly identical under the microscope, while more diagnostic plant remains are usually much rarer. Using superresolution pollen images and our deep-learning model, we were able for the first time to reconstruct both grass diversity and photosynthetic pathway composition from morphology alone, without relying on ancient DNA, chemical analyses, or the identification of larger plant remains. We then tested how those changes were associated with climate and fire from the last ice age into the Holocene.

Ecological and environmental change

A key theme in my research is how ecosystems respond to environmental change over long timescales, across environments ranging from the High Arctic to tropical grasslands. I use biological remains preserved in sediments, such as phytoplankton and pollen grains, together with environmental DNA and geochemistry, to reconstruct changes in biodiversity, vegetation, community composition, and physiological composition. One interesting focus of my research, for example, has been to reconstruct shifts in the relative abundance of plants that use different photosynthetic pathways (C3 or C4), two alternative strategies for fixing carbon, and examine how those shifts reflect changes in local environmental conditions. Across these systems, relatively small biological shifts can reveal much larger changes in the surrounding environment and climate.

New computational methods for unresolved questions in evolution and ecology

Both types of questions depend on measuring variation that is often too subtle to capture directly. I develop statistical and computational methods, including signal processing and deep learning applied to superresolution imaging, that turn complex structures such as pollen, leaves, or cones into quantitative representations of fine phenotypic variation and integrate these patterns with molecular and phylogenetic information. These representations can then be used in very different ways: to infer phylogenetic placement, detect cryptic speciation and extinction events, reconstruct adaptations over evolutionary timescales, or quantify diversity in ecological communities where species cannot be distinguished reliably by eye. A particularly powerful aspect of this approach is that the same techniques and algorithms can be adapted to questions in both evolution and ecology, even though they operate over very different timescales and rely on different biological assumptions.

Superresolution images of pollen grains from Croton hirtus, Mabea occidentalis, and Agropyron repens (Luke Mander and Mayandi Sivaguru)
Superresolution images of pollen from Croton hirtus, Mabea occidentalis, and Agropyron repens. Images by Luke Mander and Mayandi Sivaguru, Carl R. Woese Institute for Genomic Biology.

Interests

Topics I work in