Lapworth Lecture - From Molecules to Models: Integrating Experimental Taphonomy and AI to Resurrect Ancient Life

How do we know what an extinct animal looked like, how it behaved, or how it relates to the creatures alive today? The fossil record is far richer than bones alone. Exceptionally preserved fossils can retain traces of the original pigments that once coloured skin, scales and feathers, while three-dimensional preservation can lock away the internal anatomy that records an animal’s form, function, growth and evolutionary relationships. The difficulty is that these signals are faint, fragmentary and easily misread, and recovering them reliably across very different scales of preservation demands both careful experiments and new technology.

This lecture follows two research programmes that, between them, reach from individual molecules to whole-body models of extinct vertebrates.

The first asks a deceptively simple question: can we reconstruct the colours of animals that died many millions of years ago? Colour-bearing structures do survive in some fossils, but burial, heat and chemical change alter them over deep time, so what we measure now is not necessarily what was originally there. By recreating these processes in the laboratory, through controlled decay and maturation experiments, we can build frameworks that separate a genuine biological signal from the imprint of fossilisation itself. Used with care, this approach allows us to infer not only ancient colouration but also clues to ecology and behaviour, from camouflage and signalling to choice of habitat.

The second programme tackles a very different bottleneck. Computed tomography (CT) scanning now lets us look inside fossils without damaging them, but converting a stack of scan images into a usable three-dimensional model is slow and painstaking, and is often the rate-limiting step in research. Using Jurassic vertebrate fossils from the Isle of Skye, Scotland, this work develops automated deep learning pipelines that segment CT data far faster than is possible by hand, opening up anatomical information that is critical for understanding the deep-time origins of modern vertebrate groups, including the earliest mammals and their close relatives.

Both stories share a common thread, and a common caution. These methods can recover remarkable detail, but each carries its own assumptions and limits, and a central part of this research is being honest about what such data can, and cannot, reliably tell us. Taken together, experimental taphonomy, geochemistry and artificial intelligence are changing how we read the fossil record, allowing us to reconstruct the appearance, biology and life histories of ancient organisms with a resolution and rigour that would have been out of reach a generation ago.

The lecture is free to attend and open to everyone, and is available to watch in-person on campus or on Zoom.

About Arindam

Dr Arindam Roy is a Marie Skłodowska-Curie Actions postdoctoral fellow at the University of Birmingham, based in the School of Geography, Earth and Environmental Sciences. He takes a deeply interdisciplinary approach to palaeobiology, developing new experimental and computational methods to extract biologically meaningful information from the fossil record. His work bridges experimental taphonomy, biogeochemistry and artificial intelligence, with a particular focus on the preservation of soft tissues and pigments, and on deep learning-based automated segmentation of fossil CT data.

His research spans fossil colour reconstruction, taphonomy, functional morphology and CT-based anatomy, and he collaborates with museums and research institutions in the UK and internationally. Alongside his research, he maintains an active public engagement record that includes film consultancy, media coverage and outreach.