Learning a mechanical growth model of flower morphogenesis

Argyris Zardilis, Alexandra Budnikova, Henrik Jönsson

Track: Tiny Paper Track

Keywords: multi-modal data, unsupervised learning, biophysical model, plant morphogenesis

TL;DR: Learning a mechanical growth model of flower morphogenesis by learning a combined representation of morphology and transcription data and then refining to a more physical representation using mechanical information.

Abstract: Morphogenesis, the process by which organs take their shape, is a fundamental problem in biology and requires a complex combination of genetic, biochemical, and mechanical signals. The data that we collect for this process are usually multi-scale and multi-modal, typically image timeseries of tissue morphology and transcription data. As the data becomes larger and more diverse, the harder it becomes to gain meaningful understanding in terms of theoretical models. Here, and tackling this problem in the developing flower organ, we learn a mechanical model of growth starting from combined morphology and transcription data across the tissue by firstly learning and connecting independent representations for the two spaces and secondly refining the representation using the mechanical state of the tissue to learn a mechanical growth model. This approach can link multiple scales, in this case proposing the physical action of genes, and can act as a vehicle to accommodate diverse data and hypotheses in developmental biology of plants and beyond.

Attendance: Argyris Zardilis

Submission Number: 58