In both embryonic development and disease, the same genetic mutation can lead to highly variable outcomes in different individuals. Dr. Lammers aims to shed light on the drivers of this nongenetic variability using the developing zebrafish embryo as a model system. By combining fluorescence microscopy and single-cell sequencing, he will test whether subtle differences in gene expression within individual cells can explain why some embryos with a given genetic mutation survive to adulthood, while others perish within the first 24 hours of their development. His findings will provide a quantitative foundation for understanding the genetic and molecular basis of cancer outcomes in human patients where, for instance, tumors with the same underlying mutations often exhibit dramatically different disease courses.

Dr. Lammers will train Variational Autoencoders to learn low-dimensional latent space representations of whole-embryo transcriptomes and grayscale images depicting embryonic morphology. He will then train a third neural network to translate from transcriptional latent space to morphological latent space. Together, these three networks will comprise a new computational method, morphSeq, that takes single-cell transcriptomes of mutant and wildtype embryos as input and produces predictions for corresponding embryo morphologies as its output.