Albert Dominguez Mantes

Identifying spot-like structures in large and noisy microscopy images is a crucial step to produce high quality results in various life-science applications. Imaging-based spatial transcriptomics (iST) methods, in particular, critically depend on the precise detection of millions of transcripts in images with low signal-to-noise ratio. Despite advances in computer vision that have revolutionized many biological imaging tasks, currently adopted spot detection techniques are mostly still based on classical signal processing methods that are fragile and require tedious manual tuning per dataset. In this work, we introduce Spotiflow, a deep learning method that achieves subpixel-accurate localizations by formulating the spot detection task as a multiscale heatmap and stereographic flow regression problem. Spotiflow can be used for 2D images and 3D volumetric stacks and can be trained to generalize across different imaging conditions, tissue types and chemical preparations, while being substantially more time- and memory-e cient than existing methods. We show the eficacy of Spotiflow via extensive quantitative experiments on a variety of diverse datasets and demonstrate that the enhanced accuracy of Spotiflow leads to meaningful improvements in the biological insights obtained from iST and live imaging experiments. Spotiflow is available as an easy-to-use Python library as well as a napari plugin at https://github.com/weigertlab/spotiflow.

Characterizing the genetic structure of large cohorts has become increasingly important as genetic studies extend to massive, increasingly diverse biobanks. Popular methods decompose individual genomes into fractional cluster assignments with each cluster representing a vector of DNA variant frequencies. However, with rapidly increasing biobank sizes, these methods have become computationally intractable. Here we present Neural ADMIXTURE, a neural network autoencoder that follows the same modeling assumptions as the current standard algorithm, ADMIXTURE, while reducing the compute time by orders of magnitude surpassing even the fastest alternatives. One month of continuous compute using ADMIXTURE can be reduced to just hours with Neural ADMIXTURE. A multi-head approach allows Neural ADMIXTURE to offer even further acceleration by computing multiple cluster numbers in a single run. Furthermore, the models can be stored, allowing cluster assignment to be performed on new data in linear time without needing to share the training samples.