David Van Valen

This paper develops a framework combining a protein language model and a screening assay to predict kinase-substrate interactions and map the human kinome.

CellTune is a software tool for cell classification in spatial proteomics that outperforms existing methods with high accuracy and resolution.

This paper introduces Cryosearch, a GPU-accelerated framework that uses Monte Carlo tree search to assemble protein complexes from intermediate-resolution cryo-EM maps.

This paper introduces phospho-PCA, a method to quantify kinase-substrate interactions, and applies it to study kinase variants, revealing mutations that affect activity and substrate specificity.

The paper introduces Kanzi, a flow-based tokenizer for protein structures that simplifies training and outperforms existing tokenizers on reconstruction metrics.

The paper introduces Nimbus, a deep learning model for predicting marker positivity in multiplexed imaging data using a pre-trained dataset, and shows its integration with clustering for cell subtype identification.

The paper develops tools for analyzing T cell behavior from live-cell imaging to improve T cell therapies by studying the effects of genetic modifications on T cell-cancer cell interactions.

This paper introduces DeepCell Types, a language-informed vision model for spatial proteomics cell phenotyping that generalizes across diverse datasets and marker panels.

Human-in-the-loop labeling of nuclear live-cell movies produced DynamicNuclearNet (over 2,621 divisions); Caliban, a tracker trained on it, beat four published trackers at division detection.

Polaris provides a deep learning-based pipeline for analyzing spatial transcriptomics data, improving spot detection and gene expression quantification.

Bounding boxes from a transformer detector let the Segment Anything Model segment cells automatically; one generalist model beats generalist Cellpose and matches inter-annotator agreement across mammalian, yeast and bacterial images.

The paper introduces Mesmer, a deep learning algorithm for cell segmentation in tissue images, achieving human-level performance and enabling automated feature extraction.

The paper constructs a spatiotemporal atlas of the maternal-fetal interface using multiomic data to show how gestational age influences immune and stromal cell interactions and spiral artery remodeling.

DeepCell Kiosk scales deep learning workflows for large imaging datasets, processing 106 images in ~5.5 hours at a low cost.