Try it here: https://github.com/dchaley/deepcell-imaging
Follow the README instructions to submit Google Batch jobs, and use the public data provided to start. Happy Segmenting 🫡
About
DeepCell is an AI tool that uses TensorFlow to predict cellular segmentation, aka drawing circles around cells on a microscope image. This is a key step in bioinformatics cancer research and patient treatment. Currently, DeepCell is hard to run at scale for a typical bioinformatics team. The DeepCell team (the Van Valen Lab at CalTech) provides a Kubernetes deployment, however K8s is inaccessible to most bioinformaticians who don't have DevOps or cloud practitioners on-hand. This work seeks to provide tools & guidance to scale DeepCell on Google Cloud, with minimal infrastructure knowledge required. Along the way I've been working with the Google Batch team getting their recommendations, and testing upcoming features. Here is a post I coauthored with Lynn Langit & Weihao Ge, using Batch Node Pools and DeepCell: https://medium.com/@dchaley/running-deepcell-on-google-batch-with-node-pools-8fe7d1c121a2 The current project work is testing in our target environment (where we don't have admin privileges). We've successfully run test workloads and need to productionize with more parameters, larger inputs, etc.
