Cell Maps VNN (Cell Maps Visible Neural Network) Tool
Cell Maps VNN builds interpretable neural networks from biological hierarchies to predict cellular responses to drugs—while keeping every step transparent, reproducible and FAIR.
It enables the creation, training, and application of neural network models whose architecture mirrors a biological hierarchy. This makes the model’s internal structure both visible and interpretable, helping researchers connect predictions to biological context.
A hierarchy in HCX format—such as one generated by the cellmaps_generate_hierarchy tool—defines the structure of the visible neural network.
The tool produces separate output directories for training, prediction, and annotation. Results in these directories are stored and registered as Research Object Crates (RO-Crate) using the FAIRSCAPE-cli framework.
Overview of Cell Maps VNN Flow
Free software: MIT license
Source code: https://github.com/idekerlab/cellmaps_vnn