The main objective of the Tools group is to create tools to enable 1) the construction and interrogation of multimodal cell maps, which can serve as a foundation for structural and functional genomics (Figure 1), and 2) the use of these maps toward interpretable genotype-phenotype learning.
Figure 1. Overview of end-to-end data integration to produce multi-modal cell maps. From Schaffer et al. (2025) Nature
To this end, the Tools group is focused on:
- Creating a Cell Mapping Toolkit (Figure 2) for assembling multi-scale spatiotemporal maps of human cell architecture through integration of ‘omics data, including protein immunofluorescence images, protein biophysical interactions, and genetic perturbation data. Building on the success of the MuSIC v1.0 [Qin et al. (2021) Nature] and the U2OS MuSIC map [Schaffer et al. (2025) Nature] projects, these tools will be used to construct cell maps relevant to various disease contexts including cancer, neurodevelopmental, and cardiac disorders. We will leverage existing databases and tools, such as NDEx and Cytoscape web, to persist and render the generated maps.
Figure 2. Overview of Cell Mapping Pipeline
- Assembling a toolbox for the design, training, and execution of ‘visible’ neural network (VNN) models for use in genotype-phenotype predictions. For example, the VNN Tool (Figure 3) enables the creation, training, and usage of an interpretable neural network-based models that predict cell response to a drug Users of the toolbox will be able to build VNN models with structures based on multi-scale maps, enabling the interpretation of model predictions in biomedical applications. Training of the models in our pilot studies will use a variety of genotype-phenotype data, either generated by the Data Acquisition module or derived from public resources.
Figure 3. Overview of VNN Tool Pipeline
