SciML Receives Chan Zuckerberg Institute Funding: Spatial SSAs, Identifiability, and Compile Times

The NumFOCUS-sponsored SciML organization is pleased to announce that we have received a Chan Zuckerberg Initiative Essential Open Source Software for Science grant as part of Cycle 4! As a leading organization developing the mathematical techniques being used in software for clinical pharmacology by top firms like Moderna, demonstrating 175x accelerations in preclinical analysis by firms like Pfizer, and an core part of the Heta language tools from InSysBio, SciML and the Julia programming language have become a standard in the field of pharmacology. This grant is to accelerate the open source development in ways that improve the ecosystem for this essential industry.

As part of this grant, Professor Samuel Isaacson and his lab will work to expand SciML's biological and chemical modeling functionality to enable the study of spatially distributed systems. University of Washington Ph.D. student Vasily Ilin has already begun a first effort to add spatial stochastic simulation tooling to DiffEqJump as part of a GSoC with Dr. Isaacson. This grant will contribute to SciML's ability to make a sustained, multiyear effort to advance these starter pieces to a full-fledged spatial modeling ecosystem, enabling the investigation of cell signaling and the internal effects of drugs on cellular processes. As a part of the SciML ecosystem, it will focus on the scalability and performance required to handle the largest models, along with making sure to compose with the rest of the ecosystem.

Additionally, we are pleased to announce that as part of this work we will be funding Dr. Tim Holy to help improve the compile times for these essential packages. We have already started to investigate and solve some of the main compile time issues, bringing the compile times of widely used tools from 22 to 3 seconds . With Tim onboard, we plan to comb through a large portion of the Julia ecosystem used throughout pharmaceutical modeling and simulation to improve the general usability of the ecosystem. Additionally, we hope to document this work through issues and workshops to make these changes replicable by other Julia package organizations, with a goal of changing the general developer mindset to include compile times as a priority.

Lastly, we plan to integrate tools for identifiability analysis directly into the analytics workflows. This will accelerate the life of scientists by making it quick and easy to answer questions like "are there multiple sets of parameters which equally fit the model?". We have already begun to integrate StructuralIdentifiability.jl into the SciML ecosystem, and with this work we will improve the methods, documentation, and tutorials so that structural and practical identifiability can be easy as running just a few functions on a standard ODESystem and ODEProblem.

We thank CZI for this opportunity and hope we can make a clear difference in the productivity of pharmaceutical science.