When a Cell paper on April 23, 2025, proposed a previously unrecognized transcriptional role for PHGDH in amyloid pathology of Alzheimer's Disease, the UAB Systems Pharmacology AI Research Center (SPARC) treated it as a relay handoff rather than a finish line.
Led by Sixue Zhang, Ph. D., associate director of SPARC, and Jake Y. Chen, Ph.D., director of SPARC, the team used its systems-pharmacology and generative-AI pipeline—built around GeneTerrain Knowledge Maps (GTKM)—to pressure-test the mechanism as a real-time case study. In roughly 90 days, SPARC advanced from reading the paper to filing a U.S. provisional patent on a first batch of small-molecule candidates. This demonstrates how a decentralized model can rapidly translate open science into actionable drug discovery. Within these 90 days, the team at SPARC developed a collection of multi-scale, multi-modality deep learning AI models that not only aided the discovery of novel PHGDH inhibitors but could also serve as foundation models for other AI drug discovery projects.
The center's approach downplays any single target and focuses instead on the process: integrating a fresh mechanism into GTKM to understand the network context, generating families of CNS-suitable chemotypes, and triaging them with multi-objective AI for potency, brain penetration, and early safety flags.
"Our aim wasn't to claim a discovery; it was to show the relay works," said Chen. "A strong finding appears in the literature, and our systems-pharmacology AI can turn it into make-ready molecules fast—then invite the community to help validate them." That community invitation is by design.
In a recent Drug Discovery AI Talk, Chen articulated the philosophy behind the effort. "AI empowers startups, academic labs, and smaller organizations to drive therapeutic innovation," he said, adding that "agentic AI systems can automate experimental workflows"—two shifts that let groups like SPARC move early discovery forward without waiting on legacy bottlenecks.
Here, the decentralized relay flows from publication to design to bench researchers publish a mechanism; SPARC picks it up, validates context, and generates compounds; and partners across academia, startups, and industry run the experiments that determine which candidates advance. In this Alzheimer's case study, SPARC's provisional filing documents 18 synthetically accessible compounds prioritized for follow-up assays.
"I ran the end-to-end analysis—link the literature to GTKM, generate and down-select compounds, and clear early ADME/safety flags—so collaborators receive a short list worth making," said Fuad Al Abir, a doctoral student in Chen's lab who led the in-silico triage.
The continued design and development of those analogs will also leverage Zhang's expertise, who has rich experience in computer and AI-aided drug discovery for Alzheimer's Disease (see his latest publication on a novel drug candidate showing in vivo therapeutic efficacy for Alzheimer's Disease).
The team is now seeking partners to quickly evaluate the 18 compounds across enzymatic and cellular assays, ADMET, and early in vivo models, understanding that success would trigger company formation and capital to drive IND-enabling work and accelerate a path to clinical trials.
SPARC frames this work as a template for decentralized drug discovery across chronic diseases: treat each high-quality mechanism paper as a starting gun, run the AI-enabled relay to drug-like matter, and share the baton with experimentalists. It gives due credit to the Cell authors for the mechanistic insight, positions SPARC as the systems-AI translator, and invites rapid, distributed validation.