Work

Integrating Dynamical Systems Learning with Foundational Models: A Meta-Evolutionary AI Framework for Clinical Trials

Year: 2025

Type: article

Abstract: Artificial intelligence (AI) is evolving into an ecosystem of specialized “species.” This manuscript compares two: DeepSeek-V3, a Mixture of Experts LLM, and NetraAI, a dynamical-system-based framework designed for small, structured datasets in clinical trials. We formalize NetraAI’s contraction-based learning architecture, its ability to efficiently discover causal bundles of variables, and introduce a meta-evolutionary loop where a large language model acts as a strategist, orchestrating NetraAI’s subpopulation discovery process. Across datasets in schizophrenia, depression, and pancreatic cancer, NetraAI uncovers high-effect subgroups defined by 2–4 variables—Personas—that significantly outperform LLMs and other machine learning methods at this task. This work advances explainable, agentic AI architectures and proposes a new design philosophy for interpretable, modular systems for clinical trial enrichment and derisking.

Source: Zenodo (CERN European Organization for Nuclear Research)

DOI: 10.5281/zenodo.15645097

Authors: Geraci, Joseph, Qorri, Bessi, Cumbaa, Christian, Tsay, Mike, Leonczyk, Paul, Pani, Luca

Institutions: Centre for Addiction and Mental Health, Queen's University, University of California San Diego, University of Miami

Language: English

Topic: Artificial Intelligence in Healthcare and Education

Subfield: Health Informatics

Field: Medicine

Domain: Health Sciences

Open Access status: green

OpenAlex ID: W6911982376

FWCI:

Cites:

Cited by:

Related to: