New preprint: DASyR-LLM — domain-aware symbolic regression with LLMs for kinetic model discovery

Mon, 10 Aug 2026 00:00:00 +0000

How can we discover the equations that govern a chemical or biological process — directly from noisy experimental data, and without wasting experiments? 🤖

Our new preprint, DASyR-LLM: Domain-Aware Symbolic Regression with LLMs for Kinetic Model Discovery, by Roberto Aliaga Medina, Paulina Quintanilla and Antonio del Rio Chanona, tackles exactly this.

The idea

DASyR-LLM closes the loop between experimentation and modelling. Starting from noisy experimental data, it runs an iterative, LLM-guided symbolic regression cycle:

Run a new experiment, chosen via model-based design of experiments so each one is maximally informative;
Symbolic regression proposes candidate kinetic models as interpretable equations;
an LLM critiques those candidates using scientific domain knowledge — for example, recognising that a rate law should use Monod-type saturation terms;
the LLM proposes new candidate models, which feed the next round.

Because the LLM injects domain awareness, the search is steered toward models that are not only accurate but also physically meaningful.

The result

Across the case studies, DASyR-LLM recovers interpretable kinetic models with equivalent predictive accuracy (R² > 0.98) while needing up to 79% fewer experiments than conventional symbolic regression.

📄 Preprint: arXiv:2608.05120
💻 Code: Zenodo record
🔗 View the announcement on LinkedIn

Huge congratulations to Roberto Aliaga Medina, who led this work while finishing his MSc in Chile! 🎉

Model Discovery | LOOPS Research Group

New preprint: DASyR-LLM — domain-aware symbolic regression with LLMs for kinetic model discovery

The idea

The result

Read more