Research map / Probabilistic and scientific modeling / Dynamical systems

Position: Biology is the Challenge Physics-Informed ML Needs to Evolve

NeurIPS 2025 · Probabilistic and scientific modeling › Dynamical systems · approach: Stochastic population dynamics · 1 citations (OpenAlex)

Read the paperCheck your idea against 108,967 papers

AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Position paper arguing that biological dynamical-systems modeling demands an evolved PIML, "Biology-Informed ML" (BIML), built on four pillars (uncertainty quantification, contextualization, constrained latent structure inference, scalability) with foundation models/LLMs as an integrative layer, plus benchmark and community recommendations.

Problem

Physics-informed ML assumes well-known governing equations, full observability, low noise, and low dimensionality, which rarely hold in biological dynamical systems, and there are no principled biology-tailored benchmarks to evaluate PIML for them.

Approach

Reframe PIML for biology as BIML: retain mechanistic structure but operate under softer, probabilistic, multi-source, context-dependent priors with explicit uncertainty, constrained latent variables, and scalable inference, with LLMs/foundation models supplying priors and hypotheses.

Key results

• Bibliometric analysis reports a marked rise of physics-informed ML terminology in biomedical literature from 2015 to 2024.
• ODEBench includes 63 systems, of which only 12 have dimensionality D in {3,4} and only 2 are biologically motivated; LLM-SRBench has 28 biology-related examples, all one-dimensional.

Surprising findings

• On several perturbation-response tasks, simple baselines still outperform LLM-based methods.

Limitations

• No experiments or method implementation; the four pillars are proposals without full methodological validation.
• LLMs hallucinate in underspecified biological settings, so factual reliability is a concern.

Experimental setup

Builds on: physics-informed machine learning SINDy neural ODEs ODEFormer

Closest papers by meaning

• Physics-Guided Foundation Model for Scientific Discovery: An Application to Aquatic Science (AAAI 2025 · similarity 0.85)
• FunctionEvolve: Structure-Guided Symbolic Regression with LLMs (NeurIPS 2026 · similarity 0.83)
• Contextualizing biological perturbation experiments through language (ICLR 2025 · similarity 0.83)
• Prediction-Augmented Trees for Reliable Statistical Inference (NeurIPS 2026 · similarity 0.83)
• Discovering Mechanistic Models of Neural Activity: System Identification in an in Silico Zebrafish (NeurIPS 2026 · similarity 0.83)
• From Mechanistic Interpretability to Mechanistic Biology: Training, Evaluating, and Interpreting Sparse Autoencoders on Protein Language Models (ICML 2025 · similarity 0.83)