The capabilities of large language models (LLMs) have expanded beyond natural language processing to scientific prediction tasks, including molecular property prediction. However, their effectiveness in in-context learning remains ambiguous, particularly given the potential for training data contamination in widely used benchmarks. This paper investigates whether LLMs perform genuine in-context regression on molecular properties or rely primarily on memorized values. Furthermore, we analyze the interplay between pre-trained knowledge and in-context information through a series of progressively blinded experiments. We evaluate nine LLM variants across three families (GPT-4.1, GPT-5, Gemini 2.5) on three MoleculeNet datasets (Delaney solubility, Lipophilicity, QM7 atomization energy) using a systematic blinding approach that iteratively reduces available information. Complementing this, we utilize varying in-context sample sizes (0-, 60-, and 1000-shot) as an additional control for information access. This work provides a principled framework for evaluating molecular property prediction under controlled information access, addressing concerns regarding memorization and exposing conflicts between pre-trained knowledge and in-context information.
@article{arxiv.2603.25857,
title = {In-Context Molecular Property Prediction with LLMs: A Blinding Study on Memorization and Knowledge Conflicts},
author = {Matthias Busch and Marius Tacke and Sviatlana V. Lamaka and Mikhail L. Zheludkevich and Christian J. Cyron and Christian Feiler and Roland C. Aydin},
journal= {arXiv preprint arXiv:2603.25857},
year = {2026}
}