ALIGNing Learned Representations: Domain-General Models Triangulate Physicochemical Interactions in Analytical Separations
9 affiliations
- 1Department of Chemistry
- 2University of Waterloo
- 3University of British Columbia
- 4Department of Electrical and Computer Engineering
- 5Enveda Therapeutics, Inc.
- 6Matterworks Inc.
- 7WaterFEL Free Electron Laser Laboratory
- 8Centre for Eye and Vision Research
- 9Hong Kong Science Park
Abstract
Despite rapid progress, foundation models for chemistry have yet to consistently deliver broadly transferable, experiment-grounded representations for structure-to-measurement tasks in practice. This reflects both the scarcity and fragmentation of experimental datasets and the weak alignment between common molecular representations (e.g., SMILES and graphs) and measured chemical behaviour. Here, we formalize domain-general models: pretrained models constrained across multiple measurement contexts within a physically coherent domain to learn transferable chemical representations that capture how molecular structure gives rise to experimental observables. Using liquid chromatography as a testbed, we develop ALIGN and show that it outperforms conventional baselines, including foundation models and transfer learning approaches, across diverse chromatographic contexts while retaining performance in data-limited and chemically out-of-distribution settings. By training across multiple chromatographic methods, ALIGN triangulates an underlying physicochemical interaction space, with each method providing a complementary projection that constrains the learned representation. We show that ALIGN learns embeddings consistent with physicochemical properties and chromatographic principles and provide evidence for a mechanism underlying improved method-specific adaptation. These representations transfer to related analytical modalities, including GC, CCS, and DMS, suggesting that these measurements encode overlapping physicochemical relationships. Together, these results demonstrate the value of physically grounded representation learning and suggest that domain-general approaches may extend to other fragmented scientific domains.
Related
- Ferritin-Based Nanocarriers: Analytical Methods for Physicochemical Characterization
- Oligoglycine-functionalized gold nanoparticles: physicochemical characterisation and serum protein interactions
- Powder-to-membrane transformation affording 2D/3D COF membranes for molecular separations
- Confined Mg²⁺-Directed Interfacial Evolution of Primary Structural Units of Mg-MOF-74 on Platelet Clinoptilolite for CO₂/CH₄, CH₄/N₂, and CO₂/N₂ Separations.
- Modeling Transport and Mechanics in Anion Exchange Membranes to Optimize Continuous CO2 Separations
BibTeX
@misc{stienstra2026aligning,
title = {{ALIGNing Learned Representations: Domain-General Models Triangulate Physicochemical Interactions in Analytical Separations}},
author = {Cailum M. K. Stienstra and Yen-Nhi Hoang and Chen Zhao and Alisa Hui and Emir Nazdrajić and Pelle Simpson and Aurelie Roux and August Allen and Michael Widrich and Daniel Domingo-Fernández and W. Scott Hopkins},
howpublished = {ChemRxiv},
year = {2026},
month = sep,
doi = {10.26434/chemrxiv.15007288/v3},
url = {https://chemrxiv.org/doi/full/10.26434/chemrxiv.15007288/v3},
note = {CC BY},
}
RIS
TY - UNPB
TI - ALIGNing Learned Representations: Domain-General Models Triangulate Physicochemical Interactions in Analytical Separations
AU - Cailum M. K. Stienstra
AU - Yen-Nhi Hoang
AU - Chen Zhao
AU - Alisa Hui
AU - Emir Nazdrajić
AU - Pelle Simpson
AU - Aurelie Roux
AU - August Allen
AU - Michael Widrich
AU - Daniel Domingo-Fernández
AU - W. Scott Hopkins
PY - 2026
DA - 2026/09/01
PB - ChemRxiv
DO - 10.26434/chemrxiv.15007288/v3
UR - https://chemrxiv.org/doi/full/10.26434/chemrxiv.15007288/v3
AB - Despite rapid progress, foundation models for chemistry have yet to consistently deliver broadly transferable, experiment-grounded representations for structure-to-measurement tasks in practice. This reflects both the scarcity and fragmentation of experimental datasets and the weak alignment between common molecular representations (e.g., SMILES and graphs) and measured chemical behaviour. Here, we formalize domain-general models: pretrained models constrained across multiple measurement contexts within a physically coherent domain to learn transferable chemical representations that capture how molecular structure gives rise to experimental observables. Using liquid chromatography as a testbed, we develop ALIGN and show that it outperforms conventional baselines, including foundation models and transfer learning approaches, across diverse chromatographic contexts while retaining performance in data-limited and chemically out-of-distribution settings. By training across multiple chromatographic methods, ALIGN triangulates an underlying physicochemical interaction space, with each method providing a complementary projection that constrains the learned representation. We show that ALIGN learns embeddings consistent with physicochemical properties and chromatographic principles and provide evidence for a mechanism underlying improved method-specific adaptation. These representations transfer to related analytical modalities, including GC, CCS, and DMS, suggesting that these measurements encode overlapping physicochemical relationships. Together, these results demonstrate the value of physically grounded representation learning and suggest that domain-general approaches may extend to other fragmented scientific domains.
ER -