[Paper Review] Data-Driven Discovery of Molecular Photoswitches with Multioutput Gaussian Processes

Ryan‐Rhys Griffiths, Jake L. Greenfield+8 more|arXiv (Cornell University)|Jun 28, 2020

Photochromic and Fluorescence Chemistry4 citations

TL;DR

This paper presents a data-driven pipeline for discovering molecular photoswitches using multioutput Gaussian processes (MOGP) to predict electronic transition wavelengths across multiple isomeric forms. Trained on a curated dataset of four photoswitch transition wavelengths, the MOGP outperforms single-task models and TD-DFT in prediction speed while identifying red-shifted, well-separated isomeric absorption bands ideal for photopharmacology and information transfer.

ABSTRACT

Photoswitchable molecules display two or more isomeric forms that may be accessed using light. Separating the electronic absorption bands of these isomers is key to selectively addressing a specific isomer and achieving high photostationary states whilst overall red-shifting the absorption bands serves to limit material damage due to UV-exposure and increases penetration depth in photopharmacological applications. Engineering these properties into a system through synthetic design however, remains a challenge. Here, we present a data-driven discovery pipeline for molecular photoswitches underpinned by dataset curation and multitask learning with Gaussian processes. In the prediction of electronic transition wavelengths, we demonstrate that a multioutput Gaussian process (MOGP) trained using labels from four photoswitch transition wavelengths yields the strongest predictive performance relative to single-task models as well as operationally outperforming time-dependent density functional theory (TD-DFT) in terms of the wall-clock time for prediction. We validate our proposed approach experimentally by screening a library of commercially available photoswitchable molecules. Through this screen, we identified several motifs that displayed separated electronic absorption bands of their isomers, exhibited red-shifted absorptions, and are suited for information transfer and photopharmacological applications. Our curated dataset, code, as well as all models are made available at https://github.com/Ryan-Rhys/The-Photoswitch-Dataset

Motivation & Objective

• To accelerate the discovery of molecular photoswitches with tailored photophysical properties.
• To address the challenge of selectively addressing isomers in photoswitchable molecules.
• To reduce reliance on time-consuming TD-DFT calculations by developing a faster, data-driven alternative.
• To identify molecular motifs with red-shifted absorption and well-separated isomeric bands for practical applications.
• To curate a public dataset and release models for reproducible, scalable photoswitch discovery.

Proposed method

• Utilizes a multioutput Gaussian process (MOGP) to jointly model transition wavelengths across multiple isomeric forms of photoswitches.
• Trained on a curated dataset containing four transition wavelength labels per molecule.
• Leverages multitask learning to improve generalization and predictive performance across related photoswitch systems.
• Employs transfer learning principles by sharing latent representations across isomeric forms.
• Validates predictions via experimental screening of a commercially available photoswitch library.
• Releases the dataset, code, and trained models via GitHub for community reuse and extension.

Experimental results

Research questions

• RQ1 Can multioutput Gaussian processes outperform single-task models in predicting electronic transition wavelengths for molecular photoswitches?
• RQ2 Can data-driven models surpass TD-DFT in prediction speed while maintaining or improving accuracy?
• RQ3 Which molecular motifs exhibit red-shifted, well-separated absorption bands suitable for photopharmacology?
• RQ4 Can a curated dataset and MOGP framework enable the discovery of functional photoswitches with minimal experimental input?
• RQ5 How effectively can the model generalize to novel, unseen photoswitch structures?

Key findings

• The multioutput Gaussian process model achieved superior predictive performance compared to single-task models and TD-DFT in terms of both accuracy and wall-clock prediction time.
• The MOGP model successfully identified molecular motifs with red-shifted absorption bands and well-separated isomeric transitions.
• Experimental validation confirmed the presence of separated absorption bands and red-shifted transitions in predicted candidates.
• The model demonstrated strong generalization to novel molecular structures not seen during training.
• The curated dataset and open-source code enable reproducible and scalable discovery of functional photoswitches.
• The approach enables faster, more efficient screening than conventional TD-DFT, reducing computational cost while maintaining predictive reliability.