Predicting Solvent-Induced Changes in Fluorescence Quantum Yield: Direct ClassificationVersus Regression

Changing the solvent can alter a molecule’s fluorescence quantum yield. This study compares two structure-based methods for predicting such changes: direct classification and prediction of two quantum yields followed by subtraction. From FluoDB-Lite, we retained 12,157 measurements for 2,712 molecules, resulting in 29,550 within-paper solvent comparisons. We categorized changes as reduced, approximately unchanged, or enhanced using a provisional absolute tolerance of 0.10. Molecules were divided into training, validation, and test sets without identity overlap. Validation identified a histogram gradient-boosting classifier and a Random Forest regressor as optimal models. On 407 test molecules, molecule-weighted macro-F1 scores were 0.577 and 0.569, respectively, compared with 0.248 for the majority baseline. The paired bootstrap interval for their difference included zero. The mean absolute error for individual quantum-yield prediction was 0.185. Small external checks and a design screen demonstrate the use and limitations of these approaches. Both methods capture useful information, but neither reliably improves experimental results for new compounds.

Authors

Institutions

Publication Details

Journal
ChemRxiv
Published
2026-11-17
DOI
https://doi.org/10.26434/chemrxiv.15012408/v1
Primary Topic
Computational Drug Discovery Methods
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Predicting Solvent-Induced Changes in Fluorescence Quantum Yield: Direct ClassificationVersus Regression

Neethu Anand, Jibin Joseph
ChemRxiv
Computational Drug Discovery Methods
preprint

Predicting Solvent-Induced Changes in Fluorescence Quantum Yield: Direct ClassificationVersus Regression

Neethu Anand, Jibin Joseph
preprint en

Abstract

Changing the solvent can alter a molecule’s fluorescence quantum yield. This study compares two structure-based methods for predicting such changes: direct classification and prediction of two quantum yields followed by subtraction. From FluoDB-Lite, we retained 12,157 measurements for 2,712 molecules, resulting in 29,550 within-paper solvent comparisons. We categorized changes as reduced, approximately unchanged, or enhanced using a provisional absolute tolerance of 0.10. Molecules were divided into training, validation, and test sets without identity overlap. Validation identified a histogram gradient-boosting classifier and a Random Forest regressor as optimal models. On 407 test molecules, molecule-weighted macro-F1 scores were 0.577 and 0.569, respectively, compared with 0.248 for the majority baseline. The paired bootstrap interval for their difference included zero. The mean absolute error for individual quantum-yield prediction was 0.185. Small external checks and a design screen demonstrate the use and limitations of these approaches. Both methods capture useful information, but neither reliably improves experimental results for new compounds.

ChemRxiv
Pohang University of Science and Technology (KR), The University of Texas at Austin (US)
Computational Drug Discovery Methods
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon