CDEF-Tg: Machine Learning-Based Descriptor–Embedding Fusion for Glass Transition Temperature Prediction in a Polyimide-Oriented Polymer Dataset
- 1Department of Computer Engineering, Faculty of Engineering and Architecture, Bingol University, 12000 Bingol, Türkiye
- 2Department of Computer Technologies, Bitlis Eren University, 13000 Bitlis, Türkiye
Abstract
Predicting the glass transition temperature (Tg) of polymers from structures remains difficult, and whether descriptors and molecular-language embeddings carry complementary information is unclear. CDEF-Tg (Compact Descriptor–Embedding Fusion for Glass Transition Temperature Prediction) concatenates 205 RDKit descriptors (RDKit205) with a frozen 384-dimensional ChemBERTa embedding (ChemBERTa384) and maps the 589-dimensional vector to Tg with an ExtraTreesRegressor. It was evaluated on 10,065 identity-resolved records of a polyimide-oriented dataset under three leakage-safe protocols: IID Group-Safe (IID, independent and identically distributed), Scaffold OOD (OOD, out-of-distribution), and Similarity OOD (T = 0.90). The model was selected on validation evidence and frozen before the final-test sets were opened. In the locked final test, CDEF-Tg reached a macro mean absolute error (MAE) of 48.84 °C, versus 50.17 °C (RDKit205 alone) and 54.33 °C (ChemBERTa384 alone). Group-aware bootstrap analyses separated CDEF-Tg from RDKit205 alone in MAE only under Similarity OOD and showed no MAE advantage over the comparator RDEF-Tg. Additional validation-only controls showed a lower macro MAE for a Morgan-fingerprint partner than for CDEF-Tg (49.11 versus 53.15 °C), and ChemBERTa384 lowered LightGBM error; the ExtraTrees ordering was stable across random states (split seeds were not varied). The results suggest a modest, protocol-dependent benefit within this dataset; the additional analyses provide descriptive evidence.
Related
- Predicting Glass Transition Temperature of Polyimide Homopolymers and Copolymers Using Interpretable QSPR-Based Machine Learning: The MFKit Framework
- AqSol-Blend: A CPU-Trained Descriptor Ensemble for Aqueous Solubility Prediction
- Descriptor-Based Pre-Training Improves Reaction Property Prediction
- Safety margin on the ductile to brittle transition temperature after hydrogen embrittlement on X65 steel
- Temperature-Responsive Ionic Liquids: Polymer Structure, Phase Transition Behaviors and Applications
BibTeX
@article{calısan2026cdeftg,
title = {{CDEF-Tg: Machine Learning-Based Descriptor–Embedding Fusion for Glass Transition Temperature Prediction in a Polyimide-Oriented Polymer Dataset}},
author = {Mücahit Çalışan and İbrahim Ayaz},
journal = {Polymers},
year = {2026},
month = nov,
doi = {10.3390/polym18212783},
url = {https://www.mdpi.com/2073-4360/18/21/2783},
note = {CC BY 4.0},
}
RIS
TY - JOUR
TI - CDEF-Tg: Machine Learning-Based Descriptor–Embedding Fusion for Glass Transition Temperature Prediction in a Polyimide-Oriented Polymer Dataset
AU - Mücahit Çalışan
AU - İbrahim Ayaz
PY - 2026
DA - 2026/11/07
JO - Polymers
DO - 10.3390/polym18212783
UR - https://www.mdpi.com/2073-4360/18/21/2783
AB - Predicting the glass transition temperature (Tg) of polymers from structures remains difficult, and whether descriptors and molecular-language embeddings carry complementary information is unclear. CDEF-Tg (Compact Descriptor–Embedding Fusion for Glass Transition Temperature Prediction) concatenates 205 RDKit descriptors (RDKit205) with a frozen 384-dimensional ChemBERTa embedding (ChemBERTa384) and maps the 589-dimensional vector to Tg with an ExtraTreesRegressor. It was evaluated on 10,065 identity-resolved records of a polyimide-oriented dataset under three leakage-safe protocols: IID Group-Safe (IID, independent and identically distributed), Scaffold OOD (OOD, out-of-distribution), and Similarity OOD (T = 0.90). The model was selected on validation evidence and frozen before the final-test sets were opened. In the locked final test, CDEF-Tg reached a macro mean absolute error (MAE) of 48.84 °C, versus 50.17 °C (RDKit205 alone) and 54.33 °C (ChemBERTa384 alone). Group-aware bootstrap analyses separated CDEF-Tg from RDKit205 alone in MAE only under Similarity OOD and showed no MAE advantage over the comparator RDEF-Tg. Additional validation-only controls showed a lower macro MAE for a Morgan-fingerprint partner than for CDEF-Tg (49.11 versus 53.15 °C), and ChemBERTa384 lowered LightGBM error; the ExtraTrees ordering was stable across random states (split seeds were not varied). The results suggest a modest, protocol-dependent benefit within this dataset; the additional analyses provide descriptive evidence.
SN - 2073-4360
ER -