Rotamer-Dependent Amino Acid Descriptors as Protein and Peptide Representation for Reactivity and Property Prediction
Abstract
We present RotADeL, a Rotamer-dependent Amino acid Descriptors Library designed to facilitate the use of chemically interpretable representations in machine learning models for peptide and protein reactivity. RotADeL contains precomputed steric and electronic descriptors for the rotamer ensembles of the canonical amino acids, accounting for their different protonation states. These descriptors can be queried either from a known protein structure, by selecting the closest library rotamer, or directly from an amino acid sequence, by averaging over the rotamer ensemble according to backbone and side-chain conformational probabilities. The latter approach avoids the need for computationally demanding protein geometry generation. We evaluated the resulting features in two case studies: first, the comparison of RotADeL descriptors to commonly used peptide representations for antimicrobial activity and toxicity classification; secondly, the application of RotADeL descriptors to predict the enantioselectivity of cyclisation reactions catalysed by GluER-T36A. Together, these results show that RotADeL provides a transferable and interpretable amino acid representation for machine learning predictions of peptide property and enzyme reactivity.
Related
- Weighted Views Representation with Multi-Head Coulomb Attention for Molecular Property Prediction
- A Machine Learning Approach to logBB Prediction Combining Molecular Docking and Quantum-Chemical Descriptors
- Charge-Transfer-State Descriptors from Donor–Acceptor Interfaces Enable Improved Prediction of Indoor Organic Photovoltaic Efficiency
- Machine learning prediction of DFT-derived electrical conductivity in perovskite and double perovskite materials using physically motivated descriptors
- Carbon Nanocones Geometry-Dependent Descriptors Evidence Levels and Polymer Interfaces
BibTeX
@misc{jacotdescombes2026rotamerdependent,
title = {{Rotamer-Dependent Amino Acid Descriptors as Protein and Peptide Representation for Reactivity and Property Prediction}},
author = {Lauriane Jacot-Descombes and Kjell Jorner},
howpublished = {ChemRxiv},
year = {2026},
month = sep,
doi = {10.26434/chemrxiv.15009507/v1},
url = {https://chemrxiv.org/doi/full/10.26434/chemrxiv.15009507/v1},
note = {CC BY},
}
RIS
TY - UNPB
TI - Rotamer-Dependent Amino Acid Descriptors as Protein and Peptide Representation for Reactivity and Property Prediction
AU - Lauriane Jacot-Descombes
AU - Kjell Jorner
PY - 2026
DA - 2026/09/27
PB - ChemRxiv
DO - 10.26434/chemrxiv.15009507/v1
UR - https://chemrxiv.org/doi/full/10.26434/chemrxiv.15009507/v1
AB - We present RotADeL, a Rotamer-dependent Amino acid Descriptors Library designed to facilitate the use of chemically interpretable representations in machine learning models for peptide and protein reactivity. RotADeL contains precomputed steric and electronic descriptors for the rotamer ensembles of the canonical amino acids, accounting for their different protonation states. These descriptors can be queried either from a known protein structure, by selecting the closest library rotamer, or directly from an amino acid sequence, by averaging over the rotamer ensemble according to backbone and side-chain conformational probabilities. The latter approach avoids the need for computationally demanding protein geometry generation. We evaluated the resulting features in two case studies: first, the comparison of RotADeL descriptors to commonly used peptide representations for antimicrobial activity and toxicity classification; secondly, the application of RotADeL descriptors to predict the enantioselectivity of cyclisation reactions catalysed by GluER-T36A. Together, these results show that RotADeL provides a transferable and interpretable amino acid representation for machine learning predictions of peptide property and enzyme reactivity.
ER -