UEFCCS: A Multimodal Deep Learning Framework with Fine-Tuned 3D Molecular Representations for Collision Cross Section Prediction
Collision cross section (CCS), derived from ion mobility spectrometry, provides a valuable reference for analyte identification. Achieving high predictive accuracy and effective extrapolation to structurally dissimilar molecules, however, remains challenging for existing machine-learning models. Here, we present UEFCCS, a multimodal deep learning framework that integrates molecular formula encoding and fingerprints with CCS-adapted three-dimensional (3D) molecular representations, obtained by fine-tuning a pretrained Transformer encoder using only [M+H]+ CCS data from the training set. Formula and fingerprint features retain composition and
Authors
- Liang Tian (ORCID: https://orcid.org/0000-0002-6595-445X)
- Xingye Cheng (ORCID: https://orcid.org/0000-0002-9996-455X)
- Xin Xiong
- Aiping Lyu
- Zongwei Cai
- YAPING SHI (ORCID: https://orcid.org/0009-0007-1001-2880)
- Simin Zhang
Institutions
- Hong Kong Baptist University (HK)
- Chinese University of Hong Kong (HK)
- Eastern Institute of Technology, Ningbo (CN)
Publication Details
- Journal
- ChemRxiv
- Published
- 2026-11-18
- DOI
- https://doi.org/10.26434/chemrxiv.15012449/v1
- Primary Topic
- Mass Spectrometry Techniques and Applications
- Type
- preprint