FastCCS: A Deep Learning Framework for Accurate Collision Cross-Section Prediction from SMILES with Broad Applications in Cheminformatics and Omics

One of the best ways to analyze complex samples is ion-mobility-mass spectrometry (IM-MS). A crucial capability of IM-MS is the ability to compare changes in the analyte structures over time. In analytical and structural chemistry, collision cross-section (CCS) values obtained from IM-MS serve as essential, structure-dependent descriptors. We provide FastCCS, a deep learning system that directly and accurately predicts CCS from SMILES strings and ion adduct types. It was trained on the most chemically diverse CCS dataset, which includes 26 adduct ion states and 23,636 curated molecular structures. Compared with state-of-the-art CCS prediction algorithms, FastCCS achieves a median relative error of 1.7% (<i>R</i>² = 0.99). FastCCS could be used in mass spectrometry coupled with IMS systems to improve compound identification in complex biological and chemical matrices, which is open-access and available for free via the online interface at www.fastccs.com, and its code is available on GitHub for local running and development. Its accuracy, reproducibility, and robustness relative to existing methodologies are demonstrated by strong benchmarking data. Its universal applicability across multiple chemical spaces underlines its relevance for metabolomics, drug discovery, and environmental research.