Deep Learning–Driven Data Fusion of NMR and Mass Spectrometry for Disease Prediction
Year: 2026
Type: conference-paper
Abstract: Nuclear magnetic resonance (NMR) and Mass spectrometry (MS) are two mainly used techniques in metabolomics that can measure abundance of hundreds of metabolites. The data generated by either techniques have already allowed a huge progress in various research fields from personalized medicine to pharmaceutical sciences. Combining measurements from both platforms through a simple concatenation also led valuable discoveries. In this work, however, we propose an alternative data fusion approach based on deep learning that aims to make use of each type of measurement and combine the derived information in a different layer for disease prediction. Rather than feeding in all data as a single vector, we create dual "pipelines", operating on NMR and MS datasets separately, and then combine the outputs into a single prediction. We measure the performance of our dual-pipeline approach by using Alzheimer’s and Parkinson’s data for diagnostic prediction. We not only build custom deep learning (DL) architectures for both data sets, but also differentiate pipelines as to how to handle NMR and MS, i.e., separately or concatenated. Our results show that the dual-pipeline approach performs better than a single traditional DL pipeline which relies on traditional data fusion approach when both NMR and MS data are available.
Language: English
Subfield: Molecular Biology
Domain: Life Sciences
Open Access status: hybrid
OpenAlex ID: W7202269203