Skip to main content

Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.

  • Article
  • Published:

Deep spectral component filtering as a foundation model for spectral analysis demonstrated in metabolic profiling

Nature Machine Intelligence volume 7, pages 743–757 (2025)Cite this article

Subjects

Abstract

Analysing metabolites in bioliquids through various spectroscopic methods provides valuable insights into the metabolic phenotypes. Deciphering spectral data has greatly benefited from deep-learning methods; however, data-driven solutions often struggle with data dependence on different devices, samples and spectral modalities. Most current task-specific methods have limited generalizability to different spectral analysis problems, including preprocessing, quantification and interpretation. Here, we developed a pretrained foundation model, termed deep-spectral component filtering (DSCF) through a self-supervised approach termed spectral component resolvable learning. By acquiring general spectral knowledge, DSCF achieved state-of-the-art performance for five distinct spectral analysis tasks on 11 datasets. Notably, the general pretraining led to zero-shot spectral denoising and trace-level quantification in complex mixtures. DSCF achieved molecule-level interpretation of surface-enhanced Raman spectra and mapped serum metabolic profiles from nearly 600 individuals for various diseases, including stroke, Alzheimer’s disease and prostate cancer. Overall, the proposed foundation model illustrates promising generalizability for spectral analysis and offers a clear and feasible pathway for general spectral analysis.

Access through your institution
Buy or subscribe

This is a preview of subscription content, access via your institution

Access options

Access through your institution

Buy this article

  • Purchase on SpringerLink
  • Instant access to full article PDF
Buy now

Prices may be subject to local taxes which are calculated during checkout

Fig. 1: Overview of DSCF.
Fig. 2: Results of DSCF dewaxing via IR histological spectroscopy.
Fig. 3: Results of DSCF-based preprocessing in SERS: NP background subtraction, and noise elimination in SERS, Raman and IR spectra.
Fig. 4: Results of spectral quantification.
Fig. 5: Results of the explanation efficacy of the ComFilE approach.
Fig. 6: Biomarker discovery of stroke via SERS spectra of serum samples.

Similar content being viewed by others

Data availability

All the data included in the validation of the DSCF model are open source. The IR spectra of liver tissues are available at figshare https://doi.org/10.6084/m9.figshare.28107236 (ref. 60). The serum SERS spectra for PCa research are available at figshare https://doi.org/10.6084/m9.figshare.28107395 (ref. 61). The serum SERS spectra for stroke research are available at figshare https://doi.org/10.6084/m9.figshare.28107431 (ref. 62). The serum SERS spectra for AD research are available at figshare https://doi.org/10.6084/m9.figshare.28107578 (ref. 63). The SERS spectra of synthetic solution for quantification are at available figshare https://doi.org/10.6084/m9.figshare.28107281 (ref. 64). The SERS spectra for background removal are available at figshare https://doi.org/10.6084/m9.figshare.28107326 (ref. 65) and https://doi.org/10.6084/m9.figshare.28107305. The simulation spectra of QM9S are available at figshare https://figshare.com/articles/dataset/QM9S_dataset/24235333 (ref. 66). Source data are provided with this paper.

Code availability

All the codes for the DSCF model have been made public at Zenodo https://doi.org/10.5281/zenodo.15013288 (ref. 19).

References

  1. Holmes, E., Wilson, I. D. & Nicholson, J. K. Metabolic phenotyping in health and disease. Cell 134, 714–717 (2008).

    Article  Google Scholar 

  2. Nicholson, J. K. Global systems biology, personalized medicine and molecular epidemiology. Mol. Syst. Biol. 2, 52 (2006).

    Article  Google Scholar 

  3. Alseekh, S. et al. Mass spectrometry-based metabolomics: a guide for annotation, quantification and best reporting practices. Nat. Methods 18, 747–756 (2021).

    Article  Google Scholar 

  4. Julkunen, H. et al. Atlas of plasma NMR biomarkers for health and disease in 118,461 individuals from the UK Biobank. Nat. Commun. 14, 604 (2023).

    Article  Google Scholar 

  5. Hu, J. et al. RSPSSL: a novel high-fidelity Raman spectral preprocessing scheme to enhance biomedical applications and chemical resolution visualization. Light Sci. Appl. 13, 52 (2024).

    Article  Google Scholar 

  6. He, H. et al. Noise learning of instruments for high-contrast, high-resolution and fast hyperspectral microscopy and nanoscopy. Nat. Commun. 15, 754 (2024).

    Article  Google Scholar 

  7. Guo, S., Popp, J. & Bocklitz, T. Chemometric analysis in Raman spectroscopy from experimental design to machine learning–based modeling. Nat. Protoc. 16, 5426–5459 (2021).

    Article  Google Scholar 

  8. Felten, J. et al. Vibrational spectroscopic image analysis of biological material using multivariate curve resolution–alternating least squares (MCR-ALS). Nat. Protoc. 10, 217–240 (2015).

    Article  Google Scholar 

  9. Bi, X. et al. Molecule-resolvable SERSome for metabolic profiling. Chem https://doi.org/10.1016/j.chempr.2025.102528 (2025).

  10. Su, H. et al. Surface-enhanced Raman spectroscopy study on the structure changes of 4-mercaptophenylboronic acid under different pH conditions. Spectrochim. Acta A Mol. Biomol. Spectrosc. 185, 336–342 (2017).

    Article  Google Scholar 

  11. Giese, B. & McNaughton, D. Surface-enhanced Raman spectroscopic study of uracil. The influence of the surface substrate, surface potential, and pH. J. Phys. Chem. B 106, 1461–1470 (2002).

    Article  Google Scholar 

  12. Zarei, M. et al. Machine learning analysis of Raman spectra to quantify the organic constituents in complex organic-mineral mixtures. Anal. Chem. 95, 15908–15916 (2023).

    Article  Google Scholar 

  13. Koyun, O. C. et al. RamanFormer: a transformer-based quantification approach for raman mixture components. ACS Omega 9, 23241–23251 (2024).

    Article  Google Scholar 

  14. Brown, T. et al. Language models are few-shot learners. In Proc. Advances in Neural Information Processing Systems Vol. 33, 1877–1901 (NeurIPS, 2020).

  15. Bi, X., Czajkowsky, D. M., Shao, Z. & Ye, J. Digital colloid-enhanced Raman spectroscopy by single-molecule counting. Nature 628, 771–775 (2024).

    Article  Google Scholar 

  16. Luo, Y. et al. Component identification for the SERS spectra of microplastics mixture with convolutional neural network. Sci. Total Environ. 895, 165138 (2023).

    Article  Google Scholar 

  17. Bi, X. et al. SERSomes for metabolic phenotyping and prostate cancer diagnosis. Cell Rep. Med. https://doi.org/10.1016/J.XCRM.2024.101579 (2024).

  18. Ye, J. et al. Hypoxanthine is a metabolic biomarker for inducing GSDME-dependent pyroptosis of endothelial cells during ischemic stroke. Theranostics 14, 6071–6087 (2024).

    Article  Google Scholar 

  19. Xue, B. Source code for deep spectral component filtering: DSCF_V1 (v1.2). Zenodo https://doi.org/10.5281/zenodo.15013288 (2025).

  20. Isensee, F., Jaeger, P. F., Kohl, S. A. A., Petersen, J. & Maier-Hein, K. H. nnU-Net: a self-configuring method for deep learning-based biomedical image segmentation. Nat. Methods 18, 203–211 (2020).

    Article  Google Scholar 

  21. Liu, Z. et al. Swin transformer: hierarchical vision transformer using shifted windows. In Proc. IEEE/CVF International Conference on Computer Vision 9992–10002 (IEEE, 2021).

  22. Hughes, C., Gaunt, L., Brown, M., Clarke, N. W. & Gardner, P. Assessment of paraffin removal from prostate FFPE sections using transmission mode FTIR-FPA imaging. Anal. Methods 6, 1028–1035 (2014).

    Article  Google Scholar 

  23. Meuse, C. W. & Barker, P. E. Quantitative infrared spectroscopy of formalin-fixed, paraffin-embedded tissue specimens: paraffin wax removal with organic solvents. Appl. Immunohistochem. Mol. Morphol. 17, 547–552 (2009).

    Article  Google Scholar 

  24. Nallala, J., Lloyd, G. R. & Stone, N. Evaluation of different tissue de-paraffinization procedures for infrared spectral imaging. Analyst 140, 2369–2375 (2015).

    Article  Google Scholar 

  25. De Lima, F. A. et al. Digital de-waxing on FTIR images. Analyst 142, 1358–1370 (2017).

    Article  Google Scholar 

  26. Bai, B. et al. Deep learning-enabled virtual histological staining of biological samples. Light Sci. Appl. 12, 57 (2023).

    Article  Google Scholar 

  27. Lotfollahi, M., Daeinejad, D., Berisha, S. & Mayerich, D. Digital staining of high-resolution FTIR spectroscopic images. In Proc. 2018 IEEE Global Conference on Signal and Information Processing, GlobalSIP 973–977 (IEEE, 2018).

  28. Gobinet, C. et al. Automatic identification of paraffin pixels on FTIR images acquired on FFPE human samples. Anal. Chem. 93, 3750–3761 (2021).

    Article  Google Scholar 

  29. Bi, X., Fang, Z., Deng, B., Zhou, L. & Ye, J. Ultrahigh Raman-fluorescence dual-enhancement in nanogaps of silver-coated gold nanopetals. Adv. Opt. Mater. 11, 2300188 (2023).

    Article  Google Scholar 

  30. Wang, H. P. et al. Recent advances of chemometric calibration methods in modern spectroscopy: algorithms, strategy, and related issues. Trends Anal. Chem. 153, 116648 (2022).

    Article  Google Scholar 

  31. Zou, Z. et al. A deep learning model for predicting selected organic molecular spectra. Nat. Comput. Sci. 3, 957–964 (2023).

    Article  Google Scholar 

  32. Savitzky, A. & Golay, M. J. E. Smoothing and differentiation of data by simplified least squares procedures. Anal. Chem. 36, 1627–1639 (1964).

    Article  Google Scholar 

  33. You, L. et al. An exposome atlas of serum reveals the risk of chronic diseases in the Chinese population. Nat. Commun. 15, 2268 (2024).

    Article  Google Scholar 

  34. Oh, H. S. H. et al. Organ aging signatures in the plasma proteome track health and disease. Nature 624, 164–172 (2023).

    Article  Google Scholar 

  35. Enroth, C. et al. Crystal structures of bovine milk xanthine dehydrogenase and xanthine oxidase: structure-based mechanism of conversion. Proc. Natl Acad. Sci. USA 97, 10723–10728 (2000).

    Article  Google Scholar 

  36. Schwedhelm, E. et al. Trimethyllysine, vascular risk factors and outcome in acute ischemic stroke (MARK–STROKE). Amino Acids 53, 555–561 (2021).

    Article  Google Scholar 

  37. Farthing, D. E., Farthing, C. A. & Xi, L. Inosine and hypoxanthine as novel biomarkers for cardiac ischemia: from bench to point-of-care. Exp. Biol. Med. 240, 821–831 (2015).

    Article  Google Scholar 

  38. Dudka, I. et al. Comprehensive metabolomics analysis of prostate cancer tissue in relation to tumor aggressiveness and TMPRSS2-ERG fusion status. BMC Cancer 20, 437 (2020).

    Article  Google Scholar 

  39. Chen, M. M. & Meng, L. H. The double faced role of xanthine oxidoreductase in cancer. Acta Pharmacol. Sin. 43, 1623–1632 (2021).

    Article  Google Scholar 

  40. Sreekumar, A. et al. Metabolomic profiles delineate potential role for sarcosine in prostate cancer progression. Nature 457, 910–914 (2009).

    Article  Google Scholar 

  41. Clark, D. E. et al. The serine/threonine protein kinase, p90 ribosomal S6 kinase, is an important regulator of prostate cancer cell proliferation. Cancer Res. 65, 3108–3116 (2005).

    Article  Google Scholar 

  42. Maitre, M., Klein, C., Patte-Mensah, C. & Mensah-Nyagan, A. G. Tryptophan metabolites modify brain Aβ peptide degradation: a role in Alzheimer’s disease? Prog. Neurobiol. 190, 101800 (2020).

    Article  Google Scholar 

  43. Horgan, C. C. et al. High-throughput molecular imaging via deep-learning-enabled Raman spectroscopy. Anal. Chem. 93, 15850–15860 (2021).

    Article  Google Scholar 

  44. Ribeiro, M. T., Singh, S. & Guestrin, C. ‘Why should I trust you?’: explaining the predictions of any classifier. In Proc. 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (ed. Balaji, K.) 1135–1144 (ACM, 2016).

  45. Radford, A. et al. Language models are unsupervised multitask learners. OpenAI 1, 9 (2019).

    Google Scholar 

  46. Liu, B., Yuan, Y., Pan, X., Shen, H-.B. & Jin, C. AttSiOff: a self-attention-based approach on siRNA design with inhibition and off-target effect prediction. Med-X https://doi.org/10.1007/s44258-024-00019-1 (2024).

  47. Leopold, N. & Lendl, B. A new method for fast preparation of highly surface-enhanced raman scattering (SERS) active silver colloids at room temperature by reduction of silver nitrate with hydroxylamine hydrochloride. J. Phys. Chem. B 107, 5723–5727 (2003).

    Article  Google Scholar 

  48. Lee, P. C. & Meisel, D. Adsorption and surface-enhanced Raman of dyes on silver and gold sols. J. Phys. Chem. 86, 3391–3395 (1982).

    Article  Google Scholar 

  49. Yuan, H. et al. Gold nanostars: surfactant-free synthesis, 3D modelling, and two-photon photoluminescence imaging. Nanotechnology 23, 075102 (2012).

    Article  Google Scholar 

  50. Kaur, V., Tanwar, S., Kaur, G. & Sen, T. DNA-origami-based assembly of Au@Ag Nanostar dimer nanoantennas for label-free sensing of pyocyanin. Chem. Phys. Chem. 22, 160–167 (2021).

    Article  Google Scholar 

  51. He, K. et al. Masked autoencoders are scalable vision learners. In Proc. IEEE/CVF International Conference on Computer Vision 15979–15988 (IEEE, 2022).

  52. Chau, S. L., Hu, R., London, A., Gonzalez, J. & Sejdinovic, D. RKHS-SHAP: Shapley values for kernel methods. Adv. Neural Inf. Process Syst. 35, 13050–13063 (2022).

    Google Scholar 

  53. Shrikumar, A., Greenside, P. & Kundaje, A. Learning important features through propagating activation differences. In Proc. 34th International Conference on Machine Learning Vol. 70 (eds Precup, D. & Teh, Y. W.) 3145–3153 (PMLR, 2017).

  54. Selvaraju, R. R. et al. Grad-CAM: visual explanations from deep networks via gradient-based localization. In Proc. IEEE/CVF International Conference on Computer Vision 618–626 (IEEE, 2017).

  55. Zeiler, M. D. & Fergus, R. Visualizing and understanding convolutional networks. In Proc. European Conference on Computer Vision 818–833 (Springer, 2014).

  56. Lundberg, S. M., Allen, P. G. & Lee, S.-I. A unified approach to interpreting model predictions. In Proc. Advances in Neural Information Processing Systems Vol. 30 (eds Guyon, I. et al.) 4765–4774 (Curran Associates, 2017).

  57. Sundararajan, M., Taly, A. & Yan, Q. Axiomatic attribution for deep networks. In Proc. 34th International Conference on Machine Learning (JMLR, 2017).

  58. Breiman, L. Random forests. Mach. Learn. 45, 5–32 (2001).

    Article  Google Scholar 

  59. Zhou, B., Khosla, A., Lapedriza, A., Oliva, A. & Torralba, A. Learning deep features for discriminative localization. In Proc. IEEE/CVF International Conference on Computer Vision 2921–2929 (IEEE, 2016).

  60. Xue, B. FTIR of liver tissue. figshare https://doi.org/10.6084/m9.figshare.28107236.v1 (2024).

  61. Xue, B. ComFilE for PCa. figshare https://doi.org/10.6084/m9.figshare.28107395.v1 (2024).

  62. Xue, B. ComFilE for stroke. figshare https://doi.org/10.6084/m9.figshare.28107431.v1 (2024).

  63. Xue, B. ComFilE for AD. figshare https://doi.org/10.6084/m9.figshare.28107578.v1 (2024).

  64. Xue, B. SERS quantification. figshare https://doi.org/10.6084/m9.figshare.28107281.v1 (2024).

  65. Xue, B. SERS nanoparticle background removal (PCa). figshare https://doi.org/10.6084/m9.figshare.28107326.v1 (2024).

  66. Zou, Z. QM9S dataset. figshare https://doi.org/10.6084/m9.figshare.24235333.v3 (2023).

Download references

Acknowledgements

We gratefully acknowledge the financial support from the National Key Research and Development Program of China (grant numbers 2024YFF1502600 and 2022YFB4702702), the Major R&D Plan of Shanghai Municipal Science and Technology Commission (grant number 1DZ1100301), the National Natural Science Foundation of China (grant numbers 82272054, 81627801, 31971151, 623B2070 and 82373358), the Science and Technology Commission of Shanghai Municipality (grant numbers 24DIPA00300, 24490710800, 24490790900, 21511102100 and BI0820067/002), the Fundamental Research Funds for the Central Universities (grant numbers YG2024LC09 and YG2025ZD25), the Clinical Research Plan of Shanghai Hospital Development Center (grant number SHDC2020CR3014A), the ‘Clinic Plus’ Outstanding Project (grant number 2023ZYA007) from the Shanghai Key Laboratory for Nucleic Acid Chemistry and Nanomedicine and the Shanghai Key Laboratory of Gynecologic Oncology. Qiqihar Science and Technology Plan Joint Guidance Project (grant number LHYD-202016). We thank Z. Zhao and B. Han at Qingdao University Affiliated Hospital for providing the hepatocellular carcinoma FFPE tissues. Our gratitude also extends to Z. Zou and W. Hu from the Department of Chemistry at Qilu University of Technology for their dataset contributions. We thank C. Wu at the School of Biomedical Engineering, Shanghai Jiao Tong University, for providing animal materials. We thank J. Zhao, Department of Neurology, Minhang Hospital, Fundan University, for providing serum samples from the stroke cohort. We are grateful to J. Pan and W. Xue, Department of Urology, Ren Ji Hospital, School of Medicine, Shanghai Jiao Tong University, for providing the serum samples from the PCa cohort. We thank H. Tang, Department of Geriatrics, Rui Jin Hospital, School of Medicine, Shanghai Jiao Tong University, for serum samples from the AD cohort.

Author information

Author notes

  1. These authors contributed equally: Bingsen Xue, Xinyuan Bi.

Authors and Affiliations

  1. Institute of Medical Robotics, School of Biomedical Engineering, Shanghai Jiao Tong University, Shanghai, China

    Bingsen Xue, Xinyuan Bi, Xin Fang, Yizhe Yuan, Ruoxi Wang, Shuyu Liu, Rushi Jiao, Yuze Chen, Weitao Zu, Chengxiang Wang, Jianhao Zhang, Jian Ye & Cheng Jin

  2. Institute of Digital Medicine, Shanghai Jiao Tong University, Shanghai, China

    Bingsen Xue & Cheng Jin

  3. Shanghai Artificial Intelligence Laboratory, Shanghai, China

    Bingsen Xue, Ya Zhang, Yanfeng Wang & Cheng Jin

  4. National Clinical Research Center for Kidney Diseases, Beijing, China

    Zheyi Dong & Cheng Jin

  5. School of Electronic Information and Electronical Engineering, Shanghai Jiao Tong University, Shanghai, China

    Yunzhe Xu, Ya Zhang & Yanfeng Wang

  6. School of Medical Technology, Qiqihar Medical University, Heilongjiang, China

    Minghui Liang

  7. Department of Electrical and Computer Engineering, Johns Hopkins University, Baltimore, USA

    Jiang Liu

  8. Department of Radiation Oncology, Shanghai Chest Hospital, Shanghai Jiao Tong University School of Medicine, Shanghai, China

    Qin Zhang

  9. Key Laboratory of Biopharmaceutical Preparation and Delivery, Chinese Academy of Sciences, Beijing, China

    Ye Yuan

  10. Department of Pathology, Fudan University, Shanghai Cancer Center, Shanghai, China

    Midie Xu

  11. Shanghai Key Laboratory of Gynecologic Oncology, Ren Ji Hospital, School of Medicine, Shanghai Jiao Tong University, Shanghai, China

    Jian Ye

  12. Beijing Anding Hospital Capital Medical University, Beijing, China

    Cheng Jin

Authors
  1. Bingsen Xue
    View author publications

    Search author on:PubMed Google Scholar

  2. Xinyuan Bi
    View author publications

    Search author on:PubMed Google Scholar

  3. Zheyi Dong
    View author publications

    Search author on:PubMed Google Scholar

  4. Yunzhe Xu
    View author publications

    Search author on:PubMed Google Scholar

  5. Minghui Liang
    View author publications

    Search author on:PubMed Google Scholar

  6. Xin Fang
    View author publications

    Search author on:PubMed Google Scholar

  7. Yizhe Yuan
    View author publications

    Search author on:PubMed Google Scholar

  8. Ruoxi Wang
    View author publications

    Search author on:PubMed Google Scholar

  9. Shuyu Liu
    View author publications

    Search author on:PubMed Google Scholar

  10. Rushi Jiao
    View author publications

    Search author on:PubMed Google Scholar

  11. Yuze Chen
    View author publications

    Search author on:PubMed Google Scholar

  12. Weitao Zu
    View author publications

    Search author on:PubMed Google Scholar

  13. Chengxiang Wang
    View author publications

    Search author on:PubMed Google Scholar

  14. Jianhao Zhang
    View author publications

    Search author on:PubMed Google Scholar

  15. Jiang Liu
    View author publications

    Search author on:PubMed Google Scholar

  16. Qin Zhang
    View author publications

    Search author on:PubMed Google Scholar

  17. Ye Yuan
    View author publications

    Search author on:PubMed Google Scholar

  18. Midie Xu
    View author publications

    Search author on:PubMed Google Scholar

  19. Ya Zhang
    View author publications

    Search author on:PubMed Google Scholar

  20. Yanfeng Wang
    View author publications

    Search author on:PubMed Google Scholar

  21. Jian Ye
    View author publications

    Search author on:PubMed Google Scholar

  22. Cheng Jin
    View author publications

    Search author on:PubMed Google Scholar

Contributions

C.J. and J.Y. conceived the research. B.X. designed and implemented the DSCF models and ComFilE, collected the IR spectra of human liver and rat kidney FFPE sections and organized the reader studies. X.B. synthesized the silver NPs, configured the synthetic mixtures and performed the SERS measurements. M.L., M.X., X.F. and Yizhe Yuan collected the H&E staining images of the human liver cancer FFPE sections. J.L., Y.Z., Y.C., S.L., R.W. and R.J. contributed to the model implementation. C.J. administered the project and provided guidance on the methodology. C.J. and J.Y. guided the SERS measurements of the serum samples. All the authors wrote and revised the paper. B.X. and X.B. contributed equally to this work.

Corresponding authors

Correspondence to Jian Ye or Cheng Jin.

Ethics declarations

Competing interests

The authors declare no competing interests.

Peer review

Peer review information

Nature Machine Intelligence thanks Erik Vartiainen and the other, anonymous, reviewer(s) for their contribution to the peer review of this work.

Additional information

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Extended data

Extended Data Fig. 1 Examples of human liver FFPE sections.

Panels (left to right): infrared hyperspectral images, H&E staining, paraffin abundance maps, and cell annotations. Scale bar = 2 mm.

Source data

Extended Data Fig. 2 The scheme for the simulation of the datasets and explanation evaluation.

The profiles recorded the concentrations and the interactive logic of the “biomarker” molecules and the other molecules. The spectra were simulated based on the profile and the spectra dictionary. The classification labels were decided by the concentration threshold and interactive logics among biomarkers. Multiple interactive logics (such as ‘AND’, ‘OR’, ‘SOFT’, etc.) were used to simulate the different interaction and correlation among the molecules in various diseases. Recall and accuracy were evaluated by comparing the predictive biomarkers with the ground-truth biomarkers.

Extended Data Fig. 3 The workflow of comparison between multiple classical explanations + Raman Experts and ComFilE.

a Eleven classical explanations were applied to the trained classifiers and calculated the heat maps. The Raman experts ranked the molecules as decreasing the possibility of markers, referring to heat maps and spectral database of molecules. b ComFilE was applied to trained classifiers, calculated delta accuracy, and ranked the molecules as the delta accuracy decreasing. c The evaluation of explanation efficacy.

Extended Data Fig. 4 Detail results of ComFilE applied in synthetic data and clinical data.

(a-b) The relationship between the explanation efficacy and the scale of training datasets: (a) Explanation recall versus the training dataset scale and (b) Explanation accuracy versus the training dataset scale. (c-d) The relationship between the explanation efficacy and the classification performance (n = 8). Data are all presented as means ± SD. (c) Explanation accuracy versus classification accuracy. (d) Explanation recall versus classification accuracy. In summary, the numerical test indicates that the explanation results derived from models with a classification accuracy of over 90% hold a very high level of credibility (Top-3 accuracy>98%). (e) The receiver operating characteristic (ROC) curves of prostate cancer diagnosis with different metabolites filtered. (f) The receiver operating characteristic (ROC) curves of stroke diagnosis with different metabolites filtered. Data in this figure are all presented as means ± SD.

Source data

Extended Data Fig. 5 2-nd Component Filtering Explanation.

a Workflow of 2-nd Component Filtering Explanation. b Results Demonstration of 2-nd ComFilE. c-g Coupling effects of distinct interaction logic by different classifiers. h Summary of coupling effects by different classifiers. i-m Additive effects of distinct interaction logic by different classifiers. n Summary of additive effects by different classifiers. c-n n = 72 for AND, SOFT and OR while n = 2448 for none. Data in this figure are all presented as means ± SD.

Source data

Extended Data Fig. 6 Biomarker discovery of prostate cancer and Alzheimer’s Diseases using SERS spectra of serum samples.

a Diagnostic accuracy changes (δacc) of 22 metabolic molecules for prostate cancer, assessed by ComFilE. b Abundance differences of 22 metabolic molecules between prostate cancer patients (n = 116) and controls (n = 104), analyzed by a one-sided t-test (Supplementary Table 6). No multiple comparison adjustments were applied, as only one hypothesis was tested.c Diagnostic accuracy changes (δacc) of 22 metabolic molecules for Alzheimer’s disease (AD), assessed by ComFilE. d Abundance differences of 22 metabolic molecules between AD patients (n = 60) and controls (n = 43), analyzed by a one-sided t-test (Supplementary Table 7). No multiple comparison adjustments were applied, as only one hypothesis was tested. Box plots depict medians (center lines), 25th–75th percentiles (box bounds), and whiskers extending to the furthest points within 1.5 × IQR from the box edges.

Source data

Supplementary information

Supplementary Information

Supplementary Notes 1–13 and Tables 1–7.

Reporting Summary

Source data

Source Data Fig. 1

Performances summary.

Source Data Fig. 2

Results of DSCF dewaxing via IR histological spectroscopy.

Source Data Fig. 3

Results of DSCF-based preprocessing in SERS. (a–e) NP background subtraction and (f–h) noise elimination in SERS, Raman and IR spectra.

Source Data Fig. 4

Results of spectral quantification.

Source Data Fig. 5

Results of the explanation efficacy of the ComFilE approach.

Source Data Fig. 6

Biomarker discovery of stroke via SERS spectra of serum samples.

Source Data Extended Data Fig. 1

IR hyperspectral images of FFPE sections.

Source Data Extended Data Fig. 4

Detail results of ComFilE applied in synthetic data and clinical data.

Source Data Extended Data Fig. 5

Second ComFilE.

Source Data Extended Data Fig. 6

Biomarker discovery of PCa and Alzheimer’s diseases using SERS spectra of serum samples.

Rights and permissions

Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.

Reprints and permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Xue, B., Bi, X., Dong, Z. et al. Deep spectral component filtering as a foundation model for spectral analysis demonstrated in metabolic profiling. Nat Mach Intell 7, 743–757 (2025). https://doi.org/10.1038/s42256-025-01027-5

Download citation

  • Received:

  • Accepted:

  • Published:

  • Issue date:

  • DOI: https://doi.org/10.1038/s42256-025-01027-5

Search

Advanced search

Quick links

Nature Briefing: Translational Research

Sign up for the Nature Briefing: Translational Research newsletter — top stories in biotechnology, drug discovery and pharma.

Get what matters in translational research, free to your inbox weekly. Sign up for Nature Briefing: Translational Research