SCDM-Net: Symbiotic causal debiasing for SERS-based exosome profiling in lung cancer

Area of Science:

• Biomedical Engineering
• Computational Biology
• Analytical Chemistry

Background:

• Early lung cancer detection significantly improves patient outcomes.
• Surface-enhanced Raman scattering (SERS) analysis of exosomes shows promise for molecular profiling in lung cancer detection.
• Instrumental and batch variations in SERS data can hinder the identification of crucial spectral features.

Purpose of the Study:

• To develop a robust computational method for analyzing exosome SERS spectra for lung cancer detection.
• To address and mitigate data variations inherent in SERS measurements.
• To enhance the accuracy and reliability of SERS-based molecular profiling for early lung cancer diagnosis.

Main Methods:

• Development of a Symbiotic Causal Debiasing Multi-scale Network (SCDM-Net) incorporating a symbiotic-parasitic module with gated residual fusion.
• Utilized a dual-stream architecture processing both 1D spectra and 2D STFT representations with multi-scale convolutions.
• Employed a gradient reversal layer (GRL) to effectively separate batch-associated and class-relevant spectral representations.

Main Results:

• SCDM-Net achieved 90.77% accuracy and an F1 score of 0.8980 on an independent test set.
• The model demonstrated effectiveness on 4006 exosome SERS spectra from 12 biological batches.
• Successfully separated batch-associated variations from class-relevant spectral features in complex SERS data.

Conclusions:

• The SCDM-Net provides a promising methodological framework for SERS-based exosome profiling in lung cancer detection.
• This AI-driven approach can overcome technical variations, paving the way for more reliable SERS analysis.
• Further validation with patient-derived samples is necessary for clinical translation of this SERS-based lung cancer diagnostic approach.