Fast and accurate decoding of Raman spectra-encoded suspension arrays using deep learning.

Area of Science:

• Analytical Chemistry
• Biotechnology
• Machine Learning

Background:

• Raman spectroscopy offers ideal encoding properties due to narrow bandwidths and stable signals.
• Suspension arrays (SAs) utilizing micro-quartz pieces (MQPs) with Raman reporter molecules enable multiplexed detection.
• Grafting biomolecule probes onto MQPs allows for simultaneous identification of multiple target analytes.

Purpose of the Study:

• To develop and validate a deep learning model for decoding Raman spectra-encoded suspension arrays.
• To assess the efficiency and accuracy of residual neural networks (ResNet) in classifying analytes from complex spectral data.
• To simplify the decoding process for multiplexed detection systems.

Main Methods:

• A residual neural network (ResNet) was employed to decode Raman spectra from suspension arrays.
• Micro-quartz pieces (MQPs) were functionalized with different Raman reporter molecules and biomolecule probes.
• Raman spectroscopy was used to measure mixed MQPs, and the resulting spectra were analyzed by ResNet for analyte identification.

Main Results:

• ResNet demonstrated superior classification stability and training convergence compared to other machine learning models.
• The classification accuracy for identifying analyte types reached 100%.
• A t-distributed stochastic neighbor embedding (t-SNE) diagram visually confirmed the robust classification performance of ResNet.

Conclusions:

• Residual neural networks effectively decode Raman spectra-encoded suspension arrays for highly accurate multiplexed analyte detection.
• The proposed method significantly simplifies the decoding process in complex biological assays.
• This deep learning approach offers a powerful tool for advancing high-throughput biological sample analysis.