Deep learning-based deconvolution of SERS spectra of ovine cerebral spinal fluid

In fifteen words or less, explain the significance of this contribution (Novel Aspect).: Exploration of deep learning for deconvolution of complex SERS spectra of biofluids for disease diagnostics.

Abstract Text:

Many neurological diseases are difficult to diagnose, particularly in the early stages of disease progression where treatment would be most beneficial. For various conditions, including mental health disorders, changes in neurochemical concentrations indicate the onset or progression of the disease. Patients with clinical depression are often deficient in neurotransmitters including dopamine, serotonin, and norepinephrine. While it is known that understanding neurochemical concentrations are integral to diagnosing various neurological conditions, they are rarely monitored. Methods used for measuring and quantifying neurochemical concentrations require large, expensive instrumentation, complex sample preparation, and long collection times. There is an urgent need for the development of neurochemical sensing methods for early disease detection that are rapid, label-free, selective, and involve little to no sample preparation. Surface-enhanced Raman spectroscopy (SERS) is a highly sensitive and specific vibrational spectroscopy that can be applied towards neurochemical sensing. Here we integrate SERS with Deep Learning and a supervised spectral unmixing function, non-negative least squares (NNLS), to deconvolute the neurochemical signals in SERS spectra of cerebral spinal fluid (CSF) extracted from sheep. The NNLS analysis identified approximately ten components that comprised the CSF SERS spectra. The components were then input in a pretrained neural network and compare with the SERS spectra of eight neurochemicals (dopamine, serotonin, melatonin, epinephrine, norepinephrine, 5-HIAA, and DOPAC) using cosine similarity. We will highlight the power of combining SERS with machine learning for a rapid and reliable method to deconvolute the SERS spectra of ovine CSF and determine the neurochemical contributions.