Rapid in situ bio-analysis of cellular behaviors in response
to
external stimuli remains a formidable challenge but can open crucial
opportunities in biology and medicine. The standard label-based end
point assays suffer from invasiveness and complex sample handling.
In this regard, label-free surface-enhanced Raman spectroscopy (SERS)
has emerged as a promising non-invasive in situ bio-analysis technique
for living cells. Nevertheless, achieving rapid in situ SERS bio-analysis
still faces challenges in reliable high-throughput measurements and
accurate multivariate analysis, which requires significant innovations
in bio-interfaced SERS devices and machine learning (ML) methods.
Here, we exploit cell-interfaced nanolaminate SERS substrates to demonstrate
reliable high-throughput SERS measurements using well-studied living
cancer cells with four drug dosages. Artificial neural network (ANN)
for multiclass classification of cellular drug responses provides
high accuracy (94%). Uniquely, nanolaminate SERS substrates with a
high SERS enhancement factor (>107) can rapidly generate
big SERS data sets with rich molecular information on living cells
(10,000 spectra within 3 min) that can enable the utilization of data-hungry
ML methods (e.g., ANN). By capturing additional hidden features in
high-dimensional spectroscopic data, ANN is more powerful for multiclass
classification than five other popular ML methods, including principal
component analysis combined with linear discriminant analysis (PCA-LDA),
partial least-squares discriminant analysis (PLSDA), classification
tree (CT), k-nearest neighbor (KNN), and support
vector machine (SVM). On the basis of the proof-of-concept demonstration
using drugs on living cells, we anticipate that the nanolaminate SERS
substrates can potentially monitor living cell responses to other
external stimuli in a label-free and non-invasive manner.