Raman Biosensing of Sweat Metabolites: Univariate vs. Multivariate Algorithms

Over the past eight years, there has been a remarkable surge in the field of sweat analysis, as a non-invasive, continuous, and multi-metabolite monitoring solution tailored for wearable devices. However, its full potential has yet to be fully realized due to the limitations of existing biosensing transducers. Despite years of research, wearable devices still fall short of providing biochemical insights into human functions, largely due to the longevity issues associated with colorimetric and electrochemical biosensing methods stemming from their biorecognition elements. However, optical methods such as Raman scattering measurements offer an alternative, inherently selective biosensing mechanism without the longevity issues seen in other methods. While the main hurdle in the past was the bulky instrumentation required, advancements in microengineering and laser technology have paved the way for the development of compact Raman systems. Nevertheless, research at the intersection of Raman systems and sweat analysis (or other alternative biofluids to blood) is still in its infancy, with no comparative studies to assess the efficiency of multivariate versus univariate data analysis techniques in biosensing. To address this, the present work analyzes two of these widely used data processing methods in multiplexed human sweat glucose, urea, and lactate biosensing. Experimental findings suggest that multivariate analysis, particularly Principal Components Regression (PCR), demonstrates better performance especially in datasets containing interferents, outperforming univariate analysis. This paper also delves into the potential advantages and limitations associated with the two investigated algorithms, shedding light on their applicability in sweat analysis for future wearables Raman systems.