Spectroscopy is the discipline of studying the interactions between light and matter, in
order to perform analyzes unmatched in terms of the speed of data acquisition. Depending
on the spectral ranges used by the sensors, it is possible to carry out molecular
(molecular and vibrational spectroscopy) or elementary (atomic spectroscopy) analyzes.

As part of this project, GreenTropism has selected Surface Enhanced Raman Scattering
(SERS) technology as a spectral technique. The scientific literature reports several
cases of use of SERS technology for virus analysis, under variable conditions: variable
viral loads, after amplification, use of substrates enriched in antigens.

The SERS allows an analysis of a sample deposited on a substrate on average (from fifteen
seconds to 10 minutes depending on the devices and the presence of complementary
imaging). Already proven for the identification of viruses on strains pathogenic for
humans and animals, its deployment is slowed down by the complexity of the data to be
processed.

These spectra acquisition technologies require the joint use of statistical tools and
multivariate analyzes to allow sample discrimination (classification) and / or
quantification. Until recently, the capacity and performance of statistical tools were
limited by the available computational capacities. The lifting of this technological lock
allowed the advent and democratization of Artificial Intelligence (AI) techniques
theorized in the 1960s and applied today.

GreenTropism's Kaïssa, AI tool, in addition to processing big data, has been designed and
trained specifically for processing spectral data and automating all of the algorithmic
chains needed to go from spectrum support to its interpretation, and the presentation of
the final answer.

The analysis of chemometric data, for the purpose of classification, implements several
types of algorithms that Kaïssa uses, combining them automatically to obtain the best
possible analyzes of these data. These algorithms are divided into two large groups:
mathematical preprocessing and classification models.

The combination of photonic technologies (here SERS) and AI allows real-time analyzes of
multiple substrates, without a priori knowledge of the user on the sample and without
prior expertise. These characteristics make it a valuable tool for diagnosing SARS-CoV-2
infection in the context of Point Of Care.

In a work carried out between the months of March and June 2020, several models showed,
on test databases not integrated in the learning, performance of discrimination between
positive and negative patients for SARS-CoV-2 according to RT-PCR equivalent to Youden
indices of 0.6 to 0.92. On the other hand, these models have highlighted a variability in
the use of samples which results in a drop in performance during tests on statistically
independent databases requiring additional spectral acquisitions, leading today to the
presentation of this report project.