EURO 2024 Copenhagen
Abstract Submission

EURO-Online login

2659. Shapley value analysis for revealing molecular signature in Raman Spectra

Invited abstract in session TD-27: Feature attribution and selection for XAI, stream Mathematical Optimization for XAI.

Tuesday, 14:30-16:00
Room: 047 (building: 208)

Authors (first author is the speaker)

1. Marco Piazza
Computer Science, University of Milan-Bicocca
2. Enza Messina
DISCo - Department of Informatics, Systems and Communication, University of Milano Bicocca
3. Mauro Passacantando
Department of Business and Law, University of Milano-Bicocca

Abstract

Raman Spectroscopy is a technique based on the inelastic scattering of a monochromatic light beam used to observe low-frequency modes in a target molecular system. The scattering pattern can be viewed as a sort of ”fingerprint” that encodes information about the chemical composition of a given target, including presence, concentrations, and interactions among molecules.
Recently, Raman spectra of biofluids have been used for medical diagnosis, and Machine Learning and Deep Learning models have been proposed for the development of noninvasive tests for early diagnosis.
However, the lack of interpretability of these models makes decoding and interpreting the Raman spectral fingerprint a challenging task.
In this work, we explore the application of eXplainable Artificial Intelligence (XAI) methods to extract explanations aimed at the possible identification of new biomarkers.
Among the possible approaches, feature attribution methods seem to be the most promising for analyzing spectroscopic data, and we will focus our attention on Shapley value methods based on cooperative game theory. The computational results of a comparative analysis will be presented and discussed.

Keywords

Status: accepted

Back to the list of papers