# Improving the Annotation for Spatial Proteomics: A Computational Approach to Enhance Molecular Characterization of Thyroid Nodules

## TL;DR

This paper introduces a computational workflow that improves MALDI-MSI analysis of thyroid tissue by using digital pathology to select cell-rich regions, enhancing diagnostic accuracy and biomarker discovery.

## Contribution

A novel automated workflow integrating digital pathology with spatial proteomics to enhance MALDI-MSI analysis of thyroid lesions.

## Key findings

- The pixel classifier reduced interfering signals by 15% and increased the signal-to-noise ratio of tryptic peptides by 37%.
- The approach detected 9-24% more m/z signals and improved spectral clustering for distinguishing histopathological regions.
- ROC analysis showed a 50% increase in discriminatory m/z features across thyroid nodule diagnoses compared to conventional methods.

## Abstract

The present work proposes a reproducible and automated
workflow
for integrating digital pathology in matrix-assisted laser-desorption
ionization mass spectrometry imaging (MALDI-MSI) data analysis, using
thyroid tissue as a proof-of-concept application. MALDI-MSI has shown
promise in the molecular characterization of thyroid neoplasms. Yet
challenges remain in minimizing signal interferents and improving
diagnostic discrimination. In this study, we propose an interdisciplinary
approach integrating digital pathology with spatial proteomics to
enhance MALDI-MSI analysis of thyroid lesions from formalin-fixed
paraffin-embedded tissue sections. We trained a pixel classifier to
automatically select cell-rich regions of interest (ROIs) from hematoxylin
and eosin-stained tissue microarrays, reducing interference from colloid-rich
areas. The proteomics signals obtained with the pixel classifier (PC) were compared with those obtained from the full core
(FC) and those manually annotated by the pathologist
(PAT). The results showed that PC ROIs significantly decreased interfering signals (15%) while increasing
the signal-to-noise ratio of tryptic peptides (+37%). Indeed, we detected
a greater number of m/z signals
(between 9 and 24%), improving the spectral clustering by means of
principal component analysis to distinguish different histopathological
regions. Receiver operating characteristic (ROC) analysis further
confirmed the improved classification power, with a 50% increase in
discriminatory m/z features across
different thyroid nodules diagnosis compared to conventional FC and PAT data. Using a PC to select cell-specific regions globally enhances reproducibility,
reduces operator workload, and optimizes MALDI-MSI workflows. Altogether,
the proposed approach paves the way for more accurate molecular characterization
of thyroid neoplasms and holds potential for improving biomarker discovery
and diagnostic precision in clinical pathology.