# Label-free estimation of regulatory T cell activation markers using Raman spectroscopy with machine learning

## TL;DR

This paper introduces a non-invasive method using Raman spectroscopy and machine learning to estimate regulatory T cell activation without the need for expensive and invasive lab tests.

## Contribution

A novel label-free method for monitoring regulatory T cell activation using Raman spectroscopy and machine learning is developed and validated.

## Key findings

- An L1-regularized least-squares model accurately estimated activation markers from Raman spectroscopy data.
- The model was validated on external donors, showing robust performance in predicting biomarker values.
- The method is suitable for integration into cell manufacturing devices for real-time monitoring.

## Abstract

Regulatory T cells are a class of T lymphocytes which respond to activation signals by expanding their cell numbers, and whose culturing and expansion are of significant clinical interest. Cellular activation states are used to inform process control decisions such as restimulation and can be probed with experimental measurements of cell surface markers. However, these measurements are expensive, time-consuming, and invasive, and an urgent need exists for devising a non-invasive method for activation state monitoring that could be deployed on-line. Raman spectroscopy is a label-free and information-rich optical method that, when coupled to data analytical methods, can ameliorate these experimental issues. In this work, we quantitatively estimated experimental measurements of regulatory T cell activation markers with high accuracy. We simulated a clinical manufacturing setting by building an \documentclass[12pt]{minimal}
\usepackage{amsmath}
\usepackage{wasysym}
\usepackage{amsfonts}
\usepackage{amssymb}
\usepackage{amsbsy}
\usepackage{mathrsfs}
\usepackage{upgreek}
\setlength{\oddsidemargin}{-69pt}
\begin{document}$${L}_{1}$$\end{document}-regularized least-squares model with spectroscopic data from six regulatory T cell donors. Then, we validated the constructed model by accurately estimating different experimental measurements of biomarker values from two external donors, unseen by the model. We have devised a robust program to effectively estimate the activation state of regulatory T cells. We anticipate our method to be used with on-line Raman probes integrated into cell manufacturing devices for label-free monitoring of these processes.

The online version contains supplementary material available at 10.1038/s41598-025-16002-8.