Single-Cell Biophysics: Measurement, Modulation, and Modeling

Single-Cell Biophysics: Measurement, Modulation, and Modeling

Poster Abstracts

41-POS

Board 21

The Correspondence between Raman Spectra and Total Omics

Koseki J. Kobayashi-Kirschvink

, Hidenori Nakaoka

, Arisa Oda

, Kunihiro Ohta

2,3

, Yuichi

Wakamoto

The University of Tokyo, Meguro-ku, Tokyo, Japan,

The University of Tokyo, Meguro-ku,

Tokyo, Japan,

The University of Tokyo, Bunkyo-ku, Tokyo, Japan.

Fluorescent reporters have been the de facto standard tool of

in vivo

research in the past two

decades, allowing us to monitor specific molecules in living cells with high precision. However,

the cell is a complex system with numerous molecules and interactions, and due to the specificity

of fluorescent reporters, they often fail to elucidate the cellular behavior as a whole. On the other

hand, omics techniques such as next generation sequencing or mass-spectrometry give us much

more comprehensive molecular information about the cell, but are inherently destructive. No

existing technique allows us to further analyze the dynamic molecular changes occurring at the

single-cell level.

In contrast, Raman micro-spectroscopy has advanced in recent years, and is one of the very few

imaging techniques that can potentially report on whole-cell molecular compositions at the

single-cell level in both comprehensive and non-destructive manners. However, the complexity

of Raman spectra has hampered its interpretation and thus its routine use. Previous attempts on

interpreting spectra mostly relied on preparing Raman spectral databases of purified materials,

which is usually time-consuming and laborious, especially when dealing with biological

samples. Here, we propose a method that can possibly circumvent the preparation of such

databases, and instead understand how Raman spectra and “total” omics correspond to each

other. Our method is based on the linear relationship between the two data sets that holds in

principle, and actively employs the intrinsic low-dimensionality of gene/protein expression

profiles. Example data of Raman spectra and mRNA-seq data of

Schizosaccharomyces

Pombe

under various stress conditions are obtained, and using supervised machine learning

algorithms such as the partial least squares regression, it is shown that if certain conditions are

met, mRNA-seq data can possibly be predicted from Raman spectra.