Machine learning concept in de-spiking process for nuclear resonant vibrational spectra - automation using no...

Area of Science:

• Spectroscopy
• Biochemistry
• Materials Science

Background:

• Nuclear resonant vibrational spectroscopy (NRVS) provides site-specific vibrational data crucial for biochemical research.
• NRVS is sensitive to weak signals but requires extensive scans and can be affected by occasional spikes.
• Spikes in NRVS scans can lead to artifacts in partial vibrational density of state (PVDOS) calculations, potentially causing misinterpretation of structural information.

Purpose of the Study:

• To develop a fully automated procedure for identifying and smoothing occasional spikes in NRVS spectra.
• To improve the reliability of PVDOS analysis derived from NRVS data.
• To provide a self-contained method that does not require external parameters or data from other scans.

Main Methods:

• Utilized machine learning principles to create an automated de-spiking algorithm for NRVS spectra.
• The procedure analyzes statistical information inherent to each individual NRVS scan.
• Developed an R subroutine for efficient batch processing of multiple NRVS scans.

Main Results:

• Successfully implemented a fully automated process for screening and smoothing spiky points in NRVS data.
• The developed method operates without external parameters or reliance on other scans.
• An R code is provided for practical application in processing large datasets.

Conclusions:

• The presented automated de-spiking procedure offers a robust solution for handling artifacts in NRVS spectra.
• This work represents the first automated approach to de-spiking NRVS data without external parameters.
• The method enhances the utility of NRVS for accurate structural and vibrational analysis in scientific research.