“Our two-step neural network is designed to perform two key steps necessary for the quantification of protein secondary structures from discrete frequency data,” Ghosh said to Spectroscopy. “It reconstructs the full spectra from seven wavenumbers, and it then predicts areas under the curve (AUCs) of underlying spectral components for structural quantification, which is typically done using band fitting.”
The researchers’ approach required data from just seven discrete wavenumbers rather than densely sampled spectral data. The model then performs an interpolation step to reconstruct the full spectral profile and predict the AUCs for protein components. Over the course of their study, they found that this process was found to be over 3,000 times faster than traditional spectral fitting (1).
Ghosh also discussed that their model does not rely on high-density spectral sampling. As a result, the data sets are smaller, more manageable, and less demanding in terms of storage and processing power. Importantly, the predictive accuracy of the model closely matched the results of traditional methods, meaning researchers can trust the insights generated by the neural network (1).
“To assess robustness under varying signal-to-noise (S/N) ratio conditions, the model’s performance was compared against traditional Gaussian fitting,” Ghosh said to Spectroscopy. “Across different S/N ratio levels, the model consistently outperformed Gaussian fitting, showing lower relative mean absolute error (MAE) and less degradation in predictive accuracy as noise increased.”
By significantly streamlining both data acquisition and interpretation, the work of Ghosh and his collaborators at The University of Alabama and the University of Illinois Urbana-Champaign indicates a potential way for biomedical researchers to approach protein structural analysis in complex biological samples.
Spectroscopy sat down with Ayanjeet Ghosh, as well as his coauthor Rohit Bhargava, to discuss their study more in-depth. Over the next week, we will release the three-part interview. In Part I, Ghosh and Bhargava will discuss their interest in using discrete frequency infrared (IR) imaging to analyze protein secondary structure. Part II will discuss how machine learning (ML) is used in data analysis and go into more detail about the model they developed in their study. Part III will see Ghosh and Bhargava discuss the key performance metrics they used to evaluate their model, and what the future of neurodegenerative disease research looks like.
We are looking forward to sharing their insights with all of you soon.
References
- Edmonds, H.; Mukherjee, S. S.; Holcombe, B.; et al. Quantification of Protein Secondary Structures from Discrete Frequency Infrared Images Using Machine Learning. Appl. Spectrosc. 2025, ASAP. DOI: 10.1177/00037028251325553
- Hughes, C.; Clemens, G., Bird, B.; et al. Introducing Discrete Frequency Infrared Technology for High-Throughput Biofluid Screening. Sci. Rep. 2016, 6, 20173. DOI: 10.1038/srep20173
