Shining Light on DNA Mutations through Machine Learning-Augmented Vibrational Spectroscopy.
Journal:
Analytical chemistry
Published Date:
Jun 4, 2025
Abstract
A method to directly predict the number of nucleic acid bases in a single-stranded DNA (ssDNA) or a genomic DNA has been proposed with a combination of Raman spectroscopy and an Artificial Neural Network (ANN) algorithm. In this work, the algorithm was trained by using the Raman spectroscopic signatures from a cohort of 32 ssDNAs. The algorithm could predict the number of bases in an unknown sequence with an value of more than 0.83. Chemical mutation using the hydroxylamine method was performed on a ssDNA and also a genomic herring sperm DNA, and the extent was monitored using optical absorbance measurements. The mutation of bases, such as cytosine, can introduce subtle alterations in the DNA structure, potentially leading to significant biological consequences, including neurodegenerative and epigenetic disorders. Also, during the mutation process, the unstable intermediate can undergo further transformation, converting bases such as cytosine to uracil, thus significantly altering the base-pairing properties of the DNA. A one-to-one correspondence was observed between the experimentally and computationally predicted mutated bases in both the single- and double-stranded DNA (dsDNA), thus opening up avenues for the detection of mutations in a diagnostic setup.
