**4. Discussion**

As seen in existing experimental studies of single-molecule SERS of DNA and proteins [64,65], researchers have gradually developed methods to characterize a single biological molecule with Raman spectra at the time scale of interaction with the environment and other molecules and structures. The simulation methods can create spectral libraries for vibrations of molecular species vs. different types of interaction to specify the interaction's type and strength.

We connect the importance of the proposed theoretical calculations with the development of machine learning in various fields of biomolecular modeling and simulation.

AI algorithms are employed in protein structure prediction based on the data in the form of trajectories that can be used to train machine learning algorithms [66] or applied to the analysis of the SERS/TERS spectra measurements of the presence of drug-in-solution [67]. All such algorithms need data to learn from. Another aspect is that machine learning methods generally require extra preprocessing or knowledge of key features in data of measured structures to handle large-scale data. The proposed simulation approach can be the method to produce a sufficient amount of data to train machine learning algorithms, not only on structural differences in spectra of measured molecules, but also on conformational and interaction changes.

The data on the changes in spectra relative to the interactions in the measured system will provide knowledge of key features in spectra as a part of data sets. Such data in combination with machine learning can reduce spectra accumulation and averages over a few seconds' acquisition time applied currently, and open the way to a single pulse measurement of vibrational spectra. In the present study, the possibility of the simulation of Raman spectra as depending on the transient interactions between the parts of a SERS system is investigated. A simple system that contains the molecule itself, an Au nanoparticle as Raman enhancer and graphene sheet with nanopore that can control the localization of the measured molecule relative to enhancer if it is attached to graphene, is selected. The position-controlled agglomeration of nanoparticles can be a key to the successful collection of SERS spectra [39] of proteins and DNA as shown in the experiment. We suppose that when the positioning of nanoparticles can be controlled, and detailed knowledge about spectral changes due to interaction with the nanoparticles and substrate such as graphene with nanopore can be used for the training of machine learning algorithms, fast transient single-molecule SERS measurements can be realized.

The question of extension of the achieved accuracy and simulation' sensitivity of building block molecules to the 3D structures of biomacromolecules such as nucleic acids and proteins are also important for applications. In the case of DNA molecules, a slow translocation of the single or double strand through the nanopore has a timescale around a microsecond in the experiment. The duration time of the nucleotide presence in the vicinity of the graphene nanopore with attached Au NPs should be far below the conventional time resolution of current SERS and TERS techniques that accumulate pulse spectra. However, the single pulse measurement will ge<sup>t</sup> a spectrum of the nucleotide located in the vicinity and currently enhanced by agglomerated Au NPs at the translocation time. Each such single pulse spectrum is relatively weak and has a fingerprint of the interaction with the enhancer NP, therefore, preliminary knowledge of the spectral changes relative to interaction could help to isolate the signal in the spectrum from the noise level.
