Studies On Recognition Of Small-molecule Binding Sites In Proteins Based On SERS And Deep Learning

Identification of small-molecule binding sites in proteins is of great significance in analysis of protein function and drug design.Modified sites can be recognized via proteolytic cleavage followed by liquid chromatography-mass spectrometry（LC-MS）,however,this always impeded by the complexity of peptide mixtures.Here,we presented a strategy based on surface-enhanced Raman spectroscopy（SERS）and deep learning to identify binding sites of proteins and small molecules.Without any secondary tag,the small molecules are directly complexed with proteins.After proteolysis and LC,SERS signals of all LC-fractions are collected and inputted into the model,whereby the fractions containing the small-molecule-modified peptides can be recognized by the model and sent to MS/MS to identify the binding site（s）.We successfully identified the binding sites of small molecules in different proteins using the strategy,indicating the potential of deep learning-based SERS in protein-active molecule screening and drug discovery.Therefore,based on the combination of SERS and deep learning technology,this paper focuses on the identification of binding sites of different proteins and small molecules,and the main research works are as follows:（1）Aiming at the SERS detection of the protein enzymatic hydrolyzate separated by LC,a SERS multi-well plate system based on Fe²⁺-induced adsorption was designed and fabricated to realize the detection of all components in the HPLC fractions.The SERS multi-well plate was fabricated from mass-produced Si@Ag/Au substrates（one substrate to one well）with good reproducibility,stability（RSD=3.32%）,and SERS-enhanced performance（AEF=1.2×10⁹）.The whole LC fractions were collected in the SERS multi-well plate embedded with Si@Ag/Au substrates for SERS detection,and the peptides in the fractions could be uniformly adsorbed on the substrates and localized to hot spots by Fe²⁺ions induction.Through the SERS detection of the mixture of insulin B and bacitracin after LC separation,the potential of this platform to be applied to the SERS detection of LC separation products of protein enzymatic hydrolysates was further verified.（2）Taking the coordination-bound fomepizole and ADH as the research objects,a label-free SERS and deep learning-based strategy was created for the identification of binding sites for ADH and fomepizole.The SERS spectral library of small molecules such as fomepizole was obtained through the Si@Ag/Au substrate to train the deep learning model based on the Res Net network.After validation,the model had a specificity of 96%and a sensitivity of 88%.The Fe²⁺ions-induced adsorption-based SERS multi-well plate system was used to detect the fractions of the ADH-fomepizole mixture enzymatic hydrolysate after HPLC separation.The SERS spectra were input into the deep learning model for prediction,and the fractions containing the fomepizole SERS signal were obtained,MS/MS data confirmed the presence of fomepizole and zinc-binding peptides,indicating that the strategy successfully identified the binding sites.（3）The experiments focused on covalently bound ibrutinib and Bruton’s tyrosine kinase（BTK）,and identified ibrutinib-modified peptides in BTK based on label-free SERS and deep learning strategies.Small-molecule SERS spectral library were obtained through the Si@Ag/Au substrate to train the deep learning model.After verification,the specificity of the model was 98%and the sensitivity was 88%.The fractions of the BTK-ibrutinib mixture enzymatic hydrolyzate after HPLC separation were detected by the Fe²⁺ions-induced adsorption-based SERS multi-well plate system,and the SERS spectra of the fractions were input into the deep learning model to obtain fractions containing the ibrutinib SERS signal.The MS/MS results of this fraction showed that the fraction contained ibrutinib-modified peptides,demonstrating the feasibility of this strategy in identifying covalent binding sites for small molecules and proteins.