Application Of Machine Learning Algorithm Based On Fluorescence Signal In High Selectivity Rapid Analysis And Detection

Environmental disinfection and antibiotic use are common means for human to resist the threat of living environment,among which tetracycline antibiotics and hypochlorite are widely used to treat bacterial infections and prevent virus transmission.However,the abuse of antibiotics and the excessive disinfection of living environment also bring serious negative effects.These residual drugs will be washed into rivers and lakes with the cleaning of city streets and rainwater.Excessive drug stimulation poses a potential threat to the health of humans and animals.Therefore,the development of a highly selective detection platform for tetracycline antibiotics and hypochlorite in the environment is conducive to the risk assessment of water environment and provides an important reference for further treatment measures.To solve the above problems,several fluorescent probes specifically recognizing hypochlorite and tetracycline antibiotics were designed and synthesized in this study,and the machine learning algorithm was applied to the spectral analysis.The results show that the fluorescent probe based on machine learning algorithm has broad application prospects in complex environment and visual sensing.The main research contents are as follows:(1)Specific identification of hypochlorite by isomeric small molecule fluorescent probe.In this paper,four small isomers with highly similar structures were synthesized by regulating the methyl position.It was found that only one of these structures was specifically quenched by sodium hypochlorite in a specific solvent environment,and was not affected by other reactive oxygen species(ROS).The molecular weight and structure of several isomers were determined by high resolution mass spectrometry and nuclear magnetic resonance,and the molecular structure of the reaction products between probe molecules and sodium hypochlorite was further investigated.Density functional theory(DFT)was used to determine the molecular frontier orbital distribution and energy gap of the probe and the oxidation product,so as to verify the mechanism of luminescence of the probe and fluorescence quenching of the product under alkaline conditions.(2)Identification of hypochlorite by UV and fluorescence dual signal probes based on resonance energy transfer.Although small molecule fluorescent probes have certain advantages,the fluorescence quenching sensing probes based on single signal quenching still have disadvantages in selectivity and sensitivity.Therefore,a kind of purplish red carbon sphere with the absorption spectrum overlapping with the small molecule fluorescence emission spectrum was designed and synthesized.The carbon sphere and the stable small molecule fluorescence sphere formed the hypochlorite probe with the energy transfer mechanism.After the probe reacts with sodium hypochlorite,the purplish red color of the carbon sphere fades,and the small molecule fluorescence turns on,thus forming a hypochlorite probe with ultraviolet and fluorescent dual signals.Compared with the quenching mode of single signal,this recognition method greatly improves the reliability and sensitivity of the probe.(3)Recognition of hypochlorite by 3D sensor array based on machine learning.Compared with the above two conventional detection methods,the 3D array sensing based on machine learning performs better in the detection of substances with complex environments and highly similar structural properties.Only when the target substances are detected can all signals be changed accordingly.By improving the synthesis scheme of the above-mentioned small molecule fluorophore,an array probe was designed to simultaneously generate three kinds of characteristic signals(fluorescence enhancement,UV absorption characteristic peak shift and 3D fluorescence characteristic value change)in response to hypochlorite.To solve the problem of large amount of data and complex structure of multidimensional array,We apply the Hierarchical clustering analysis(HCA)and Density clustering analysis(DCA)methods in machine learning to the analysis of 3D array data.The results showed that the clustering effect of the same ROS was excellent and the array specificity was significant in a variety of ROS detection.Finally,the Principal component analysis(PCA)method in machine learning was used to reduce the dimensionality of complex three-dimensional data and fit the multiple linear regression equation for hypochlorite detection,so as to realize the multivariable linear hypochlorite detection.(4)Highly selective detection of doxycycline by bimetal doped Metal-organic framework(MOF)materials.Tetracycline antibiotics are easy to coordinate,aggregate and fluoresce in the channels of metal organic frame materials such as zinc(Zn)and aluminum(Al).Based on this mechanism,a kind of indium(In)and europium(Eu)double-doped metal-organic framework(MOF)probe was designed and synthesized.When the probe reacted with tetracycline doxycycline,the green fluorescence gradually changed to red fluorescence after a period of incubation,so as to detect doxycycline by specific ratio sensing.Finally,the probe was applied to real samples such as fish and urine.The recovery rate and relative standard deviation of the probe were within the error range,indicating that the probe has potential practical application value.(5)Online detection of tetracycline antibiotics based on machine learning.The accuracy of traditional visual sensing or smart phone recognition mode has a large error.In this study,an organic ligand emitting blue light was designed and synthesized to coordinate with europium(Eu)ions to form a stable ratio fluorescent probe.The cross-platform computer vision library in Python is used to extract the RGB values in the fluorescence visualization images,and the RGB channel values in the images are preprocessed to construct the sample fingerprint information feature values.Then,k-means clustering model(KCM)and Hierarchical clustering model(HCM)were constructed by machine learning method to analyze the regularity of fingerprint information of samples with different concentrations.The clustering results showed good clustering characteristics of similar concentration samples,which provided reliable support for rapid detection.A more accurate detection of tetracycline can be achieved by analyzing the color change of images through multiple linear regression model.In conclusion,the analysis and detection mode combining machine learning and fluorescence sensing in complex environment and visual sensing field has more advantages and potential than the traditional fluorescence probe detection method.