Controlled Synthesis Of Carbon Dots And The Application Of Fluorescence Sensing Of Heavy Metals In Traditional Chinese Medicine Assisted By Machine Learning

In recent years,the application of traditional chinese medicine（TCM）in disease prevention and treatment has become increasingly popular.However,problems such as excessive levels of multiple heavy metals in the cultivation and processing of TCM not only affect its quality,but also limit its application development.The rapidly advancing sensing array technology can detect multiple heavy metals in complex samples with the advantages of low device dependence and abundant output signals.However,the construction of sensor arrays with electrochemical and photoelectrochemical techniques has problems such as complicated preparation of sensing elements and high instrumentation dependence,which limit their further applications.The sensor arrays constructed by fluorescence approach have the advantages of easy operation,low cost and high sensitivity,and play an important role in the simultaneous analysis and identification of multiple heavy metals in TCM.Carbon dots（CDs）are a kind of zero-dimensional fluorescent nanomaterials with excellent luminescence stability,excellent water solubility and biocompatibility,which can be used as ideal materials for constructing fluorescent sensor arrays.At present,the trial-and-error approach based on artificial experience is the main way to prepare CDs,however,this traditional strategy is inevitable to avoid subjectivity,randomness and coincidence problems.More importantly,the formation process and fluorescence properties of CDs are susceptible to multiple factors.At the same time,there is a lack of more suitable approaches to reveal the hidden laws between the synthesis conditions and the luminescence properties of CDs.Therefore,it is still necessary to develop efficient and rational approaches to investigate the main factors affecting the luminescence properties of CDs,and to achieve the controlled synthesis of CDs with ideal luminescence properties.Based on this,this thesis is devoted to the controlled synthesis of fluorescent CDs and its application in the analysis of heavy metal sensing arrays of TCM.The specific content is as follows:（1）Controlled synthesis of multicolor carbon dots under the guidance of machine learning.The formation process and photoluminescence properties of CDs can be affected by a variety of factors,and the ambiguous relationship between the synthesis conditions and luminescence properties of CDs has greatly hindered the development of the CDs preparation technology and innovative applications.Inspired by the application of machine learning（ML）in materials science and other fields,we proposed data-driven machine learning strategy to achieve the controlled synthesis of CDs.The strategy can explore the correlation between the reaction parameters and photoluminescence properties of CDs in multiple dimensions and optimize the synthesis route of CDs.First,we selected p-phenylenediamine,urea and citric acid as precursors and obtained a dataset of 270 kinds of CDs by varying the synthesis parameter preparation.Five of the experimental parameters were identified as the main input features,including precursor type and ratio,solvent type,reaction temperature and reaction time,and the maximum emission wavelength,stokes shift and fluorescence quantum yield of CDs as the output signals,from which a learning model was established.The experimental results demonstrated that the random forest（RF）model could accurately predict the photoluminescence properties of CDs in the test set and unknown samples with an accuracy of 80%.Most importantly,with the assistance of the permutation importance algorithm,it was clarified that solvent,precursor ratio and precursor type were major factors affecting the maximum emission wavelength,quantum yield and stokes shift of CDs,respectively.The effects of specific properties of reaction solvents on the fluorescence properties of CDs were further investigated,and it was demonstrated that boiling point,specific heat capacity and dielectric constant were major factors affecting the maximum emission wavelength,quantum yield and stokes shift of CDs,respectively.Finally,multicolor CDs were applied to efficient and secure information encryption with the advantages of large theoretical information capacity and abundant color information.This study indicated that the ML algorithm can screen the optimal reaction parameters for the preparation of multicolor CDs,provide reliable and optimal reaction conditions,thus optimize the preparation route and finally establish an intelligent and controllable synthesis platform for CDs.（2）Construction of machine learning-guided"chemical nose/tongue"sensor and the application to the identification of multiple heavy metal ions in traditional chinese medicine.At present,the sensor arrays based on fluorescence approach for multi-analyte identification and detection suffer from the problems of complex sensing element construction and high cost.At the same time,the number of sensing elements can seriously affect the sensitivity and discrimination ability of the array.Based on this,we constructed a CDs"chemical nose/tongue"sensor under the guidance of ML strategy to identify and detect heavy metals in TCM.Firstly,we used blue carbon dots（B-CDs）,yellow carbon dots（Y-CDs）and 5,10,15,20-tetra（4-aminophenyl）porphyrins（TAPP）with certain spatial location and coordination ability to construct the sensing element.The sensor array then had three independently emitting and significantly separated fluorescence emission peaks under excitation at 410 nm and could produce significantly different fluorescence response signals after addition of heavy metals.The experimental results demonstrated that the CDs"chemical nose/tongue"sensor can accurately identify12 kinds of heavy metal and binary mixtures of heavy metals,including Cu²⁺、Fe²⁺、As⁵⁺、Ni²⁺、Cr³⁺、Mn²⁺、Pb²⁺、Zn²⁺、As³⁺、Fe³⁺、Cd²⁺、Co²⁺.The RF model predicted a single heavy metal category in the test set with an accuracy of 94.5%,Cd²⁺and Pb²⁺concentrations with an accuracy of 91.3%and 93.2%,respectively,and a single heavy metal ion category in the unknown data set with an accuracy of 83.3%.The results showed that the model had good prediction performance and generalization ability when predicting unknown data.Finally,the accuracy of the RF model in predicting the binary mixed metal ion categories in radix codonopsis and honeysuckle was up to 100%and93.3%,and the accuracy in predicting the single metal ion and binary mixture in radix codonopsis and honeysuckle was 85.3%and 92.5%,respectively.It was shown that the prediction accuracy of the model was not affected by the complex detection environment.These results suggested that the CDs"chemical nose/tongue"sensor can rapidly identify and accurately quantify a variety of heavy metals in TCM with the assistance of the RF model,and the sensor has the advantages of low preparation cost and no complex chemical modifications.In summary,this thesis constructed an intelligent platform for the synthesis of multicolor CDs and a heavy metal"chemical nose/tongue"sensor based on ML strategy to realize the controlled synthesis of multicolor CDs and accurately predict the categories and concentrations of single and mixed heavy metal ions,which can expand the potential of ML for the controlled preparation of CDs and accurate detection of multiple analytes.