An intelligent method for constructing a priority control list for serotonin reuptake inhibitors

Abstract

The present invention relates to a method for intelligently constructing a list of preferred 5-hydroxytryptamine reuptake inhibitors. The method comprises the following steps: constructing a functional receptor protein of a 5-hydroxytryptamine reuptake inhibitor by a homology modeling method, screening its key protein of interference with human health based on a harmful outcome pathway method, and calculating the binding energy between it and the receptor protein. The method further calculates the comprehensive effect value as the dependent variable by using an information weighting method, screening the main features of the target as the independent variable based on a variance filtering coupled Pearson correlation coefficient method, realizing the establishment of a data set, and coupling a GRU neural network with a 1D-CNN neural network to realize the intelligent construction of a list of preferred 5-hydroxytryptamine reuptake inhibitors. The present invention aims to overcome the inability of current traditional methods to comprehensively evaluate the comprehensive selection effect of the functionality of 5-hydroxytryptamine reuptake inhibitors and interference with human health, and to realize the intelligent construction of a list of preferred 5-hydroxytryptamine reuptake inhibitors with high priority by establishing a deep learning model.

Description

Technical Field

[0001] The present invention relates to a method for intelligently constructing a 5-hydroxytryptamine reuptake inhibitor optimal control list. Background Art

[0002] The worldwide prevalence of depression has increased, leading to a surge in the prescription and consumption of antidepressants. Based on the monoamine neurotransmitter hypothesis, three classes of serotonin reuptake inhibitors (5-HT-RIs): selective serotonin reuptake inhibitors (SSRIs), dual serotonin-norepinephrine reuptake inhibitors (SNRIs), and triple serotonin-norepinephrine-dopamine reuptake inhibitors (SNDRIs)—all exert their pharmacological effects by competing with 5-HT for the serotonin transporter (SERT) binding site, acting as SERT inhibitors to stimulate intracellular 5-HT efflux. However, it has been observed that threonine (T) at position 276 of SERT can mutate to aspartic acid (D), further inducing the mutation of isoleucine (I) at position 425 to valine (V), thereby inhibiting its transport capacity and conferring drug resistance. 5-HT reuptake inhibitors (5-HT reuptake inhibitors), as non-antibiotic drugs and persistent micropollutants, have a long half-life in the human body, and their active ingredients are not fully metabolized. Municipal and hospital wastewater treatment plants are one of the main sources of 5-HT reuptake inhibitor release into the environment. Existing wastewater treatment processes are inefficient in removing 5-HT reuptake inhibitors, leading to their widespread detection in urban and non-urban water bodies and drinking water.

[0003] 5-HT reuptake inhibitors (5-HT reuptake inhibitors) have biological targeting effects, and the potential ecological and environmental safety hazards and human health risks they may cause require sufficient attention. This is particularly evident in terms of chronic biological toxicity. Studies have shown that environmental concentrations of 5-HT reuptake inhibitors can affect the endocrine system of female Japanese medaka or goldfish, the behavior, development, and reproduction of zebrafish; significantly inhibit acetylcholinesterase activity in mussels or changes in zebrafish neurons, causing neurotoxic effects; and hinder goldfish's sense of smell by interfering with the generation, transmission, and processing of olfactory signals, affecting their sensory behavior. Therefore, long-term exposure to 5-HT reuptake inhibitors can result in drug residues and bioaccumulation in tissues such as muscle, liver, brain, and plasma in target or non-target aquatic organisms, rats, and even humans. These drugs can also significantly interfere with physiological changes such as reproduction, growth, behavioral responses, metabolism, nervous system, endocrine system, sensory perception, and oxidative stress responses.

[0004] The Gated Recurrent Unit (GRU) is a deep learning method that, compared to traditional recurrent natural network (RNN) neural network methods, establishes closer connections between independent molecular features, effectively avoiding problems such as vanishing gradients and long-term dependencies. Compared to long short-term memory (LSTM) neural network methods, the GRU model has a simpler structure and higher computational efficiency. Studies have found that GRUs coupled with convolutional neural networks (CNNs) or LSTM models can effectively predict depression, breast cancer, protein structure, PM2.5 concentrations, and water quality. The development of a list of "high-priority" chemicals or eco-markers for toxicological risks from pharmaceuticals and personal care products (PPCPs) based on a "prioritization strategy" has attracted the attention of global environmental monitoring programs. Estimating potential impacts and prioritizing the list based on toxicological information from PPCP occurrence and toxicity testing or quantitative structure-activity relationship (QSAR) models can effectively assess and control the human and environmental risks of a large number and variety of PPCPs. For example, a control list of 15 "high-priority" chemicals was developed by studying the toxicity quotient (TQ) of PPCPs in surface rivers near Washington, D.C. However, the types and amount of data on 5-HT reuptake inhibitors studied in this paper are relatively small, and traditional machine learning models are not applicable due to problems such as overfitting and insufficient generalization. In addition, 5-HT reuptake inhibitors are non-antibiotic drugs, and there is a lack of systematic integrated research on the identification and evaluation of their human health risks. Therefore, how to achieve the construction of a priority control list for emerging pollutants such as 5-HT reuptake inhibitors with a small sample size of data, and to achieve the prediction of the comprehensive effects of functionality and human health interference, is a technical problem that researchers in this field urgently need to solve. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide a method for constructing a 5-HT-RI optimal control list based on a CNN-GRU neural network, construct a functional receptor protein of a 5-hydroxytryptamine reuptake inhibitor by a homology modeling method, screen its key proteins that interfere with human health based on the harmful outcome pathway method, calculate the binding energy between the 5-hydroxytryptamine reuptake inhibitor and the receptor protein by a molecular dynamics method, further calculate the comprehensive effect value as the dependent variable by using the information weight method, screen the main features of the 5-hydroxytryptamine reuptake inhibitor as an independent variable based on the variance filtering coupled Pearson correlation coefficient method, realize the establishment of a data set, and based on the SE module idea, establish a CNN-GRU deep learning network structure by coupling the GRU neural network through the 1D-CNN neural network, and use the K-fold crossover method to enhance the generalization ability of the network, thereby realizing the construction of a 5-hydroxytryptamine reuptake inhibitor optimal control list. The present invention is achieved in this way: a method for constructing a 5-hydroxytryptamine reuptake inhibitor optimal control list based on a CNN-GRU neural network:

[0006] The steps include:

[0007] 1) Determine the type and molecular structure of 5-HT reuptake inhibitor drugs and use homology modeling to construct the amino acid sequences of the pharmacodynamic receptor protein (SERT) and its double-mutated drug-resistant receptor protein structure (I425V-T276D-SERT), i.e., the functional (pharmacodynamic and drug-resistant) protein structures;

[0008] 2) Based on the Reactome database, we used the adverse outcome pathway method to construct the human health interference pathway of 5-HT reuptake inhibitors, screened the key receptor proteins involved in olfactory toxicity, neurotoxicity transmission, and the positive feedback regulation of the gut-brain axis by intestinal microorganisms, and used molecular dynamics methods to calculate the binding free energy between 5-HT reuptake inhibitor molecules and key receptor proteins. The binding free energy was normalized and the information weight method was used to calculate the comprehensive effect value.

[0009] 3) Gaussian 09 and Chembiodraw 12.0 software were used to calculate the molecular features of 5-HT reuptake inhibitors, and the linear correlation between molecular features was calculated based on the variance-filtered coupled Pearson correlation method, and 10 major molecular features with reasonable correlation were selected;

[0010] 4) Using the 10 main molecular characteristics of 5-HT reuptake inhibitors as independent variables and the comprehensive effect value of their functionality and human health interference as the dependent variable, a data set for neural network learning was constructed;

[0011] 5) Use K-fold crossover to divide the dataset into training set and test set, and define hyperparameters such as number of iterations, number of folds, loss function, etc.

[0012] 6) Through the four-layer 1D-CNN neural network, combined with the GELU activation function and normalization layer, the data is expanded and contracted, thereby amplifying the features with greater influence. Finally, it passes through the GRU neural network and performs back propagation through the loss function;

[0013] 7) Verify and predict the training results through the test set. If the requirements are met, the trained neural network is output to achieve intelligent prediction and construction of the optimal control list of 5-HT reuptake inhibitors;

[0014] In step 2), the calculation formula for binding free energy is:

[0015] ΔG b =ΔG vdW +ΔG E +ΔG PS +ΔG SASA (1)

[0016] Where ΔG vdW is the van der Waals force, ΔG E is the electrostatic force, ΔG PS is the polar solvation energy, ΔG SASA Yes, SASA can;

[0017] In the normalization process, drug resistance, olfactory toxicity, and neurotoxicity key protein binding energy were used as positive indicators, while drug efficacy and intestinal microbial interference key protein binding energy were used as negative indicators. The normalization formula is:

[0018] Positive indicators:

[0019] Contrarian indicators:

[0020] Where i represents a 5-HT reuptake inhibitor molecule (i = 1, 2, ..., 17), j represents a key receptor protein (j = 1, 2, ..., 9), and Z ij + Z is a positive indicator after normalization of 5-HT reuptake inhibitor resistance, olfactory toxicity or neurotoxicity. ij - represents the reverse index after normalization of the efficacy of 5-HT reuptake inhibitors or intestinal microbial interference, X ij represents the binding energy value between the i-th molecule and the j-th protein;

[0021] In step 3), the Pearson correlation coefficient can measure the linear correlation between two random variables, and the calculation formula is:

[0022]

[0023] Where cov(a,b) represents the covariance of the values ​​of the molecular characteristics a and b of the 5-HT reuptake inhibitor, σa and σb represent the standard deviations of the values ​​of the molecular characteristics a and b of the 5-HT reuptake inhibitor, respectively, and E represents the mathematical expectation;

[0024] In step 5), the loss function is defined as:

[0025]

[0026] Where y m Indicates the effect value of the current molecule, p m Represents the predicted value of the current molecule, and m represents the number of molecules that need to be predicted. By taking the absolute value and then accumulating it, the accuracy of the prediction of each molecule can be effectively guaranteed;

[0027] In step 6), the data is first expanded through a layer of 1D-CNN neural network, GELU and normalization layers, and then the data is contracted through a layer of 1D-CNN neural network. This process is repeated, and finally the effect value is predicted through the GRU neural network.

[0028] By calculating the loss function and realizing back propagation, the purpose of accurately predicting the molecular effect value of 5-HT reuptake inhibitor is achieved.

[0029] Compared with the existing technology, the present invention has the beneficial effect of overcoming the inability of current traditional methods to comprehensively evaluate the comprehensive selective effects of serotonin reuptake inhibitor functionality and interference with human health. By establishing a CNN-GRU neural network model, the intelligent construction of a high-priority serotonin reuptake inhibitor priority control list is realized. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 It is a schematic diagram of a method for intelligently constructing a priority list of 5-HT reuptake inhibitors;

[0031] Figure 2 This is a schematic diagram of the structure of the CNN-GRU neural network used to construct the 5-HT reuptake inhibitor priority list;

[0032] Figure 3 This is a schematic diagram comparing the molecular effect values ​​and predicted values ​​under the comprehensive effect model of 5-HT reuptake inhibitors;

[0033] Figure 4 This is a diagram showing the relative errors between the molecular effect values ​​and the predicted values ​​under the single-effect model of eight 5-HT reuptake inhibitors; DETAILED DESCRIPTION

[0034] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0035] Example

[0036] See also Figure 1 , a method for intelligently constructing a priority list of 5-HT reuptake inhibitors, comprising:

[0037] 1) Determine the types and molecular structures of 17 common 5-HT reuptake inhibitor drugs, and query and obtain the complete amino acid sequence of the SERT protein structure based on the NCBI database (https: / / www.ncbi.nlm.nih.gov / protein / NP_001036.1). Using this as a template, using the homology modeling method, the complete amino acid sequence of the SERT protein and its double mutant protein (276 threonine mutations to aspartic acid, further induced 425 isoleucine mutations to valine) were submitted to the SWISS-MODEL server to construct the amino acid sequences of the pharmacodynamic receptor protein (SERT) and its double mutant drug-resistant receptor protein structure (I425V-T276D-SERT), i.e., the functional (pharmacological and drug-resistant) protein structure of the 5-HT reuptake inhibitor;

[0038] 2) Using the Reactome database, we searched for the olfactory signaling pathways of the sensory system (Reactome: R-HSA-381753.6) and the neurotransmitter release and synthesis pathways of the nervous system (Reactome: R-HSA-264622.2 and R-HSA-372519.5) to screen for key receptor proteins involved in the transmission of olfactory toxicity and neurotoxicity in humans. Furthermore, based on the principle of microbial positive feedback regulation of the gut-microbiota-brain axis, we screened for key receptor proteins that interfere with gut microbial activity. Furthermore, we used the adverse outcome pathway (AOP) approach to construct pathways for the human health interference of serotonin reuptake inhibitors (olfactory toxicity, neurotoxicity, and gut microbial interference). The molecular initiation of olfactory toxic AOPs is the formation of a complex (OR_GNAL) between the olfactory receptor protein (OR) and the guanine nucleotide-binding protein G(olf) α-subunit (GNAL), which then binds to the odorant molecule isovaleric acid. The key event is the activation and binding of adenylate cyclin type III (ADCY3). The deleterious outcome is the inhibition of both OR_GNAL and ADCY3 activity, interrupting the initiation of olfactory signals. The molecular initiation of neurotoxic AOPs is the acetylation of choline by O-acetylcholine transfer protein (ChAT) to acetylcholine. The key event is the hydrolysis of acetylcholine by acetylcholinesterase (AChE). The deleterious outcome is the inhibition of ChAT and AChE activity, inducing neurotoxicity. The molecular initiation of gut microbial disruptive AOPs is the transmembrane transport of butyrate by monocarboxylate transporters (MCTs). The key event is binding to free fatty acid receptors (FFARs) and the inhibition of histone deacetylase (HDAC) activity. The deleterious outcome is involvement in positive feedback regulation of the central nervous system. The molecular dynamics method was used to calculate the binding free energy between the 5-HT reuptake inhibitor molecules and the above-mentioned key receptor proteins, and the binding free energy was normalized. On this basis, the information weight method was used to calculate the comprehensive effect value;

[0039] 3) Gaussian 09 and Chembiodraw 12.0 software were used to calculate 29 molecular features of 5-HT reuptake inhibitors, including geometric parameters, electronic parameters, physicochemical parameters, spectral parameters, and topological parameters. The linear correlation between molecular features was calculated based on the variance-filtered coupled Pearson correlation method, and 10 main molecular features with reasonable correlation were selected;

[0040] 4) The 10 molecular features of 5-HT reuptake inhibitors and their corresponding comprehensive effect values ​​were used as the data set. The number of iterations was set to 1000, and cross-validation was performed using the K-fold crossover method to ensure that the data would not be overfitted.

[0041] 5) Set the first layer of CNN neural network to expand the data set to 320*64, the second layer of CNN neural network to shrink the data set to 16*64, the third layer of CNN neural network to expand the data set to 64*64 again, the fourth layer of CNN neural network to compress the data to 1*64, and filter the data to 1*16 through the GRU neural network, and set the output to 1;

[0042] 6) Ensure the accuracy and universality of the output neural network through verification of multiple training sets and test sets, as well as multiple iterations.

[0043] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for intelligently constructing a priority list of serotonin reuptake inhibitors, characterized in that: The steps include: 1) Identify the types and molecular structures of 5-HT reuptake inhibitors and use homology modeling to construct the amino acid sequences of the pharmacodynamic receptor protein SERT and its double-mutated drug-resistant receptor protein structure I425V-T276D-SERT, i.e., the functional protein structures; 2) Based on the Reactome database, we used the adverse outcome pathway method to construct the human health interference pathway of 5-HT reuptake inhibitors, screened the key receptor proteins involved in olfactory toxicity, neurotoxicity transmission, and the positive feedback regulation of the gut-brain axis by intestinal microorganisms, and used molecular dynamics methods to calculate the binding free energy between 5-HT reuptake inhibitor molecules and key receptor proteins. The binding free energy was normalized and the information weight method was used to calculate the comprehensive effect value. 3) Gaussian 09 and Chembiodraw 12.0 software were used to calculate the molecular features of 5-HT reuptake inhibitors. The linear correlation between molecular features was calculated based on the variance-filtered coupled Pearson correlation method, and 10 major molecular features with reasonable correlation were selected. 4) Using the 10 main molecular characteristics of 5-HT reuptake inhibitors as independent variables and the comprehensive effect value of their functionality and human health interference as the dependent variable, a data set for neural network learning was constructed; 5) Use K-fold crossover to divide the dataset into training set and test set, and define hyperparameters such as number of iterations, number of folds, loss function, etc. 6) Through the four-layer 1D-CNN neural network, combined with the GELU activation function and normalization layer, the data is expanded and contracted, thereby amplifying the features with greater influence. Finally, it passes through the GRU neural network and performs back propagation through the loss function; 7) Verify and predict the training results through the test set. If the requirements are met, the trained neural network is output to achieve intelligent prediction and construction of the optimal control list of 5-HT reuptake inhibitors; In step 2), the calculation formula for binding free energy is: （1） Where, It's the van der Waals force. is the electrostatic force, is the polar solvation energy, Yes, SASA can; In the normalization process, drug resistance, olfactory toxicity, and neurotoxicity key protein binding energy were used as positive indicators, while drug efficacy and intestinal microbial interference key protein binding energy were used as negative indicators. The normalization formula is: Positive indicators: (2) Contrarian indicators: (3) Where i represents the 5-HT reuptake inhibitor molecule, i=1,2,...,17, j represents the key receptor protein, j=1,2,...,9, Z ij + Z is a positive indicator of normalized drug resistance, olfactory toxicity, or neurotoxicity to 5-HT reuptake inhibitors. ij - represents the reverse index after normalization of the efficacy of 5-HT reuptake inhibitors or intestinal microbial interference, X ij represents the binding energy value between the i-th molecule and the j-th protein; In step 3), the Pearson correlation coefficient can measure the linear correlation between two random variables, and the calculation formula is: （4） Where cov(a, b) represents the covariance of the values ​​of the molecular characteristics a and b of the 5-HT reuptake inhibitor, σa and σb represent the standard deviations of the values ​​of the molecular characteristics a and b of the 5-HT reuptake inhibitor, respectively, and E represents the mathematical expectation; In step 5), the loss function is defined as: （5） Where y m Indicates the effect value of the current molecule, p m Represents the predicted value of the current molecule, and m represents the number of molecules that need to be predicted. By taking the absolute value and then accumulating it, the accuracy of the prediction of each molecule can be effectively guaranteed; In step 6), the data is first expanded through a layer of 1D-CNN neural network, GELU and normalization layers, and then the data is contracted through a layer of 1D-CNN neural network. This process is repeated, and finally the effect value is predicted through the GRU neural network. By calculating the loss function and realizing back propagation, the purpose of accurate and intelligent prediction of the molecular effect value of 5-HT reuptake inhibitors is achieved.

Images