Tobacco dependence prediction method and device based on omics and brain cognition

Abstract

According to the tobacco dependence prediction method and device based on omics and brain cognition, early screening of high-risk groups can be achieved, an objective basis is provided for individualized treatment, the real-time monitoring intervention effect is improved, and finally the smoking cessation success rate is increased. The method comprises the following steps: (1) data acquisition: performing standardized multi-modal data acquisition, including whole genome sequencing, electroencephalogram characteristic recording and computerized cognitive behavior evaluation; and (2) data analysis: designing a multi-modal deep learning algorithm based on a Transform architecture, and realizing deep interaction among different modal features through a cross attention mechanism.

Description

Technical Field

[0001] The present invention relates to the technical field of bioinformatics and artificial intelligence, and in particular to a tobacco dependence prediction method based on omics and brain cognition, and a tobacco dependence prediction device based on omics and brain cognition according to this method. Background Art

[0002] Tobacco dependence, a complex, chronic brain disease, has become one of the most serious global public health challenges. The latest statistics from the World Health Organization show that there are approximately 1.3 billion tobacco users worldwide, and tobacco-related diseases cause over 8 million deaths each year. The essence of this addictive disease is that nicotine activates nicotinic acetylcholine receptors (nAChRs) in the central nervous system, particularly the α4β2 subtype, triggering the release and concentration changes of multiple neurotransmitters such as dopamine, γ-aminobutyric acid (GABA), and glutamate. Ultimately, this activates the mesolimbic dopamine reward circuitry, leading to persistent physiological and psychological dependence. This dependence is characterized by four key features: gradually developing tolerance, pronounced withdrawal symptoms, compulsive drug-seeking behavior, and a high relapse rate.

[0003] Currently, the main diagnostic tools used in clinical practice include the Test for Nicotine Dependence (FTND), the DSM-5 diagnostic criteria, and the Minnesota Nicotine Withdrawal Scale (MNWS). While these methods can assess the severity of tobacco dependence to a certain extent, they have significant limitations: First, these scales rely entirely on subjective patient reports and are susceptible to recall bias and subjective will. Second, these tools have low sensitivity and specificity, making it difficult to accurately identify patients with early-stage or mild dependence. Third, these methods cannot distinguish between different types of dependence (such as physical and psychological dependence). Most importantly, they rely on long-term observations and cannot dynamically monitor the degree of dependence, making it difficult to provide real-time guidance for clinical intervention. Although some biomarker-based assays have emerged in recent years, such as blood nicotine metabolite (cotinine) testing and exhaled carbon monoxide (CO) testing, these indicators only reflect recent nicotine exposure levels and cannot represent the extent of long-term dependence. More importantly, there is a lack of direct evidence linking these biomarkers with changes in brain function.

[0004] The rapid development of modern brain science and technology, especially advances in functional neuroimaging and electrophysiology, has provided new perspectives for understanding the neural mechanisms of tobacco dependence. Structural imaging studies have found that long-term smokers have significantly reduced gray matter volume in the prefrontal cortex; functional imaging studies have shown that the availability of dopamine D2 receptors in the striatum of smokers is significantly reduced; and electroencephalogram (EEG) studies have observed that tobacco-dependent patients exhibit characteristically decreased theta wave power and increased beta wave power. However, existing research has three significant limitations: First, the vast majority of studies are based on data from Western populations, lacking specific studies targeting the Chinese population; second, sample sizes are generally small (usually less than 100 cases), resulting in insufficient statistical power; and third, research methods are often limited to single-modality analysis, lacking systematic integration of multi-omics data.

[0005] In recent years, the breakthrough development of omics technology has brought new opportunities for tobacco dependence research. Genomic studies have identified multiple gene loci significantly associated with tobacco dependence through genome-wide association analysis (GWAS), the most prominent of which is the CHRNA5-CHRNA3-CHRNB4 gene cluster (pigment-related genes) located in the 15q25.1 region. These genes encode key subunits of nicotine receptors. Epigenomic studies have found that the methylation level of the CG05575921 site of the AHRR gene in the peripheral blood of smokers is significantly reduced, and this change is closely related to the intensity of smoking. Transcriptomics studies have revealed through RNA sequencing technology that the expression profiles of inflammation-related genes in the peripheral blood of smokers have changed significantly, and these changes may be related to the maintenance and development of dependence. However, the bridge between these omics findings and changes in brain function has not yet been fully established, which is also one of the scientific problems to be solved by the present invention. Summary of the Invention

[0006] To overcome the shortcomings of the existing technology, the technical problem to be solved by the present invention is to provide a tobacco dependence prediction method based on omics and brain cognition, which can achieve early screening of high-risk groups, provide an objective basis for individualized treatment, improve the real-time monitoring intervention effect, and ultimately improve the success rate of smoking cessation.

[0007] The technical solution of the present invention is: this tobacco dependence prediction method based on omics and brain cognition includes the following steps:

[0008] (1) Data collection: Standardized multimodal data collection, including whole genome sequencing, EEG feature recording, and computerized cognitive behavioral assessment;

[0009] (2) Data Analysis: Design a multimodal deep learning algorithm based on the Transformer architecture,

[0010] Deep interaction between different modal features is achieved through the cross-attention mechanism.

[0011] This paper proposes an innovative multimodal fusion approach that integrates genomic data, EEG features, and cognitive behavioral data to construct an AI-based prediction model. This establishes the first predictive model combining genomic and EEG features, enabling early screening of high-risk populations. This provides an objective basis for personalized treatment, improves real-time monitoring of intervention effectiveness, and ultimately increases smoking cessation success rates.

[0012] Also provided is a tobacco dependence prediction device based on omics and brain cognition, which includes:

[0013] a data acquisition module configured to perform standardized multimodal data collection, including whole genome sequencing, EEG feature recording, and computerized cognitive behavioral assessment;

[0014] The data analysis module is configured to design a multimodal deep learning algorithm based on the Transformer architecture, and to achieve deep interaction between different modal features through the cross-attention mechanism. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 The overall computational process is demonstrated, including a portable EEG headband and mobile app for data collection and preliminary visualization; a server deployed in the hospital for data cleaning and feature extraction; and a cloud server running the core neural network algorithm, supporting multi-center data integration and model updates.

[0016] Figure 2 This is an example of a website that is used to analyze collected EEG signals. The displayed content includes the extracted EEG signals, the distribution of electrodes, extracted features, etc.

[0017] Figure 3 Schematic diagram of the overall process of the tobacco dependence prediction method based on omics and brain cognition according to the present invention. DETAILED DESCRIPTION

[0018] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and specific 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.

[0019] To provide a more complete and detailed description of this disclosure, the following illustrative descriptions of embodiments and examples of the present invention are provided; however, these descriptions are not intended to be the only ways to implement or use the embodiments of the present invention. The embodiments cover features of various embodiments, as well as the method steps and sequences for constructing and operating these embodiments. However, other embodiments may be used to achieve the same or equivalent functionality and sequence of steps.

[0020] like Figure 3 As shown in FIG, this tobacco dependence prediction method based on omics and brain cognition includes the following steps:

[0021] (1) Data collection: Standardized multimodal data collection, including whole genome sequencing, EEG feature recording, and computerized cognitive behavioral assessment;

[0022] (2) Data Analysis: Design a multimodal deep learning algorithm based on the Transformer architecture,

[0023] Deep interaction between different modal features is achieved through the cross-attention mechanism.

[0024] This paper proposes an innovative multimodal fusion approach that integrates genomic data, EEG features, and cognitive behavioral data to construct an AI-based prediction model. This establishes the first predictive model combining genomic and EEG features, enabling early screening of high-risk populations. This provides an objective basis for personalized treatment, improves real-time monitoring of intervention effectiveness, and ultimately increases smoking cessation success rates.

[0025] Preferably, in step (1), an 8-channel dry electrode EEG headband is used in conjunction with an adaptive impedance matching circuit to obtain EEG feature records, wherein the headband has a built-in low-power Bluetooth module for real-time wireless data transmission.

[0026] Preferably, in step (1), the computerized cognitive behavioral assessment adopts an event-related potential paradigm, presents visual stimulation through a mobile terminal, requires the subject to respond to specific buttons and pictures, and simultaneously records EEG signals and behavioral responses.

[0027] Preferably, in step (1), the EEG feature records exclude periods with severe artifacts and then enter preprocessing, including: 50 Hz power frequency filtering to remove mains interference; independent component analysis (ICA) to separate the sources of various neural electrical activities; wavelet transform and reconstruction to eliminate myoelectric and eye movement artifacts; using Hamming distance to remove window effects; arranging the frequency spectrum into bands and taking the average.

[0028] Preferably, in step (1), the preprocessed signal enters the feature extraction stage to calculate the power spectral density, coherence connectivity and microstate parameters of each frequency band, the frequency bands including: first wave δ: 0.5-4Hz, second wave θ: 4-8Hz, third wave α: 8-13Hz, fourth wave β: 13-30Hz, and these features are analyzed together with the genomic data.

[0029] Preferably, in step (1), the EEG features include:

[0030] The energy and amplitude of δ: reflect the state of basic neural activity;

[0031] Theta power and amplitude: related to cognitive control and addiction craving;

[0032] Alpha power and amplitude: indicators of resting-state brain function;

[0033] Beta power and amplitude: characterize cortical excitability;

[0034] θ / β: key indicators of prefrontal cortex regulatory function;

[0035] (α+θ) / β: comprehensive assessment of brain inhibition balance;

[0036] Theta frequency band coherence between DLPFC and nucleus accumbens.

[0037] Preferably, in step (2), Transformer is used as the basic framework, which includes three key components: the genomic feature encoder uses a multi-layer fully connected network to process SNP data; the EEG feature encoder uses a temporal convolutional network to extract spatiotemporal features; the multimodal fusion module realizes information interaction between different modalities through a cross-attention mechanism; the model training adopts an end-to-end approach, and the loss function combines classification cross entropy and consistency constraints to ensure the balance of contributions from each modality.

[0038] Preferably, the step (2) includes an interpretability module to quantify the contribution of each feature through SHAP value.

[0039] Preferably, the input and output of the Transformer include:

[0040] SNP and EEG feature embedding: time-frequency features are used to extract local patterns through 1D convolutional layers, and the functional connectivity matrix uses a graph attention network to encode brain region relationships;

[0041] Multimodal cross-attention module: takes genomic features and EEG features as input and calculates cross-modal attention weights;

[0042] Output prediction layer: output dependency risk score and key feature contribution.

[0043] Also provided is a tobacco dependence prediction device based on omics and brain cognition, which includes:

[0044] a data acquisition module configured to perform standardized multimodal data collection, including whole genome sequencing, EEG feature recording, and computerized cognitive behavioral assessment;

[0045] The data analysis module is configured to design a multimodal deep learning algorithm based on the Transformer architecture, and to achieve deep interaction between different modal features through the cross-attention mechanism.

[0046] The generated report is rich and intuitive, consisting of three core parts: risk assessment, feature interpretation, and intervention recommendations. Risk assessment uses a continuous score of 0-1 and is divided into three levels: low risk (0-0.3), medium risk (0.3-0.7), and high risk (0.7-1.0). The feature interpretation section highlights key abnormal indicators, such as specific genotypes, abnormal theta wave power, weakened prefrontal connectivity, etc. Intervention recommendations are personalized based on the risk level, with a step-by-step approach from health education, behavioral intervention to drug treatment. The report can be exported in PDF format for easy integration into the electronic medical record system.

[0047] In clinical application scenarios, this invention demonstrates unique advantages. For physical examination centers, it enables rapid screening of large populations; for smoking cessation clinics, it provides an objective basis for individualized treatment; and for research institutions, standardized data collection processes facilitate multicenter research. Compared with traditional scale assessments, the objective indicators provided by this invention significantly improve patient acceptance of diagnostic results, thereby enhancing treatment compliance.

[0048] The following specific examples are provided to illustrate the present invention in more detail.

[0049] Example 1: Multicenter clinical validation study

[0050] To comprehensively evaluate the validity and reliability of this invention, we conducted a large-scale validation study across three clinical research centers. The study employed a prospective cohort design and enrolled 1,200 participants, including 600 smokers who met DSM-5 diagnostic criteria for tobacco dependence and 600 healthy controls. The smokers had a mean age of 37.6 ± 8.7 years, a mean smoking history of 20.3 ± 9 years, and a daily cigarette consumption of 21.4 ± 6.8 cigarettes. The control group was closely matched for age, sex, and education. All participants underwent genome sequencing, high-density electroencephalography, and standardized assessments.

[0051] Genomic analysis was performed using the Illumina NovaSeq 6000 platform for whole-genome sequencing (average depth 30X). Quality control included: sample detection rate >98%, SNP detection rate >95%, minor allele frequency (MAF) >0.01, and Hardy-Weinberg equilibrium test p>1×10^-4. After quality control, a total of 5,738,989 high-quality SNPs were retained for subsequent analysis. GWAS analysis used a linear mixed model to correct for age, sex, and the first 10 principal components. The results showed a significant signal in the 15q25.1 region, among which SNP rs16969968 (CHRNA5) reached genome-level significance p=1.2×10^-7. This site was significantly negatively correlated with theta wave power (β=-0.21). GWAS analysis process:

[0052] (1) Data quality control

[0053] Sample filtering:

[0054] Sample detection rate>98%

[0055] Kinship test (PI_HAT<0.25)

[0056] SNP Filtering:

[0057] Minor allele frequency (MAF) ≥ 0.01

[0058] Hardy-Weinberg equilibrium p-value ≥ 1×10 -4

[0059] Typing success rate>95%

[0060] (2) Correlation Analysis Model

[0061] Using mixed linear model (MLM):

[0062] y=μ+SNP+PC1-10+Age+Sex+ε

[0063] in:

[0064] y: phenotype (probability of addiction)

[0065] Significance threshold (preliminary): p < 1 × 10 -6 (Genome-wide level)

[0066] PC1-10 (the first 10 variables of principal component analysis, corrected population division)

[0067] Functional annotation

[0068] ANNOVAR annotation variant function

[0069] Gene set enrichment analysis using FUMA

[0070] EEG analysis revealed that the tobacco dependence group exhibited characteristically decreased theta wave power (p<0.013) and increased beta wave power (p=0.023). Functional connectivity analysis revealed a significant decrease in prefrontal-striatal connectivity and increased connectivity within the default mode network. In task-based analysis, the dependence group exhibited delayed striatal activation and diminished prefrontal regulatory function during a reward task.

[0071] The performance of the multimodal model was evaluated by rigorous cross-validation. The dataset was randomly divided into a training set (n=800), a validation set (n=200), and a test set (n=200). On the independent test set, the model achieved an AUC of 0.85 (95%) for distinguishing tobacco dependence, and an AUC of 0.79 using EEG data alone. Compared with the traditional FTND scale, the classification performance of the new method was significantly improved (p<0.031). The model had a prediction accuracy of 78.5% for 3-month relapse after quitting smoking, demonstrating good clinical predictive value.

[0072] Example 2: Portable Device Performance Evaluation

[0073] To validate the reliability of the portable EEG device, a comparative methodological study was conducted across three hospitals. A total of 150 participants (75 smokers and 75 controls) were recruited and each participant underwent testing with a laboratory-grade EEG device (Neuroscan SynAmps2) followed by a portable device, with a one-hour interval to eliminate sequential effects.

[0074] Signal quality assessment showed that the EEG signals collected by the portable device were highly consistent with those collected by the laboratory device. Time-domain analysis revealed that the difference in the average amplitude of the theta wave at the Pz electrode between the two devices was only 0.35 μV (95% CI: -0.12 to 0.88); frequency-domain analysis showed that the correlation coefficient of power in the main frequency bands was r>0.93 (p<0.001). The consistency of functional connectivity measurements was 0.89, an excellent level. In terms of operational performance, the portable device demonstrated a clear advantage. The average wearing time was only 5.3 minutes, reducing signal preparation time by 80%.

[0075] Compared to traditional diagnostic methods, this method offers significant cost-effectiveness. While the direct cost of a single test is slightly higher than questionnaire screening, the multiple benefits it offers include: more accurate early identification can reduce the incidence of severe addiction by 30%; and personalized treatment recommendations can increase smoking cessation success rates by 20%.

[0076] The above description is merely a preferred embodiment of the present invention and does not constitute any form of limitation to the present invention. Any simple modifications, equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention are still within the scope of protection of the technical solution of the present invention.

Claims

1. A tobacco dependence prediction method based on omics and brain cognition, characterized by: It includes the following steps: (1) Data collection: Standardized multimodal data collection, including whole genome sequencing, EEG feature recording, and computerized cognitive behavioral assessment; (2) Data Analysis: Design a multimodal deep learning algorithm based on the Transformer architecture, and achieve deep interaction between different modal features through the cross-attention mechanism.

2. The tobacco dependence prediction method based on omics and brain cognition according to claim 1, characterized in that: In step (1), an 8-channel dry electrode EEG headband is used in conjunction with an adaptive impedance matching circuit to obtain EEG feature records, wherein the headband has a built-in low-power Bluetooth module for real-time wireless data transmission.

3. The tobacco dependence prediction method based on omics and brain cognition according to claim 2, characterized in that: In step (1), the computerized cognitive behavioral assessment adopts the event-related potential paradigm, presents visual stimulation through a mobile terminal, requires the subject to make specific keystrokes and picture responses, and simultaneously records EEG signals and behavioral responses.

4. The tobacco dependence prediction method based on omics and brain cognition according to claim 3, characterized in that: In the step (1), the EEG feature records exclude the period with serious artifacts, and then enter preprocessing, including: 50Hz power frequency filtering to remove mains interference; independent component analysis (ICA) to separate the sources of various neural electrical activities; wavelet transformation and reconstruction to eliminate myoelectric and eye movement artifacts; Use Hamming distance to remove the window effect; organize the frequency spectrum into bands and take the average.

5. The tobacco dependence prediction method based on omics and brain cognition according to claim 4, characterized in that: In step (1), the preprocessed signal enters the feature extraction stage to calculate the power spectral density, coherence connectivity and microstate parameters of each frequency band. The frequency bands include: first wave δ: 0.5-4Hz, second wave θ: 4-8Hz, third wave α: 8-13Hz, and fourth wave β: 13-30Hz. These features are analyzed together with the genomic data.

6. The tobacco dependence prediction method based on omics and brain cognition according to claim 1, characterized in that: In step (1), the EEG features include: The energy and amplitude of δ: reflect the state of basic neural activity; Theta power and amplitude: related to cognitive control and addiction craving; Alpha power and amplitude: indicators of resting-state brain function; Beta power and amplitude: characterize cortical excitability; θ / β: key indicators of prefrontal cortex regulatory function; (α+θ) / β: comprehensive assessment of brain inhibition balance; Theta frequency band coherence between DLPFC and nucleus accumbens.

7. The tobacco dependence prediction method based on omics and brain cognition according to claim 6, characterized in that: In step (2), Transformer is used as the basic framework, which includes three key components: the genomic feature encoder uses a multi-layer fully connected network to process SNP data; the EEG feature encoder uses a temporal convolutional network to extract spatiotemporal features; the multimodal fusion module realizes information interaction between different modalities through a cross-attention mechanism; the model training adopts an end-to-end approach, and the loss function combines classification cross entropy and consistency constraints to ensure the balance of contributions from each modality.

8. The tobacco dependence prediction method based on omics and brain cognition according to claim 7, characterized in that: The step (2) includes an interpretability module that quantifies the contribution of each feature through the SHAP value.

9. The tobacco dependence prediction method based on omics and brain cognition according to claim 8, characterized in that: The input and output of Transformer include: SNP and EEG feature embedding: time-frequency features are used to extract local patterns through 1D convolutional layers, and the functional connectivity matrix uses a graph attention network to encode brain region relationships; Multimodal cross-attention module: takes genomic features and EEG features as input and calculates cross-modal attention weights; Output prediction layer: output dependency risk score and key feature contribution.

10. A tobacco dependence prediction device based on omics and brain cognition, characterized by: It includes: a data acquisition module configured to perform standardized multimodal data collection, including whole genome sequencing, EEG feature recording, and computerized cognitive behavioral assessment; The data analysis module is configured to design a multimodal deep learning algorithm based on the Transformer architecture, and to achieve deep interaction between different modal features through the cross-attention mechanism.

Images