Predictive model for discriminating alcohol use disorder from alcoholic hepatitis and applications thereof

Abstract

This invention discloses a predictive model and its application for differentiating between alcohol use disorder (AUD) and alcoholic hepatitis (AH). By integrating the patient's basic clinical information, gut bacterial community characteristics, and gut fungal community structure, a multi-dimensional differentiation model is constructed. This model, through the joint analysis of gut microbiota (bacteria + fungi) and patient baseline data, provides an innovative and practical combination of molecular markers for the accurate differential diagnosis of AUD and AH.

Description

Technical Field

[0001] This invention relates to the field of disease prediction model technology, specifically to a prediction model and its application for differentiating between alcohol use disorder and alcoholic hepatitis. Background Technology

[0002] Alcoholic liver disease (ALD) encompasses a range of liver pathological changes, including simple steatosis, alcoholic hepatitis (AH), alcoholic cirrhosis (ALC), and hepatocellular carcinoma (HCC), posing a significant threat to human health. Epidemiological studies show that over 90% of alcoholics have hepatic steatosis, of whom approximately 10%-35% will develop AH; among AH patients, approximately 10%-20% will progress to ALC; and among ALC patients, approximately 3%-10% will eventually develop HCC. Clinical diagnosis of ALD relies on a history of alcohol consumption, signs of liver disease, biochemical abnormalities, and imaging characteristics, and other causes of liver disease must be ruled out. Early diagnosis of ALD is difficult, as it is often asymptomatic or presents with only nonspecific symptoms. The core treatment for AH is compulsory abstinence from alcohol, supplemented by nutritional support and management of complications; short-term glucocorticoids may be considered for severe AH. In addition, targeting the gut microbiota (such as probiotics and fecal microbiota transplantation) has become an emerging strategy in recent years, with its core mechanism based on the vicious cycle of the gut-hepatic axis.

[0003] Alcohol use disorder (AUD) is a state of physical and psychological dependence on alcohol caused by long-term alcohol consumption, requiring abstinence to cut off the continuous damage of alcohol to the nervous system. Abstinence is also a decisive factor in improving the prognosis of AH (damage is reversible in mild to moderate cases), and gut microbiota interventions (probiotics, fecal microbiota transplantation, phage therapy) show potential. However, current methods have three major limitations: 1. Diagnostic lag: lack of sensitive biomarkers for early identification of AH and AUD populations; 2. Mechanistic blind spots: the role of gut bacteria-fungus interactions in the evolution of AUD to AH needs to be elucidated; 3. Intervention bottlenecks: existing therapies cannot precisely block disease progression. Therefore, in-depth analysis of the key mechanisms of gut bacteria-fungus interactions and the development of novel non-invasive predictive models based on this are crucial for early screening of high-risk AUD populations, accurate identification of AH, and the development of targeted intervention strategies. Summary of the Invention

[0004] This invention provides a multi-dimensional identification model that integrates gut bacterial community characteristics and gut fungal community structure. This model uses gut microbiota (bacteria + fungi) to obtain a set of specific and sensitive biomarkers, providing an innovative and practical molecular biomarker combination scheme for the accurate differential diagnosis of alcohol use disorder (AUD) and alcoholic hepatitis (AH).

[0005] In view of this, the solution of the present invention is as follows:

[0006] The first aspect of the invention is to propose biomarkers for differentiating alcohol use disorder from alcoholic hepatitis, said biomarkers including oral streptococci ( s__Streptococcus_oralis ), Senegalese anaerobic bacteria ( s__ Anaeromassilibacillus_senegalensis ), Clostridium botulinum ( s__Enterocloster_bolteae ), Trichosporium curvatureum ( s__Cutaneotrichosporon_curvatum ), unidentified species of the genus *Volleyi* ( s__Wallemia_ sp ) and Penicillium cinnamon ( s__Penicillium_cinnamopurpureum ).

[0007] Furthermore, the biomarkers also include aspartate aminotransferase (AST).

[0008] A second aspect of the invention is the application of a detection reagent for the biomarker described in the first aspect above in the preparation of a predictive product for differentiating between alcohol use disorder and alcoholic hepatitis.

[0009] Furthermore, the detection reagent is an abundance detection reagent, including primers for specific amplification of the V3-V4 variable region of the 16S rRNA gene for detecting bacteria, and primers for specific amplification of the ITS region for detecting fungi.

[0010] Furthermore, the nucleotide sequences of the primers used for detecting bacteria are as follows: upstream primer: 5'-AGRGTTYGATYMTGGCTCAG-3', downstream primer: 5'-RGYTACCTTGTTACGACTT-3';

[0011] And / or, the nucleotide sequences of the primers used for detecting fungi are: upstream primer: 5'-CTTGGTCATTTAGAGGAAGTAA-3'; downstream primer: 5'-TCCTCCGCTTATTGATATGC-3'.

[0012] Furthermore, the predicted product is a reagent kit, detection system, or instrument.

[0013] A third aspect of the invention is to provide a kit for differentiating alcohol use disorder from alcoholic hepatitis, the kit comprising a microbial abundance detection reagent, a prediction formula, and a prediction standard; the microbial abundance detection reagent is used to detect the abundance of the biomarker described in the first aspect; the prediction formula is: P(AH) = 1 / (1+exp(-(-2.66+180*)) s__Streptococcus_oralis -67.3* s__Enterocloster_bolteae -1.22e+03* s__ Anaeromassilibacillus_senegalensis +64.9* s__Wallemia_sp -1.8* s__ Cutaneotrichosporon_curvatum +902* s__Penicillium_cinnamopurpureum +0.085*AST), where AST represents the serum aspartate aminotransferase level in the patient; the prediction criterion is that when P(AH) is greater than 0.527, the patient is identified as having alcoholic hepatitis (AH); otherwise, the patient is identified as having alcohol use disorder (AUD).

[0014] Furthermore, the kit also includes reagents for extracting the biomarkers from the sample.

[0015] A fourth aspect of the invention is to provide an apparatus for differentiating between alcohol use disorder and alcoholic hepatitis, comprising a detection unit and a data analysis unit, wherein:

[0016] The detection unit is used to acquire samples, detect and determine the abundance of microorganisms in the samples; the microorganisms are the biomarkers described in the first aspect.

[0017] The data analysis unit is used to analyze the detection results of the detection unit in combination with the serum aspartate aminotransferase content to make identification results.

[0018] Furthermore, the data analysis unit distinguishes between alcohol use disorder and alcoholic hepatitis based on a prediction formula, which is: P(AH)=1 / (1+exp(-(-2.66+180*) s__Streptococcus_oralis -67.3* s__ Enterocloster_bolteae -1.22e+03* s__Anaeromassilibacillus_senegalensis +64.9* s__ Wallemia_sp -1.8* s__Cutaneotrichosporon_curvatum +902* s__Penicillium_ cinnamopurpureum +0.085*AST), where AST represents the serum aspartate aminotransferase (AST) level in the patient; the prediction criterion is that when P(AH) is greater than 0.527, the patient is diagnosed with alcoholic hepatitis.

[0019] Furthermore, the detection unit performs detection and obtains detection results according to the following operations:

[0020] Test fecal samples to obtain DNA data;

[0021] The data analysis unit is used to analyze and process the detection results of the detection unit, including: sequencing the DNA data of the fecal sample to obtain intestinal microbial genome data;

[0022] The relative abundance information of each biomarker was obtained by analyzing the gut microbiome genome data.

[0023] Based on the relative abundance information, gut microbiota characteristic data are determined.

[0024] Compared with the prior art, the present invention has the following beneficial effects:

[0025] The combined biomarkers provided by this invention include bacteria with excellent specificity and fungi with high sensitivity. By integrating the two types of biomarkers, it is of great significance to synergistically differentiate between alcoholic hepatitis and alcoholic cirrhosis.

[0026] This invention demonstrates a significant performance improvement through a multimodal joint diagnostic model where bacteria and fungi cannot combine with AST (assay response). On the training set, the model's AUC value reaches 0.995, exhibiting superior discriminative ability compared to a simple microbial model. Simultaneously, the model achieves an excellent balance between a sensitivity of 97.5% and a specificity of 100%, realizing a high disease detection rate and a low misdiagnosis rate. These advantages make the model particularly suitable for clinical applications, effectively identifying potential AH patients while minimizing overdiagnosis of AUD (assay dysplasia). Attached Figure Description

[0027] Figure 1 This is a bar graph showing the Shannon and Invsimpson indices and the Wilcoxon rank-sum test results for fecal bacteria from the four groups of people described in the example.

[0028] Figure 2 The results of the β-diversity analysis of fecal bacteria from the four groups of people described in the examples are shown.

[0029] Figure 3 These are the main components of fecal bacteria in the four groups of people described in the examples at the phylum and genus levels.

[0030] Figure 4 These are the main components of fecal bacteria at the species level in the four groups of people described in the examples.

[0031] Figure 5 This is a bar graph showing the Shannon and Invsimpson indices and the Wilcoxon rank-sum test results for fecal fungi in the four groups of people described in the examples.

[0032] Figure 6 The results of the β-diversity analysis of fecal fungi in the four groups of people described in the examples are shown.

[0033] Figure 7 These are the main components of the fecal fungi in the four groups of people described in the examples, at the phylum and genus levels.

[0034] Figure 8 These are the main components of the fecal fungi in the four groups of people described in the examples at the species level.

[0035] Figure 9 In this example, the bacteria ranked 10th in terms of the greatest influence on AH and ALC classification were selected by random forest feature importance screening.

[0036] Figure 10 In this example, the fungi ranked 10th in terms of the greatest influence on AH and ALC classifications were selected by random forest feature importance screening.

[0037] Figure 11 The AUC curves of the logistic regression model established by the combined bacteria in the training set (11A) and the AUC curves of the logistic regression model established by the combined bacteria in the training set and the validation set (11B) are shown in the examples.

[0038] Figure 12 The AUC curves of the logistic regression model established by the combined fungi in the training set (12A) and the AUC curves of the logistic regression model established by the combined fungi in the training set and the validation set (12B) are shown in the examples.

[0039] Figure 13 The AUC curves (13A) of the logistic regression model established by bacteria and fungi in the training set in the examples are shown; and the AUC curves (13B) of the logistic regression model established by bacteria and fungi in the training set and the validation set are shown.

[0040] Figure 14 The AUC curves (14A) of the logistic regression model established by combining bacteria and fungi with patient age in the training set are shown in the examples; and the AUC curves (14B) of the logistic regression model established by combining bacteria and fungi with patient age in the training set and validation set are shown in the examples. Detailed Implementation

[0041] The technical solution of the present invention will now be clearly and completely described in conjunction with preferred embodiments. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0042] This invention relates to a bibliographical table with an English-Chinese equivalent, as follows:

[0043]

[0044] This invention relates to a glossary of fungal classifications in English and Chinese, as shown below:

[0045]

[0046] Note: "Unclassified fungi" refers to fungi that can only be identified at a certain level and cannot be further classified into a specific known category, such as... unclassified_k__FungiThis mainly refers to the fact that the sequencing sequence can only confirm that the microorganism belongs to the kingdom Fungi, but cannot be further classified. Additionally, "sp" is an abbreviation for species, indicating that the microorganism belongs to that genus, but a specific species has not been identified. Wallemia_sp This indicates that the microorganism belongs to the genus *Vorleyella*, but no specific species was identified. The relative abundance of major unclassified fungi = (number of unclassified sequences / total number of fungal sequences) × 100%.

[0047] The sequencing method involved in this invention is as follows:

[0048] 1.1 Materials

[0049] 1) Study subjects: This study included patients with alcoholic liver disease (ALD) and healthy volunteers. The inclusion criteria were: (1) age greater than 18 years; (2) healthy controls (HC): no history of drinking or serious illness; (3) patients with alcohol use disorder (AUD), alcoholic hepatitis (AH), or alcoholic cirrhosis (ALC) with a history of drinking for more than 5 years, with a daily ethanol intake of ≥40 grams for men and ≥20 grams for women. The exclusion criteria were: (1) co-infection with hepatitis A, B, C, D, or E virus or human immunodeficiency virus; (2) co-infection with non-alcoholic fatty liver disease, drug-induced liver injury, autoimmune liver disease, congenital liver disease, etc.; (3) presence of primary liver cancer or liver metastasis; (4) presence of serious organic diseases affecting other organs; (5) pregnancy or lactation; (6) history of antibiotic or probiotic use in the past 3 months. This study has been approved by the Ethics Review Committee of Union Hospital, Tongji Medical College, Huazhong University of Science and Technology and registered in the Chinese Clinical Trial Registry ((NCT05448144)). All study participants have signed informed consent forms.

[0050] 2) Sample collection: Stool samples were collected from the subjects upon admission, placed in sterile plastic cups, and stored at -80°C until extraction.

[0051] 3) Main reagents and instruments: FastPure Stool DNA Isolation Kit (MJYH, Shanghai, China); FastPfu Polymerase (Beijing TransGen Biotech Co., Ltd.); T100 Thermal Cycler (BIO-RAD, USA); JY600C bistable electrophoresis apparatus (Beijing Junyi Oriental Electrophoresis Equipment Co., Ltd.). High-throughput sequencing was performed by Shanghai Meiji Biopharmaceutical Technology Co., Ltd.

[0052] 4) Clinical indicator measurement: Basic characteristics such as gender, age, and weight were collected from healthy volunteers and patients. Blood routine, biochemical and imaging data were obtained by consulting clinical data.

[0053] 1.2 Methods

[0054] 1) Sample DNA extraction: Total genomic DNA of microbial communities was extracted from fecal samples from the disease group and the healthy group according to the instructions of the FastPure Stool DNAIsolation Kit (MJYH, shanghai, China). The integrity of the extracted genomic DNA was detected by 1% agarose gel electrophoresis, and the DNA concentration and purity were determined by NanoDrop2000 (ThermoScientific, USA).

[0055] 2) PCR amplification and sequencing library construction: Using the extracted DNA as a template, the full-length 16S rRNA gene of bacteria was amplified using primers 27F (5'-AGRGTTYGATYMTGGCTCAG-3') and 1492R (5'-RGYTACCTTGTTACGACTT-3') with barcodes, and the full-length ITS PCR amplification of fungi was performed using primers ITS1F (5'-CTTGGTCATTTAGAGGAAGTAA-3') and ITS4R (5'-TCCTCCGCTTATTGATATGC-3'). The PCR reaction system was as follows: 4 μL of 5×FastPfu buffer, 2 μL of 2.5 mM dNTPs, 0.8 μL of upstream primer (5 uM), 0.8 μL of downstream primer (5 uM), 0.4 μL of FastPfu polymerase, 0.2 μL of BSA, 10 ng of template DNA, and the volume was brought to 20 μL. Each sample was replicated in triplicate. The amplification program was as follows: pre-denaturation at 95℃ for 3 min, 27 cycles (denaturation at 95℃ for 30 s, annealing at 60℃ for 30 s, extension at 72℃ for 30 s), followed by a stable extension at 72℃ for 10 min, and finally storage at 4℃ (PCR instrument: T100 Thermal Cycler PCR, USA). After detection by 2% agarose gel electrophoresis, the product was purified by magnetic beads and quantified using a Qubit 4.0 (Thermo Fisher Scientific, USA). The products were then mixed in appropriate proportions according to the sequencing volume requirements for each sample.

[0056] Library construction was performed using the SMRTbell prep kit 3.0: (1) DNA damage repair; (2) end repair; (3) adapter ligation. Sequencing was performed using the PacBio Sequel IIe System (Shanghai Meiji Biotechnology Co., Ltd.). HiFi reads were generated from the sequenced subreads using the CCS mode of SMRT-Link v11.0 for subsequent data analysis.

[0057] 3) Sequencing data analysis

[0058] Data from each sample was differentiated based on barcode sequences, and length filtering and orientation correction were performed, retaining sequences of 1000-1800 bp (bacteria) / 300-900 bp (fungi). Using the software platform Uparse (version 7.0.1090 http: / / drive5.com / uparse / ), OTU clustering was performed on sequences based on 97% similarity, and chimeras were removed. Sequences annotated to chloroplasts and mitochondria were removed from all samples (removal was recommended if chloroplast and mitochondrial contamination was present). To minimize the impact of sequencing depth on subsequent Alpha and Beta diversity data analysis, the sequence count of all samples was flattened to the minimum possible sequence count. After flattening, the average sequence coverage (Good's coverage) for each sample still reached 99.09%. Bacterial OTU taxonomic annotation was performed by comparing the Silva16S rRNA gene database (v138) with the RDP classifier (http: / / rdp.cme.msu.edu / , version 2.11), and fungal OTU taxonomic annotation was performed by comparing the fungal database with the Unie (Release 8.0 http: / / unite.ut.ee / index.php) database. The confidence threshold was 70%, and the community composition of each sample was statistically analyzed at different species classification levels. Species abundance information of bacteria and fungi was extracted from the sequencing data, and the following indicators were calculated: (1) Relative abundance of a specific bacterial species (percentage of total bacterial community abundance) and (2) Relative abundance of a specific fungal species (percentage of total fungal community abundance).

[0059] 4) Statistical Analysis: Microbial diversity analysis was conducted on the MajorBio Cloud Platform (https: / / cloud.majorbio.com), specifically as follows: Mothur software (http: / / www.mothur.org / wiki / Calculators) was used to calculate the Alpha diversity knowledge Shannon index, etc.; PCoA analysis (principal coordinate analysis) based on the Bray-Curtis distance algorithm was used to examine the similarity of microbial community structure among samples; based on the taxonomic analysis results, R language was used to obtain the species composition of different groups (or samples) at various taxonomic levels (such as domain, kingdom, phylum, class, order, family, genus, species, OTU, etc.).

[0060] Stratified random sampling of research subjects was implemented using the `createDataPartition` function of the `caret` package in R (version 4.4.2), dividing the dataset into training and validation sets in a 7:3 ratio. Based on the relative abundance of bacteria or fungi in the training set, random forest analysis was performed using R (version 4.4.2) to screen for bacteria or fungi that play an important role in differential diagnosis. ROC curve analysis was used to continuously optimize the multi-marker combined diagnostic model for bacteria, fungi, and clinical indicators with excellent diagnostic efficacy. The results were then validated on the training set to ensure their reliability.

[0061] Example 1: Clinical and gut microbiota sequencing results of healthy controls and ALD patients

[0062] 1. Experimental Objective: To collect basic information, clinical indicators, and stool samples from healthy controls (HC), AUD, AH, and patients with alcoholic cirrhosis (ALC), and to analyze the differences in fecal microbiota between ALD patients and healthy controls through sequencing, in order to clarify whether the intestinal bacteria and fungi in ALD patients have changed.

[0063] 2. Experimental Methods: Basic information, blood routine, biochemical and other clinical indicators of hospitalized patients (including HC, AUD, AH, ALC) were prospectively collected. The inclusion criteria were: (1) age greater than 18 years; (2) healthy controls: no history of drinking or serious illness; (3) AUD, AH and ALC patients had a history of drinking for more than 5 years, with males having a daily ethanol intake of ≥40 grams and females ≥20 grams. The exclusion criteria were: (1) co-infection with hepatitis A, B, C, D, E virus or human immunodeficiency virus; (2) co-infection with non-alcoholic fatty liver disease, drug-induced liver injury, autoimmune liver disease, congenital liver disease, etc.; (3) presence of primary liver cancer or liver metastasis; (4) presence of serious organic diseases affecting other organs; (5) pregnancy or lactation; (6) history of antibiotic or probiotic use in the past 3 months. Stool samples were collected from the subjects upon admission, placed in sterile plastic cups and stored at -80℃ until extraction.

[0064] 3. Experimental Results

[0065] 3.1 Drinking alcohol can cause liver damage.

[0066] Table 1 shows the basic information and clinical data of the patients. ALT, AST, ALP, and GGT were significantly higher in AH patients than in HC and AUD patients (Table 1 and Table 2), which is consistent with the clinical manifestations of AH patients.

[0067] Table 1: Basic Patient Information and Clinical Data

[0068] Variable HC AUD AH ALC p.value age 48.5±18 52.3±10 48.3±13.9 58.2±11.3 0.0088 BMI 22.1±2.4 23.5±3.7 23.9±3.6 23.9±3.2 0.1230 FIB4 1.7±1.8 1.6±1.6 3.3±5.7 6.8±5.6 ＜0.0001 WBC 5.6±1.8 5.8±1.6 5.8±2.1 5±2.4 0.2777 RBC 4.2±0.5 4.5±0.5 4.3±0.8 3.4±1 ＜0.0001 Hb 125.1±14.5 139.5±16.3 134.6±23 109±28.5 ＜0.0001 PLT 206.3±60 205.4±49.7 187.3±59.7 108.2±50.2 ＜0.0001 TB 12.9±10.5 16±9.3 23.9±33 49.2±68.1 ＜0.0001 ALT 21.8±18.9 24.5±15.9 94.9±116.5 54.1±86 ＜0.0001 AST 27.6±28.7 27.3±20.9 101.9±153 69±70.4 ＜0.0001 ALP 71.5±26.5 70±20.9 98.9±55.9 128.3±77.4 ＜0.0001 GGT 40.8±92.7 42.2±66.7 225.7±264.2 192.4±245.9 ＜0.0001 TP 69.7±6.4 68±5.7 68.4±8.2 63.5±9.2 0.0250 ALB 43.1±2.9 42.9±3.8 41.7±6 33.9±6.5 ＜0.0001 D2 0.4±0.6 0.4±0.7 0.7±1.2 2.5±3.2 ＜0.0001 PT 13.6±1.1 13.1±1.1 13.2±1.3 15.5±2.6 ＜0.0001 INR 1.1±0.2 1±0.2 1±0.2 1.3±0.3 ＜0.0001 PTA 92.8±19.2 104.8±19.1 102.3±25 77.5±26.5 ＜0.0001 APTT 37.2±4.8 37.4±5.3 36.7±8.8 38.4±7.7 0.8286 FIB 3±0.8 3.1±0.9 3±1 2.5±0.9 0.0035 TT 18.5±1.7 18.4±1.8 18.2±2.5 19.1±2.2 0.0662

[0069] Table 2: Post-hoc comparison results of AUD and AH

[0070] Variable p.Adj (AUD VS AH) Test method age 0.2388 Dunn test FIB4 0.0082 Dunn test RBC 0.1656 Dunn test Hb 0.3864 Dunn test PLT 0.2955 Tukey HSD TB 0.3812 Dunn test ALT ＜0.0001 Dunn test AST ＜0.0001 Dunn test ALP 0.0312 Dunn test GGT ＜0.0001 Dunn test TP 0.8430 Dunn test ALB 1.0000 Dunn test D2 0.3067 Dunn test PT 0.7380 Dunn test INR 0.5833 Dunn test PTA 0.7440 Dunn test TT 0.4358 Tukey HSD

[0071] 3.2 Drinking alcohol leads to intestinal bacterial imbalance.

[0072] Figure 1-4 This study investigated the effects of alcohol consumption on human gut microbiota using 16S rRNA gene sequencing. Results showed that alcohol intake significantly altered the structure and composition of the gut microbiota. Regarding α-diversity, compared to healthy controls, patients with ALD had significantly higher Shannon index (α-diversity). Figure 1 A) and the inverse Simpson exponent ( Figure 1 B) Both showed a decreasing trend, indicating that the richness and evenness of the gut microbiota significantly decreased with increasing disease severity. β-diversity analysis showed significant differences in gut microbiota composition between ALD patients of different severities and healthy controls. Figure 2 At the phylum level, the dominant bacterial groups included Bacillota, Pseudomonadota, Bacteroidota, Actinomycetota, and Verrucomicrobiota. The relative abundance of Bacillota and Bacteroidota decreased with disease progression, while Pseudomonadota showed an increasing trend. Figure 3 A). At the genus level, potentially pathogenic bacteria such as Escherichia and Klebsiella increase with the progression of ALD, while beneficial bacteria such as Faecalibacterium decrease. Figure 3 B). Species-level analysis further confirmed this trend, with pathogenic bacteria such as *Escherichia coli* and *Klebsiella pneumoniae* increasing with the progression of ALD, while beneficial bacteria such as *Faecalibacterium prausnitzii* decreased. Figure 4 These findings suggest that alcohol consumption leads to gut microbiota dysbiosis, characterized by reduced microbial diversity, an increase in potentially pathogenic bacteria, and a decrease in beneficial bacteria, playing a significant role in the development and progression of ALD.

[0073] 3.3 Alcohol consumption leads to intestinal fungal imbalance.

[0074] Figure 5-8 This study investigated the effects of alcohol consumption on human gut fungi using ITS sequencing technology. Results showed that, although there were no significant differences in gut fungal α-diversity (including Shannon and Simpson indices) between ALD patients of different severities and HC patients,… Figure 5 A, 5B), but β-diversity analysis showed a clear separation of fungal community structure among the groups ( Figure 6 This suggests that alcohol consumption significantly alters the composition of gut fungi. At the phylum level, the gut fungal community is mainly composed of Ascomycota, Basidiomycota, unclassified_k__Fungi, and Fungi_phy_Incertae_sedis. Figure 7 A). Notably, as alcoholic liver disease (ALD) progresses, the fungal community exhibits significant dynamic changes: the relative abundance of the genus *Candida* progressively increases, while the abundance of the genus *Cutaneotrichosporon* gradually decreases. Figure 7 B). At the bacterial species level, this trend was even more pronounced, with the opportunistic pathogen Candida albicans significantly enriched in patients with AH and ALC, while the potentially beneficial bacterium Cutaneotrichosporon curvatum was significantly reduced. Figure 8 These findings indicate that alcohol consumption specifically alters the composition of the gut fungal community, manifesting as an increase in opportunistic pathogens and a decrease in potentially beneficial fungi. This change in fungal community structure may be closely related to the development of alcoholic liver disease, providing a new perspective for a deeper understanding of the impact of alcohol on the gut microbiota.

[0075] 4. Experimental conclusions:

[0076] This study found significant liver damage in patients with acute hepatocellular carcinoma (AH), primarily manifested by significantly elevated levels of ALT, AST, ALP, and GGT, indicating acute hepatocellular injury. Regarding gut microbiota, the study observed a close correlation between alcohol-induced bacterial-fungal dysbiosis and ALD progression. The AH stage already exhibited marked gut microbiota imbalance, characterized by decreased bacterial diversity, increased pathogenic bacteria, and reduced beneficial bacteria, accompanied by alterations in fungal community structure, including an increase in opportunistic pathogens and a decrease in beneficial fungi. These findings provide important evidence for a deeper understanding of the pathogenesis of ALD and the development of targeted intervention strategies.

[0077] Example 2: Establishment of the AUD vs. AH Discrimination Model

[0078] 1. Data Processing

[0079] This study included 159 participants. Stratified random sampling was implemented using the `createDataPartition` function of the `caret` package in R. The 159 participants (including 26 HC, 37 AUD, 56 AH, and 40 ALC) were divided into a training set (n=113) and a validation set (n=46) in a 7:3 ratio. The training set contained 19 HC, 26 AUD, 40 AH, and 28 ALC, while the validation set contained 7 HC, 11 AUD, 16 AH, and 12 ALC. The proportions of each group in the training and validation sets were balanced (HC 73.1% / 26.9%, AUD 70.3% / 29.7%, AH 71.4% / 28.6%, ALC 70.0% / 30.0%). All subjects underwent 16S rRNA gene sequencing and ITS sequencing. The data underwent rigorous quality control to ensure that the validation set had statistical power while maintaining a sufficient training sample size, thus providing a reliable data foundation for subsequent microbial community analysis and machine learning modeling.

[0080] 2. Differential Diagnostic Model for AUD and AH Enteric Bacteria

[0081] (1) In the training set, we evaluated the importance of gut microbiota characteristics using the random forest algorithm. Based on MeanDecrease Gini index analysis, we screened out the 10 key bacteria with the most discriminative power for AUD and AH classification: s__ Streptococcus_oralis , s__Anaeromassilibacillus_senegalensis , s__Enterocloster_ bolteae , s__Streptococcus_australis , s__Enterococcus faecium , s__Gemmiger formicilis , s__Streptococcus gordonii , s__Enterobacter hormaechei , s__Alistipes putredinis , s__Streptococcus salivarius ( Figure 9 This provides a new perspective for understanding the bacterial characteristics of AUD and AH, and also provides a theoretical basis for developing diagnostic biomarkers and intervention targets based on specific bacteria.

[0082] (2) Based on the results of random forest feature importance analysis, we further evaluated the diagnostic efficacy of the top five key bacteria. Receiver operating characteristic (ROC) curve analysis showed that these bacteria exhibited excellent differential diagnostic value for acute hepatitis (AH): s__Streptococcus orali : AUC = 0.768 (95% CI: 0.647–0.888), s__ Enterocloster bolteae: AUC = 0.73 (95% CI: 0.607–0.852), s__ Anaeromassilibacillus senegalensis : AUC = 0.718 (95% CI: 0.59–0.846), s__ Enterococcus faecium : AUC = 0.709 (95% CI: 0.577–0.841), s__Streptococcus australis AUC = 0.688 (95% CI: 0.553–0.822). It is noteworthy that in single bacteria... s__ Streptococcus orali It exhibited the best diagnostic performance (AUC=0.768), with a sensitivity of 0.775 and a specificity of 0.731. Figure 10 A, Table 3). These results suggest that specific gut bacteria not only have important pathophysiological significance, but may also serve as non-invasive biomarkers for the auxiliary diagnosis and staging of AH.

[0083] Table 3: Characteristics of the differential diagnostic model for AUD and AH intestinal bacteria

[0084] Model AUC Best_Threshold Abundance_Threshold Sensitivity Specificity PPV NPV Accuracy Precision Youden_Index 0.768 0.505 0.0407% 0.775 0.731 0.816 0.679 0.758 0.816 0.506 0.73 0.664 0.0629% 0.7 0.692 0.778 0.6 0.697 0.778 0.392 0.718 0.693 0.0111% 0.55 0.808 0.815 0.538 0.652 0.815 0.358 0.709 0.587 0.0037% 0.9 0.462 0.72 0.75 0.727 0.72 0.362 0.688 0.525 0.1666% 0.7 0.654 0.757 0.586 0.682 0.757 0.354 Combined Model 0.803 0.64 0.75 0.808 0.857 0.677 0.773 0.857 0.558 ValidationModel 0.733 0.678 0.562 0.909 0.9 0.588 0.704 0.9 0.472

[0085] (3) Based on the ROC curve analysis results, we optimized the multi-bacterial biomarker combined diagnostic model for bacterial species with excellent diagnostic efficacy. The study found that... s__Streptococcus orali , s__Enterocloster bolteae and s__Anaeromassilibacillus senegalensi The combined diagnostic model consisting of three bacteria demonstrated optimal disease prediction capability. Figure 3 B). This model integrates... s__Streptococcus orali (Optimal threshold 0.0407%) s__Enterocloster bolteae (Optimal threshold 0.0629%) and s__Anaeromassilibacillus senegalensi Abundance features (optimal threshold 0.0111%) were obtained through logistic regression, using the formula P(AH) = 1 / (1 + exp(-(0.731 + 357.493*)). s__Streptococcus oralis -199.112* s__ Enterocloster bolteae -1278.942* s__Anaeromassilibacillus senegalensis The probability of AH was calculated. That is, when P(AH) > 0.64, the patient can be identified as an AH patient. The model showed good diagnostic efficacy (AUC = 0.803, sensitivity 75.0%, specificity 80.8%, positive predictive value 85.7%, negative predictive value 67.7% (Table 3).

[0086] We validated the established multi-bacterial biomarker combined diagnostic model on an independent validation set (11 AUD patients and 16 AH patients). The validation results showed that the combined diagnostic model based on three characteristic bacterial biomarkers exhibited considerable discriminative performance on the independent validation set. ROC curve analysis indicated that the model achieved an AUC of 0.733 on the validation set. Figure 10 B). Confusion matrix analysis showed that the model's predicted classification of 27 samples differed somewhat from the actual clinical diagnosis (9 out of 16 cases in the AH group were correctly classified, and 10 out of 11 cases in the AUD group were correctly classified), with an overall accuracy of 70.37% (95% confidence interval: 0.4982–0.8625, P=0.1639). The model exhibited asymmetric recognition ability: high specificity for the AH group (90.91%), but relatively low sensitivity (56.25%); the positive predictive value reached 90.00%, while the negative predictive value was 58.82%. Based on an AH prevalence of 59.26%, the model's equilibrium accuracy was 73.58%, and the Kappa coefficient was 0.4346 (P=0.0771), indicating that the model's predictions had moderate consistency with the actual diagnosis, and there is still room for improvement in recognition performance. Considering the complex interactions between bacteria and fungi in the gut microbiota, we further explored the changing characteristics of the fungal community and its diagnostic potential.

[0087] 3. Differential Diagnostic Models for AUD and AH Enteric Fungi

[0088] (1) In the training set, we evaluated the importance of gut microbiota features using the random forest algorithm. Based on MeanDecrease Gini index analysis, we selected 10 key fungi with the strongest discriminative power for AUD and AH classification: s__ Wallemia sp , s__Cutaneotrichosporon curvatum , s__Penicillium cinnamopurpureum , s__Aspergillus penicillioides , s__unclassified_f__Dipodascaceae , s__ Moesziomyces sp , s__unclassified_g__Penicillium , s__unclassified_k__Fungi , s__ Cladosporium halotolerans , s__Candida albicans ( Figure 11 This provides a new perspective for understanding the fungal characteristics of AUD and AH, and also provides a theoretical basis for developing diagnostic biomarkers and intervention targets based on specific fungi.

[0089] (2) Based on the results of random forest feature importance analysis, we further evaluated the diagnostic efficacy of the top five key fungi. ROC curve analysis showed that these fungi exhibited excellent differential diagnostic value for AH: s__ Wallemia sp(AUC=0.8, 95%CI: 0.719-0.882), s__Cutaneotrichosporon curvatum (AUC=0.738, 95%CI: 0.617-0.858), s__Penicillium cinnamopurpureum (AUC=0.775, 95%CI: 0.697-0.853), s__Aspergillus penicillioides (AUC=0.688, 95%CI: 0.55-0.827), s__ unclassified_f__Dipodascaceae (AUC=0.738, 95%CI: 0.644–0.832). It is noteworthy that in single fungi… s__Wallemia sp It exhibited the best diagnostic performance (AUC=0.8), with a sensitivity of 62.5% and a specificity of 96.2%. Figure 12 A (Table 4); the AUC values ​​of the remaining fungal biomarkers ranged from 0.688 to 0.775, suggesting a significant association between fungal community changes and AH. These findings not only expand our understanding of ALD-related gut microbiota dysbiosis but also provide a new direction for developing a diagnostic system for AH based on fungal-bacterial combined biomarkers. Further research can focus on exploring... s__Wallemia_ sp The synergistic diagnostic value of characteristic fungi and identified bacterial markers.

[0090] (3) Based on the preliminary screening results, we further optimized and constructed a three-fungus combined diagnostic model. This model integrates... s__Wallemia sp (Optimal threshold 0.0014%) s__Cutaneotrichosporon curvatum (Optimal threshold 0.2073%) s__Penicillium cinnamopurpureum The abundance feature (optimal threshold 0.0002%) was used to calculate the probability of AH prevalence using the logistic regression equation P(AH) = 1 / (1 + exp(-(-0.497 + 59299.74 * s__Wallemia_sp - 6.407 * s__Cutaneotrichosporon_curvatum + 534584.952 * s__Penicillium_cinnamopurpureum))). That is, when P(AH) > 0.377, the patient was diagnosed with AH. This model demonstrated excellent diagnostic efficacy (AUC = 0.946), with significantly higher sensitivity (90%) and specificity (92.3%) than single fungal markers and bacterial models. The positive predictive value was 94.7%, and the negative predictive value was 84.7%. Figure 12 A, Table 4).

[0091] Table 4: Characteristics of AUD and AH Enteric Fungal Differential Diagnostic Models

[0092] Model AUC Best_Threshold Abundance_Threshold Sensitivity Specificity PPV NPV Accuracy Precision Youden_Index 0.8 0.643 0.0014% 0.625 0.962 0.962 0.625 0.758 0.962 0.587 0.738 0.717 0.2073% 0.5 0.885 0.87 0.535 0.652 0.87 0.385 0.775 0.705 0.0002% 0.55 1 1 0.591 0.727 1 0.55 0.688 0.535 0.0167% 0.9 0.5 0.735 0.765 0.742 0.735 0.4 0.738 0.648 0.0055% 0.5 0.962 0.952 0.556 0.682 0.952 0.462 CombinedModel 0.946 0.377 0.9 0.923 0.947 0.857 0.909 0.947 0.823 Validatio Model 0.736 0.139 1 0.545 0.762 1 0.815 0.762 0.545

[0093] (4) We validated the established multi-fungal biomarker combined diagnostic model on an independent validation set (11 AUD patients and 16 AH patients). Figure 12 B). Confusion matrix analysis showed that the model performed well in classifying and predicting the 27 samples. Specifically, the model excelled in identifying the AH group (positive class), correctly classifying all 16 AH samples (sensitivity 100%), while correctly identifying 6 out of 11 AUD samples (specificity 54.5%). The overall accuracy reached 81.48% (95% confidence interval: 0.6192–0.937). The model exhibited a clear asymmetric identification characteristic: it had perfect detection capability for AH cases (negative predictive value 100%), but its exclusion capability for non-AH cases was relatively limited (positive predictive value 76.2%). Based on an AH prevalence of 59.26%, the model's equilibrium accuracy was 77.27%, and the Kappa coefficient was 0.5872 (P=0.07364), indicating a moderate degree of consistency between the model's predictions and the actual diagnosis.

[0094] Based on the excellent performance of current single-modal models (AUC = 0.963 for bacterial models and AUC = 0.946 for fungal models), we propose an innovative approach to constructing a combined bacterial-fungal model. This approach is primarily based on two important findings: First, although the validation set AUC (bacteria 0.77 / fungus 0.736) is lower than the training set AUC, this performance fluctuation highlights the necessity of constructing a combined model. More importantly, the two types of microbial biomarkers exhibit significant complementary characteristics: the bacterial model demonstrates excellent specificity (92.3%), while the fungal model possesses perfect sensitivity (100%). This unique complementary characteristic provides a solid theoretical foundation for developing novel combined diagnostic models—by integrating the advantages of both types of biomarkers, not only can the model's stability on the validation set be improved, but also the synergistic optimization of sensitivity and specificity can be achieved, thereby constructing a diagnostic system with greater clinical practical value.

[0095] 4. Differential Diagnostic Model of AUD and AH Intestinal Bacteria Combined with Fungi

[0096] (1) Next, by integrating dominant bacterial and fungal biomarkers, an innovative multi-kingdom microbial joint diagnostic model was established. During the training set construction process, we used machine learning methods to screen for the most diagnostically valuable combination of microbial biomarkers, including three types of bacteria (…). s__Streptococcus oralis , s__Anaeromassilibacillus_ senegalensis , s__Enterocloster bolteae ) and three fungi ( s__Wallemia sp , s__ Cutaneotrichosporon curvatum , s__Penicillium cinnamopurpureumThe diagnostic equation established through logistic regression analysis is: P(AH) = 1 / (1 + exp(-(-0.666 + 459*)). s__Streptococcus oralis -161* s__Enterocloster bolteae -508* s__Anaeromassilibacillus senegalensis +4e+04* s__Wallemia sp -8.34* s__Cutaneotrichosporon curvatum +5.09e+05* s__Penicillium_ cinnamopurpureum The variables correspond to the standardized abundance of the aforementioned microbial markers. Specifically, when the calculated P(AH) > 0.423 based on the combined abundance of the three fungi, the patient is diagnosed with AH. This model demonstrated excellent diagnostic performance on the training set, with an AUC curve area of ​​0.963 (95% CI: 0.925-1), sensitivity of 87.5%, and specificity of 92.3%. The positive predictive value (94.6%) and negative predictive value (82.8%) were both at ideal levels. Figure 13 A, Table 5).

[0097] Table 5: Characteristics of the differential diagnostic model for AUD and AH intestinal bacteria combined with fungi

[0098] Model AUC Best_Threshold Abundance_Threshold Sensitivity Specificity PPV NPV Accuracy Precision Youden_Index 0.768 0.505 0.0407% 0.775 0.731 0.816 0.679 0.758 0.816 0.506 0.73 0.664 0.0629% 0.7 0.692 0.778 0.6 0.697 0.778 0.392 0.718 0.693 0.0111% 0.55 0.808 0.815 0.538 0.652 0.815 0.358 0.8 0.643 0.0014% 0.625 0.962 0.962 0.625 0.758 0.962 0.587 0.738 0.717 0.2073% 0.5 0.885 0.87 0.535 0.652 0.87 0.385 0.775 0.705 0.0002% 0.55 1 1 0.591 0.727 1 0.55 Combined Model 0.963 0.423 0.875 0.923 0.946 0.828 0.894 0.946 0.798 Validation Model 0.77 0.073 0.938 0.636 0.789 0.875 0.815 0.789 0.574

[0099] (2) In the independent validation set (11 AUD patients and 16 AH patients), the combined model demonstrated excellent generalization ability and clinical applicability. The model showed excellent classification performance, with an overall accuracy of 81.48% (95% CI: 0.6192-0.937). The confusion matrix showed that the model had a very strong ability to identify AH cases (positive class), with a sensitivity of 93.75%, successfully detecting 15 / 16 AH patients; while maintaining a specificity of 63.64% for AUD cases. Notably, while maintaining a high negative predictive value (87.50%), the model also achieved a positive predictive value of 78.95%, indicating its reliable diagnostic value in practical applications. The Kappa coefficient of 0.5994 (P=0.37109) showed that the model prediction and the actual diagnosis had a moderate degree of consistency. The balanced accuracy of 78.69% further confirmed the balanced performance of the model between the two classes of samples. Of particular note is that the model only misdiagnosed one case of AH (false negative) and misdiagnosed four cases of AUD (false positive). This error distribution pattern is especially valuable in clinical settings, prioritizing a high detection rate for AH. Based on an AH prevalence rate of 59.26%, this model demonstrates promising potential for clinical application. Figure 13 B, Table 5).

[0100] Comparative analysis revealed that this multi-kingdom microbial model demonstrated significantly improved diagnostic efficacy compared to single bacterial or fungal models. The combination of microbial biomarkers in the model reflected the characteristic gut microbiota dysbiosis pattern of ALD, namely, a decrease in probiotics and an increase in opportunistic pathogens. These results not only confirm that integrating bacterial and fungal biomarkers can construct a more precise ALD diagnostic system but also provide a new perspective for understanding the role of multi-kingdom gut microbial interactions in liver disease progression. This model offers an innovative microbiome solution for clinical practice.

[0101] 5. Differential diagnostic model for AST in patients with combined intestinal bacteria and fungi in patients with AUD and AH

[0102] (1) In the analysis of basic patient information, we found that AST was significantly higher in patients with AH than in patients with AUD (27.3±20.9 vs 101.9±153, P<0.0001) (Table 1, Table 2). Therefore, based on the previously established microbial biomarker model, we further integrated AST, an important clinical variable, and constructed an innovative "microbial-AST" multi-parameter diagnostic model. The penalized logistic regression prediction formula was established using the Bootstrap function of R language: P(AH)=1 / (1+exp(-(-2.66+180* s__ Streptococcus oralis -67.3* s__Enterocloster bolteae -1.22e+03* s__ Anaeromassilibacillus senegalensis +64.9* s__Wallemia sp -1.8* s__ Cutaneotrichosporon curvatum +902* s__Penicillium cinnamopurpureum +0.085*AST). This model demonstrated near-perfect diagnostic performance on the training set (AUC=0.995, 95%CI: 0.985-1.000), with a sensitivity of 97.5%, specificity of 100%, and positive and negative predictive values ​​both exceeding 95%. Specifically, when the abundance of three bacteria and three fungi is combined, and the patient's AST-calculated P(AH) is greater than 0.527, the patient is diagnosed with AH. Figure 14 A).

[0103] Model AUC Best_Threshold Best_Abundance Sensitivity Specificity PPV NPV Accuracy Youden_Index 0.768 0.505 0.0407% 0.775 0.731 0.816 0.679 0.758 0.506 0.73 0.664 0.0629% 0.7 0.692 0.778 0.6 0.697 0.392 0.718 0.693 0.0111% 0.55 0.808 0.815 0.538 0.652 0.358 0.8 0.643 0.0014% 0.625 0.962 0.962 0.625 0.758 0.587 0.738 0.717 0.2073% 0.5 0.885 0.87 0.535 0.652 0.385 0.775 0.705 0.0002% 0.55 1 1 0.591 0.727 0.55 AST 0.954 0.572 32.5 0.9 0.962 0.973 0.862 0.924 0.862 Combined Model 0.995 0.527 0.975 1 1 0.963 0.985 0.975 Validation Model 0.96 0.527 0.75 1 1 0.733 0.852 0.75

[0104] (2) In the independent validation set (11 AUD patients and 16 AH patients), the multiparameter model demonstrated excellent diagnostic performance and superior clinical applicability. Figure 14B). The overall accuracy of the model reached 85.19% (95% CI: 0.6627-0.9581). Key performance indicators showed perfect identification of AUD patients (100% specificity, 11 / 11 correct classifications) and maintained 75.00% sensitivity for AH patients (12 / 16 correct identifications). The model demonstrated excellent predictive power, with a positive predictive value of 100% and a negative predictive value of 73.33%. Statistical tests confirmed the model's high reliability (Kappa=0.7097, P=0.133614) and excellent inter-class balance (balance accuracy 87.50%). Notably, while maintaining zero false positives for AUD patients (0% false positive rate), the model only missed 4 cases of AH (25% false negative rate). Figure 14 (C) This error distribution pattern is extremely valuable in clinical screening scenarios. Based on an AH prevalence rate of 59.26%, this model provides a highly specific microbiome solution for early ALD screening.

[0105] (3) It is worth noting that this multimodal joint diagnostic model exhibits significant performance improvement: in the training set, the model's AUC value reaches 0.995 (95% CI: 0.985-1.000), demonstrating superior discriminative ability compared to the simple microbial model. Simultaneously, the model achieves an excellent balance between a sensitivity of 97.5% and a specificity of 100%, realizing a high disease detection rate and a low misdiagnosis rate. These advantages make the model particularly suitable for clinical applications, effectively identifying potential AH patients while minimizing overdiagnosis of AUD.

[0106] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A biomarker for differentiating alcohol use disorder from alcoholic hepatitis, characterized in that, The biomarkers include oral streptococci. Streptococcus oralis Senegalese anaerobic bacteria Anaeromassilibacillus senegalensis Clostridium botulinum Enterocloster bolteae Trichosporium curvatureum Cutaneotrichosporon curvatum Worley mold Wallemia sp. and Penicillium cinnamon Penicillium cinnamopurpureum .

2. The biomarker according to claim 1, characterized in that, The biomarkers also include aspartate aminotransferase (AST).

3. The application of a reagent for specifically detecting the biomarker of claim 1 in the preparation of predictive products, characterized in that, The predictive product is used to differentiate between alcohol use disorder and alcoholic hepatitis.

4. The application according to claim 3, characterized in that, The predicted product is a reagent kit or a detection system.

5. A device for differentiating between alcohol use disorder and alcoholic hepatitis, characterized in that, It includes a detection unit and a data analysis unit, wherein: The detection unit is used to acquire samples, detect and determine the abundance of microorganisms in the samples; the microorganisms are the biomarkers described in claim 1. The data analysis unit is used to analyze the detection results of the detection unit in combination with the serum aspartate aminotransferase content to make identification results; The data analysis unit differentiates between alcohol use disorder and alcoholic hepatitis based on a prediction formula, which is: P(AH)=1 / (1+exp(-(-2.66+180*) s__Streptococcus_oralis -67.3* s__Enterocloster_ bolteae -1.22e+03* s__Anaeromassilibacillus_senegalensis +64.9* s__Wallemia_sp -1.8* s__Cutaneotrichosporon_curvatum +902* s__Penicillium_cinnamopurpureum +0.085*AST), where AST represents the serum aspartate aminotransferase (AST) level in the patient; the predictive criterion is that when P(AH) is greater than 0.527, the patient is diagnosed with alcoholic hepatitis.

6. The apparatus according to claim 5, characterized in that, The detection unit performs detection and obtains detection results according to the following operations: Test fecal samples to obtain DNA data; The data analysis unit is used to analyze and process the detection results of the detection unit, including: sequencing the DNA data of the fecal sample to obtain intestinal microbial genome data; The relative abundance information of each biomarker was obtained by analyzing the gut microbiome genome data. Based on the relative abundance information, gut microbiota characteristic data are determined.

Images