Clinical domain knowledge-derived template improves post hoc AI explanations in pneumothorax classification.

OBJECTIVE: Pneumothorax is an acute thoracic disease caused by abnormal air collection between the lungs and chest wall. Recently, artificial intelligence (AI), especially deep learning (DL), has been increasingly employed for automating the diagnostic process of pneumothorax. To address the opaqueness often associated with DL models, explainable artificial intelligence (XAI) methods have been introduced to outline regions related to pneumothorax. However, these explanations sometimes diverge from actual lesion areas, highlighting the need for further improvement. METHOD: We propose a template-guided approach to incorporate the clinical knowledge of pneumothorax into model explanations generated by XAI methods, thereby enhancing the quality of the explanations. Utilizing one lesion delineation created by radiologists, our approach first generates a template that represents potential areas of pneumothorax occurrence. This template is then superimposed on model explanations to filter out extraneous explanations that fall outside the template's boundaries. To validate its efficacy, we carried out a comparative analysis of three XAI methods (Saliency Map, Grad-CAM, and Integrated Gradients) with and without our template guidance when explaining two DL models (VGG-19 and ResNet-50) in two real-world datasets (SIIM-ACR and ChestX-Det). RESULTS: The proposed approach consistently improved baseline XAI methods across twelve benchmark scenarios built on three XAI methods, two DL models, and two datasets. The average incremental percentages, calculated by the performance improvements over the baseline performance, were 97.8% in Intersection over Union (IoU) and 94.1% in Dice Similarity Coefficient (DSC) when comparing model explanations and ground-truth lesion areas. We further visualized baseline and template-guided model explanations on radiographs to showcase the performance of our approach. CONCLUSIONS: In the context of pneumothorax diagnoses, we proposed a template-guided approach for improving model explanations. Our approach not only aligns model explanations more closely with clinical insights but also exhibits extensibility to other thoracic diseases. We anticipate that our template guidance will forge a novel approach to elucidating AI models by integrating clinical domain expertise.

Optical coherence tomography choroidal enhancement using generative deep learning.

Spectral-domain optical coherence tomography (SDOCT) is the gold standard of imaging the eye in clinics. Penetration depth with such devices is, however, limited and visualization of the choroid, which is essential for diagnosing chorioretinal disease, remains limited. Whereas swept-source OCT (SSOCT) devices allow for visualization of the choroid these instruments are expensive and availability in praxis is limited. We present an artificial intelligence (AI)-based solution to enhance the visualization of the choroid in OCT scans and allow for quantitative measurements of choroidal metrics using generative deep learning (DL). Synthetically enhanced SDOCT B-scans with improved choroidal visibility were generated, leveraging matching images to learn deep anatomical features during the training. Using a single-center tertiary eye care institution cohort comprising a total of 362 SDOCT-SSOCT paired subjects, we trained our model with 150,784 images from 410 healthy, 192 glaucoma, and 133 diabetic retinopathy eyes. An independent external test dataset of 37,376 images from 146 eyes was deployed to assess the authenticity and quality of the synthetically enhanced SDOCT images. Experts' ability to differentiate real versus synthetic images was poor (47.5% accuracy). Measurements of choroidal thickness, area, volume, and vascularity index, from the reference SSOCT and synthetically enhanced SDOCT, showed high Pearson's correlations of 0.97 [95% CI: 0.96-0.98], 0.97 [0.95-0.98], 0.95 [0.92-0.98], and 0.87 [0.83-0.91], with intra-class correlation values of 0.99 [0.98-0.99], 0.98 [0.98-0.99], and 0.95 [0.96-0.98], 0.93 [0.91-0.95], respectively. Thus, our DL generative model successfully generated realistic enhanced SDOCT data that is indistinguishable from SSOCT images providing improved visualization of the choroid. This technology enabled accurate measurements of choroidal metrics previously limited by the imaging depth constraints of SDOCT. The findings open new possibilities for utilizing affordable SDOCT devices in studying the choroid in both healthy and pathological conditions.

Identification and validation of core microbes for the formation of the characteristic flavor of fermented oysters (Crassostrea gigas).

Self-fermented oyster homogenates were prepared to investigate core microbes and their correlations with flavor formation mechanisms. Five bacterial and four fungal genera were identified. Correlation analysis showed that Saccharomyces cerevisiae, Kazachstania, and L. pentosus were core species for the flavor of fermented products. Four core microbes were selected for inoculation into homogenates. Twelve key aroma compounds with odor activity values >1 were identified by gas chromatography-mass spectrometry. L. plantarum and S. cerevisiae were beneficial for producing key aroma compounds such as 1-octen-3-ol, (E,Z)-2,6-nonadienal, and heptanal. Fermentation with four microbes resulted in significant increases in contents of Asp, Glu, Lys, inosine monophosphate, and guanosine monophosphate, which provided freshness and sweetness. Fermentation with four microbes resulted in high digestibility, antioxidant abilities, and zinc contents. This study has elucidated the mechanism of flavor formation by microbial action and provides a reference for targeted flavor control in fermented oyster products.

Contribution of phospholipase B to the formation of characteristic flavor in steamed sturgeon meat.

Sensory analysis and untargeted lipidomics were employed to study the impact of phospholipase B (PLB) on lipid oxidation and flavor in steamed sturgeon meat, revealing the inherent relationship between lipid oxidation and flavor regulation. The research verified that PLB effectively suppresses fat oxidation and improves the overall taste of steamed sturgeon meat. Furthermore, the PLB group identified 52 compounds, and the content of odor substances such as isoamyl alcohol and hexanal was reduced compared with other groups. Finally, lipid substances containing eicosapentaenoic acid (EPA) or docosahexaenoic acid (DHA) were screened out from 32 kinds of differential phospholipids. Through Pearson correlation analysis, it was observed that certain differential phospholipids such as PC (22:6) and PC (22:5) exhibited varying correlations with odor substances like hexanal and isovaleraldehyde. These findings suggest that PLB specifically affects certain phospholipids, leading to the production of distinct volatile substances through oxidative degradation.

Study on antimicrobial activity of sturgeon skin mucus polypeptides (Rational Design, Self-Assembly and Application).

Despite the favorable biocompatibility of natural antimicrobial peptides (AMPs), their scarcity limits their practical application. Through rational design, the activity of AMPs can be enhanced to expand their application. In this study, we selected a natural sturgeon epidermal mucus peptide, AP-16 (APATPAAPALLPLWLL), as the model molecule and studied its conformational regulation and antimicrobial activity through amino acid substitutions and N-terminal lipidation. The structural and morphological transitions of the peptide self-assemblies were investigated using circular dichroism and transmission electron microscopy. Following amino acid substitution, the conformation of AL-16 (AKATKAAKALLKLWLL) did not change. Following N-terminal alkylation, the C8-AL-16 and C12-AL-16 conformations changed from random coil to ß-sheet or α-helix, and the self-assembly changed from nanofibers to nanospheres. AL-16, C8-AL-16, and C8-AL-16 presented significant antimicrobial activity against Pseudomonas and Shewanella at low concentrations. N-terminal alkylation effectively extended the shelf life of Litopenaeus vannamei. These results support the application of natural AMPs.

Research progress of fishy odor in aquatic products: From substance identification, formation mechanism, to elimination pathway.

Fishy odor in aquatic products has a significant impact on the purchasing decisions of consumers. The production of aquatic products is a complex process involving culture, processing, transportation, and storage, which contribute to decreases in flavor and quality. This review systematically summarizes the fishy odor composition, identification methods, generation mechanism, and elimination methods of fishy odor compounds from their origin and formation to their elimination. Fishy odor compounds include aldehydes (hexanal, heptanal, and nonanal), alcohols (1-octen-3-ol), sulfur-containing compounds (dimethyl sulfide), and amines (trimethylamine). The mechanism of action of various factors affecting fishy odor is revealed, including environmental factors, enzymatic reactions, lipid oxidation, protein degradation, and microbial metabolism. Furthermore, the control and removal of fishy odor are briefly summarized and discussed, including masking, elimination, and conversion. This study provides a theoretical basis from source to elimination for achieving targeted regulation of the flavor of aquatic products, promoting industrial innovation and upgrading.

Correlation investigation between core microbe inoculation and the evolution of flavor characteristics during the storage of sturgeon caviar (Acipenser gueldenstaedtii).

The volatile and non-volatile compounds were monitored to investigate the microbial evolution associated with the characteristic flavors for sturgeon caviar during refrigeration. The results revealed that the composition of volatile compounds changed significantly with prolonged refrigeration time, especially hexanal, nonanal, phenylacetaldehyde, 3-methyl butyraldehyde, and 1-octen-3-ol. The nonvolatile metabolites were mainly represented by the increase of bitter amino acids (Thr. Ser, Gly, Ala, and Pro) and a decrease in polyunsaturated fatty acids, especially an 18.63 % decrease in 5 months of storage. A total of 332 differential metabolites were mainly involved in the biosynthetic metabolic pathways of α-linolenic acid, linoleic acid, and arachidonic acid. The precursors associated with flavor evolution were mainly phospholipids, including oleic, linoleic, arachidonic, eicosapentaenoic (EPA), and docosahexaenoic (DHA) acids. The most abundant at the genus level was Serratia, followed by Arsenophnus, Rhodococcus, and Pseudomonas, as obtained by high-throughput sequencing. Furthermore, seven core microorganisms were isolated and characterized from refrigerated caviar. Among them, inoculation with Mammalian coccus and Bacillus chrysosporium restored the flavor profile of caviar and enhanced the content of nonvolatile precursors, contributing to the characteristic aroma attributes of sturgeon caviar. The study presents a theoretical basis for the exploitation of technologies for quality stabilization and control of sturgeon caviar during storage.

Geometric Correspondence-Based Multimodal Learning for Ophthalmic Image Analysis.

Color fundus photography (CFP) and Optical coherence tomography (OCT) images are two of the most widely used modalities in the clinical diagnosis and management of retinal diseases. Despite the widespread use of multimodal imaging in clinical practice, few methods for automated diagnosis of eye diseases utilize correlated and complementary information from multiple modalities effectively. This paper explores how to leverage the information from CFP and OCT images to improve the automated diagnosis of retinal diseases. We propose a novel multimodal learning method, named geometric correspondence-based multimodal learning network (GeCoM-Net), to achieve the fusion of CFP and OCT images. Specifically, inspired by clinical observations, we consider the geometric correspondence between the OCT slice and the CFP region to learn the correlated features of the two modalities for robust fusion. Furthermore, we design a new feature selection strategy to extract discriminative OCT representations by automatically selecting the important feature maps from OCT slices. Unlike the existing multimodal learning methods, GeCoM-Net is the first method that formulates the geometric relationships between the OCT slice and the corresponding region of the CFP image explicitly for CFP and OCT fusion. Experiments have been conducted on a large-scale private dataset and a publicly available dataset to evaluate the effectiveness of GeCoM-Net for diagnosing diabetic macular edema (DME), impaired visual acuity (VA) and glaucoma. The empirical results show that our method outperforms the current state-of-the-art multimodal learning methods by improving the AUROC score 0.4%, 1.9% and 2.9% for DME, VA and glaucoma detection, respectively.

Effects of Sous-Vide on Quality, Structure and Flavor Characteristics of Tilapia Fillets.

To investigate the effects of traditional high-temperature cooking and sous-vide cooking on the quality of tilapia fillets, muscle microstructure, texture, lipid oxidation, protein structure, and volatile compounds were analyzed. In comparison with samples subjected to traditional high-temperature cooking, sous-vide-treated samples exhibited less protein denaturation, a secondary structure dominated by α-helices, a stable and compact structure, a significantly higher moisture content, and fewer gaps in muscle fibers. The hardness of the sous-vide-treated samples was higher than that of control samples, and the extent of lipid oxidation was significantly reduced. The sous-vide cooking technique resulted in notable changes in the composition and relative content of volatile compounds, notably leading to an increase in the presence of 1-octen-3-ol, α-pinene, and dimethyl sulfide, and a decrease in the levels of hexanal, D-limonene, and methanethiol. Sous-vide treatment significantly enhanced the structural stability, hardness, and springiness of muscle fibers in tilapia fillets and reduced nutrient loss, enriched flavor, and mitigated effects on taste and fishy odor.

Identification and Verification of Novel Umami Peptides Isolated from Hybrid Sturgeon Meat (Acipenser baerii × Acipenser schrenckii).

To explore the umami mechanism in sturgeon meat, five peptides (ERRY, VRGPR, LKYPLE, VKKVFK, and YVVFKD) were isolated and identified by ultrafiltration, gel filtration chromatography, and UPLC-QTOF-MS/MS. The omission test confirmed that the five umami peptides contributed to the umami taste of sturgeon meat. Also, the peptides had the double effective role of enhancing both umami and saltiness. The threshold of ERRY was only 0.031, which exceeded most umami peptides in the last 3 years. Molecular docking results showed that five peptides could easily bind to Gly167, Ser170, and Try218 residues in T1R3 through hydrogen bonds and electrostatic interactions. Furthermore, molecular dynamics simulations indicated that hydrogen bonds and hydrophobic interactions were the main intermolecular interaction forces. This study could contribute to revealing the umami taste mechanism of sturgeon meat and provide new insights for effective screening of short umami peptides.

Uncertainty-inspired open set learning for retinal anomaly identification.

Failure to recognize samples from the classes unseen during training is a major limitation of artificial intelligence in the real-world implementation for recognition and classification of retinal anomalies. We establish an uncertainty-inspired open set (UIOS) model, which is trained with fundus images of 9 retinal conditions. Besides assessing the probability of each category, UIOS also calculates an uncertainty score to express its confidence. Our UIOS model with thresholding strategy achieves an F1 score of 99.55%, 97.01% and 91.91% for the internal testing set, external target categories (TC)-JSIEC dataset and TC-unseen testing set, respectively, compared to the F1 score of 92.20%, 80.69% and 64.74% by the standard AI model. Furthermore, UIOS correctly predicts high uncertainty scores, which would prompt the need for a manual check in the datasets of non-target categories retinal diseases, low-quality fundus images, and non-fundus images. UIOS provides a robust method for real-world screening of retinal anomalies.

GAMMA challenge: Glaucoma grAding from Multi-Modality imAges.

Glaucoma is a chronic neuro-degenerative condition that is one of the world's leading causes of irreversible but preventable blindness. The blindness is generally caused by the lack of timely detection and treatment. Early screening is thus essential for early treatment to preserve vision and maintain life quality. Colour fundus photography and Optical Coherence Tomography (OCT) are the two most cost-effective tools for glaucoma screening. Both imaging modalities have prominent biomarkers to indicate glaucoma suspects, such as the vertical cup-to-disc ratio (vCDR) on fundus images and retinal nerve fiber layer (RNFL) thickness on OCT volume. In clinical practice, it is often recommended to take both of the screenings for a more accurate and reliable diagnosis. However, although numerous algorithms are proposed based on fundus images or OCT volumes for the automated glaucoma detection, there are few methods that leverage both of the modalities to achieve the target. To fulfil the research gap, we set up the Glaucoma grAding from Multi-Modality imAges (GAMMA) Challenge to encourage the development of fundus & OCT-based glaucoma grading. The primary task of the challenge is to grade glaucoma from both the 2D fundus images and 3D OCT scanning volumes. As part of GAMMA, we have publicly released a glaucoma annotated dataset with both 2D fundus colour photography and 3D OCT volumes, which is the first multi-modality dataset for machine learning based glaucoma grading. In addition, an evaluation framework is also established to evaluate the performance of the submitted methods. During the challenge, 1272 results were submitted, and finally, ten best performing teams were selected for the final stage. We analyse their results and summarize their methods in the paper. Since all the teams submitted their source code in the challenge, we conducted a detailed ablation study to verify the effectiveness of the particular modules proposed. Finally, we identify the proposed techniques and strategies that could be of practical value for the clinical diagnosis of glaucoma. As the first in-depth study of fundus & OCT multi-modality glaucoma grading, we believe the GAMMA Challenge will serve as an essential guideline and benchmark for future research.

Correction: From 2 dimensions to 3rd dimension: Quantitative prediction of anterior chamber depth from anterior segment photographs via deep-learning.

[This corrects the article DOI: 10.1371/journal.pdig.0000193.].

Privacy-preserving continual learning methods for medical image classification: a comparative analysis.

Background: The implementation of deep learning models for medical image classification poses significant challenges, including gradual performance degradation and limited adaptability to new diseases. However, frequent retraining of models is unfeasible and raises concerns about healthcare privacy due to the retention of prior patient data. To address these issues, this study investigated privacy-preserving continual learning methods as an alternative solution. Methods: We evaluated twelve privacy-preserving non-storage continual learning algorithms based deep learning models for classifying retinal diseases from public optical coherence tomography (OCT) images, in a class-incremental learning scenario. The OCT dataset comprises 108,309 OCT images. Its classes include normal (47.21%), drusen (7.96%), choroidal neovascularization (CNV) (34.35%), and diabetic macular edema (DME) (10.48%). Each class consisted of 250 testing images. For continuous training, the first task involved CNV and normal classes, the second task focused on DME class, and the third task included drusen class. All selected algorithms were further experimented with different training sequence combinations. The final model's average class accuracy was measured. The performance of the joint model obtained through retraining and the original finetune model without continual learning algorithms were compared. Additionally, a publicly available medical dataset for colon cancer detection based on histology slides was selected as a proof of concept, while the CIFAR10 dataset was included as the continual learning benchmark. Results: Among the continual learning algorithms, Brain-inspired-replay (BIR) outperformed the others in the continual learning-based classification of retinal diseases from OCT images, achieving an accuracy of 62.00% (95% confidence interval: 59.36-64.64%), with consistent top performance observed in different training sequences. For colon cancer histology classification, Efficient Feature Transformations (EFT) attained the highest accuracy of 66.82% (95% confidence interval: 64.23-69.42%). In comparison, the joint model achieved accuracies of 90.76% and 89.28%, respectively. The finetune model demonstrated catastrophic forgetting in both datasets. Conclusion: Although the joint retraining model exhibited superior performance, continual learning holds promise in mitigating catastrophic forgetting and facilitating continual model updates while preserving privacy in healthcare deep learning models. Thus, it presents a highly promising solution for the long-term clinical deployment of such models.

Deep Neural Network Augments Performance of Junior Residents in Diagnosing COVID-19 Pneumonia on Chest Radiographs.

Chest X-rays (CXRs) are essential in the preliminary radiographic assessment of patients affected by COVID-19. Junior residents, as the first point-of-contact in the diagnostic process, are expected to interpret these CXRs accurately. We aimed to assess the effectiveness of a deep neural network in distinguishing COVID-19 from other types of pneumonia, and to determine its potential contribution to improving the diagnostic precision of less experienced residents. A total of 5051 CXRs were utilized to develop and assess an artificial intelligence (AI) model capable of performing three-class classification, namely non-pneumonia, non-COVID-19 pneumonia, and COVID-19 pneumonia. Additionally, an external dataset comprising 500 distinct CXRs was examined by three junior residents with differing levels of training. The CXRs were evaluated both with and without AI assistance. The AI model demonstrated impressive performance, with an Area under the ROC Curve (AUC) of 0.9518 on the internal test set and 0.8594 on the external test set, which improves the AUC score of the current state-of-the-art algorithms by 1.25% and 4.26%, respectively. When assisted by the AI model, the performance of the junior residents improved in a manner that was inversely proportional to their level of training. Among the three junior residents, two showed significant improvement with the assistance of AI. This research highlights the novel development of an AI model for three-class CXR classification and its potential to augment junior residents' diagnostic accuracy, with validation on external data to demonstrate real-world applicability. In practical use, the AI model effectively supported junior residents in interpreting CXRs, boosting their confidence in diagnosis. While the AI model improved junior residents' performance, a decline in performance was observed on the external test compared to the internal test set. This suggests a domain shift between the patient dataset and the external dataset, highlighting the need for future research on test-time training domain adaptation to address this issue.

Metabolomics investigation on the volatile and non-volatile composition in enzymatic hydrolysates of Pacific oyster (Crassostrea gigas).

To investigate the differences of volatile and non-volatile metabolites between oyster enzymatic hydrolysates and boiling concentrates, molecular sensory analysis and untargeted metabolomics were employed. "Grassy," "fruity," "oily/fatty," "fishy," and "metallic" were identified as sensory attributes used to evaluate different processed oyster homogenates. Sixty-nine and 42 volatiles were identified by gas chromatography-ion mobility spectrometry and gas chromatography-mass spectrometry, respectively. Pentanal, 1-penten-3-ol, hexanal, (E)-2-pentenal, heptanal, (E)-2-hexenal, 4-octanone, (E)-4-heptenal, 3-octanone, octanal, nonanal, 1-octen-3-ol, benzaldehyde, (E)-2-nonenal, and (E, Z)-2,6-nonadienal were detected as the key odorants (OAV > 1) after enzymatic hydrolysis. Hexanal, (E)-4-heptenal, and (E)-2-pentenal were significantly associated with off-odor, and 177 differential metabolites were classified. Aspartate, glutamine, alanine, and arginine were the key precursors affecting the flavor profile. Linking sensory descriptors to volatile and nonvolatile components of different processed oyster homogenates will provide information for the process and quality improvement of oyster products.

From 2 dimensions to 3rd dimension: Quantitative prediction of anterior chamber depth from anterior segment photographs via deep-learning.

Anterior chamber depth (ACD) is a major risk factor of angle closure disease, and has been used in angle closure screening in various populations. However, ACD is measured from ocular biometer or anterior segment optical coherence tomography (AS-OCT), which are costly and may not be readily available in primary care and community settings. Thus, this proof-of-concept study aims to predict ACD from low-cost anterior segment photographs (ASPs) using deep-learning (DL). We included 2,311 pairs of ASPs and ACD measurements for algorithm development and validation, and 380 pairs for algorithm testing. We captured ASPs with a digital camera mounted on a slit-lamp biomicroscope. Anterior chamber depth was measured with ocular biometer (IOLMaster700 or Lenstar LS9000) in data used for algorithm development and validation, and with AS-OCT (Visante) in data used for testing. The DL algorithm was modified from the ResNet-50 architecture, and assessed using mean absolute error (MAE), coefficient-of-determination (R2), Bland-Altman plot and intraclass correlation coefficients (ICC). In validation, our algorithm predicted ACD with a MAE (standard deviation) of 0.18 (0.14) mm; R2 = 0.63. The MAE of predicted ACD was 0.18 (0.14) mm in eyes with open angles and 0.19 (0.14) mm in eyes with angle closure. The ICC between actual and predicted ACD measurements was 0.81 (95% CI 0.77, 0.84). In testing, our algorithm predicted ACD with a MAE of 0.23 (0.18) mm; R2 = 0.37. Saliency maps highlighted the pupil and its margin as the main structures used in ACD prediction. This study demonstrates the possibility of predicting ACD from ASPs via DL. This algorithm mimics an ocular biometer in making its prediction, and provides a foundation to predict other quantitative measurements that are relevant to angle closure screening.

Capturing the impact of oral processing behavior and bolus formation on the dynamic sensory perception and composition of steamed sturgeon meat.

The effect of oral processing on flavor release and change in composition of steamed sturgeon meat was investigated. Oral processing caused changes in the concentrations of taste compounds including amino acids, 5'-nucleotides, organic acids, and Na+. Sensory omics demonstrated that the concentrations of 12 volatile compounds increased significantly (p < 0.05) during the initial stage of oral processing. There is no significant difference in microstructure, texture, and particle size of meat bolus. The top fifteen differential lipids which including eight phospholipids in all processed samples significantly (p < 0.05) correlated with the flavor release. A total of 589 differential proteins were detected in three samples with different chewing times (0, 12, and 30 s). Analysis of the correlations between odorants and 19 differential proteins was performed. Enriched pathways including fatty acid degradation, valine, leucine and isoleucine degradation, glycine, serine and threonine metabolism, and arachidonic acid metabolism were associated with flavor release during oral processing. This study aimed to investigate potential links between flavor release and biological processes during oral processing from a proteomics perspective.

Identification and validation of core microbes associated with key aroma formation in fermented pepper paste (Capsicum annuumL.).

Fermented peppers are usually obtained by the spontaneous fermentation of microorganisms attached to fresh peppers, and the variable microbial composition would lead to inconsistencies in flavor between batches. To demonstrate the roles of microorganisms in flavor formation, the core microbes closely associated with the key aroma compounds of fermented pepper paste were screened and validated in this study. Lactobacillus was the dominant bacterial genus in fermented pepper paste, whereas the main fungal genera were Alternaria and Kazachstania. Nine strains of the genera Lactobacillus, Weissella, Bacillus, Zygosaccharomyces, Kazachstania, Debaryomyces, and Pichia were isolated from fermented pepper paste. Eleven key aroma compounds were identified using gas chromatography combined with olfactometry and relative odor activity values. Correlation analysis showed that Zygosaccharomyces and Kazachstania were positively correlated with the majority of the key aroma compounds, whereas Lactobacillus was negatively correlated with them. Thus, Zygosaccharomyces and Kazachstania were identified as core genera associated with the key odorants. Finally, Zygosaccharomyces bisporus, Kazachstania humilis, and Lactiplantibacillus plantarum were used as starter cultures for fermented peppers, confirming that Z. bisporus and K. humilis were more beneficial for the key aroma compounds (e.g., acetate, linalool, and phenyl ethanol) rather than L. plantarum. This study contributed to understanding the flavor formation mechanism and provided references for the quality control of food fermentation.