Machine Learning-based Analysis Research Articles

Multiplex Microscopy Assay for Assessment of Therapeutic and Serum Antibodies against Emerging Pathogens

The emergence of novel pathogens, exemplified recently by the severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), highlights the need for rapidly deployable and adaptable diagnostic assays to assess their impact on human health and guide public health responses in future pandemics. In this study, we developed an automated multiplex microscopy assay coupled with machine learning-based analysis for antibody detection. To achieve multiplexing and simultaneous detection of multiple viral antigens, we devised a barcoding strategy utilizing a panel of HeLa-based cell lines. Each cell line expressed a distinct viral antigen, along with a fluorescent protein exhibiting a unique subcellular localization pattern for cell classification. Our robust, cell segmentation and classification algorithm, combined with automated image acquisition, ensured compatibility with a high-throughput approach. As a proof of concept, we successfully applied this approach for quantitation of immunoreactivity against different variants of SARS-CoV-2 spike and nucleocapsid proteins in sera of patients or vaccinees, as well as for the study of selective reactivity of monoclonal antibodies. Importantly, our system can be rapidly adapted to accommodate other SARS-CoV-2 variants as well as any antigen of a newly emerging pathogen, thereby representing an important resource in the context of pandemic preparedness.

Interpretable machine learning-based analysis of coal fly ash flotation by conditioning process intensification

ABSTRACT Conditioning process intensification is necessary for improving the fly ash flotation effect. Understanding the relationship between conditioning parameters and flotation results is crucial for optimizing the conditioning-flotation process. This study developed a machine learning framework for predicting the yield, loss on ignition, and removal rate of unburned carbon using data from conditioning-flotation experiments. Feature variables of conditioning speed, impeller type, stirred tank diameter, collector dosage, and frother dosage were employed to develop five machine learning models. After tuning the hyper-parameters with GridSearchCV technique, it was found that Random Forest Regression model, eXtreme Gradient Boosting Regression model, and Gradient Boosting Regression model demonstrated good agreements between the predicted data and test data for the yield, loss on ignition, and removal rate of unburned carbon. The maximum R2 values were 0.9184, 0.8339, and 0.9434, respectively, while the minimum MSE values were 3.30, 5.82, and 20.59, respectively. The interpretable analysis using SHapley Additive exPlanations (SHAP) method indicated that conditioning speed was the most significant feature variable affecting the yield, loss on ignition, and removal rate of unburned carbon, while impeller type had a secondary impact. This investigation is expected to reveal the important role of conditioning process intensification in optimizing fly ash flotation.

VivoBodySeg: Machine learning-based analysis of C. elegans immobilized in vivoChip for automated developmental toxicity testing.

Developmental toxicity (DevTox) tests evaluate the adverse effects of chemical exposures on an organism's development. While large animal tests are currently heavily relied on, the development of new approach methodologies (NAMs) is encouraging industries and regulatory agencies to evaluate these novel assays. Several practical advantages have made C. elegans a useful model for rapid toxicity testing and studying developmental biology. Although the potential to study DevTox is promising, current low-resolution and labor-intensive methodologies prohibit the use of C. elegans for sub-lethal DevTox studies at high throughputs. With the recent availability of a large-scale microfluidic device, vivoChip, we can now rapidly collect 3D high-resolution images of ~ 1,000 C. elegans from 24 different populations. In this paper, we demonstrate DevTox studies using a 2.5D U-Net architecture (vivoBodySeg) that can precisely segment C. elegans in images obtained from vivoChip devices, achieving an average Dice score of 97.80. The fully automated platform can analyze 36 GB data from each device to phenotype multiple body parameters within 35 min on a desktop PC at speeds ~ 140× faster than the manual analysis. Highly reproducible DevTox parameters (4-8% CV) and additional autofluorescence-based phenotypes allow us to assess the toxicity of chemicals with high statistical power.

Quantitative microbiology with widefield microscopy: navigating optical artefacts for accurate interpretations

Time-resolved live-cell imaging using widefield microscopy is instrumental in quantitative microbiology research. It allows researchers to track and measure the size, shape, and content of individual microbial cells over time. However, the small size of microbial cells poses a significant challenge in interpreting image data, as their dimensions approache that of the microscope’s depth of field, and they begin to experience significant diffraction effects. As a result, 2D widefield images of microbial cells contain projected 3D information, blurred by the 3D point spread function. In this study, we employed simulations and targeted experiments to investigate the impact of diffraction and projection on our ability to quantify the size and content of microbial cells from 2D microscopic images. This study points to some new and often unconsidered artefacts resulting from the interplay of projection and diffraction effects, within the context of quantitative microbiology. These artefacts introduce substantial errors and biases in size, fluorescence quantification, and even single-molecule counting, making the elimination of these errors a complex task. Awareness of these artefacts is crucial for designing strategies to accurately interpret micrographs of microbes. To address this, we present new experimental designs and machine learning-based analysis methods that account for these effects, resulting in accurate quantification of microbiological processes.

1205P A retrospective machine learning-based analysis of nationwide cancer CGP data across cancer types to identify features associated with recommendation of mutation-based therapy

1205P A retrospective machine learning-based analysis of nationwide cancer CGP data across cancer types to identify features associated with recommendation of mutation-based therapy

Multiomics and machine learning-based analysis of pancancer pseudouridine modifications

Pseudouridine widely affects the stability and function of RNA. However, our knowledge of pseudouridine properties in tumors is incomplete. We systematically analyzed pseudouridine synthases (PUSs) expression, genomic aberrations, and prognostic features in 10907 samples from 33 tumors. We found that the pseudouridine-associated pathway was abnormal in tumors and affected patient prognosis. Dysregulation of the PUSs expression pattern may arise from copy number variation (CNV) mutations and aberrant DNA methylation. Functional enrichment analyses determined that the PUSs expression was closely associated with the MYC, E2F, and MTORC1 signaling pathways. In addition, PUSs are involved in the remodeling of the tumor microenvironment (TME) in solid tumors, such as kidney and lung cancers. Particularly in lung cancer, increased expression of PUSs is accompanied by increased immune checkpoint expression and Treg infiltration. The best signature model based on more than 112 machine learning combinations had good prognostic ability in ACC, DLBC, GBM, KICH, MESO, THYM, TGCT, and PRAD tumors, and is expected to guide immunotherapy for 19 tumor types. The model was also effective in identifying patients with tumors amenable to etoposide, camptothecin, cisplatin, or bexarotene treatment. In conclusion, our work highlights the dysregulated features of PUSs and their role in the TME and patient prognosis, providing an initial molecular basis for future exploration of pseudouridine. Studies targeting pseudouridine are expected to lead to the development of potential diagnostic strategies and the evaluation and improvement of antitumor therapies.

Machine Learning-Based Analysis of Surface Hardening in Shot-Peened Superalloys

Machine Learning-Based Analysis of Surface Hardening in Shot-Peened Superalloys

Machine learning-based analysis of heavy metal contamination in Chinese lake basin sediments: Assessing influencing factors and policy implications

Sediments are important heavy metal sinks in lakes, crucial for ensuring water environment safety. Existing studies mainly focused on well-studied lakes, leaving gaps in understanding pollution patterns in specific basins and influencing factors.We compiled comprehensive sediment contamination data from literature and public datasets, including hydro-geomorphological, climatic, soil, landscape, and anthropogenic factors. Using advanced machine learning, we analyzed typical pollution factors to infer potential sources and migration pathways of pollutants and predicted pollution levels in basins with limited data availability. Our analysis of pollutant distribution data revealed that Cd had the most extensive pollution range, with the most severe pollution occurring in the Huaihe and Yangtze River basins. Furthermore, we identified distinct groups of driving factors influencing various heavy metals. Cd, Cr, and Pb were primarily influenced by human activities, while Cu and Ni were affected by both anthropogenic and natural factors, and Zn tended more towards natural sources. Our predictions indicated that, in addition to the typical highly polluted areas, the potential risk of Cd, Cu and Ni is higher in Xinjiang, and in Tibet and Qinghai, the potential risk of Cd, Cr, Cu and Ni is higher. Pb and Zn presented lower risks, except in the Huaihe and Yangtze River Basins. Temperature, wind, precipitation, precipitation rate, and the cation exchange capacity of soil significantly impacted the predictions of heavy metal pollution in sediments, suggesting that particulate migration, rainfall runoff, and soil erosion are likely the main pathways for pollutant migration into sediments. Considering the migration, pathways, and sources of pollutants, we propose strategies such as low-impact development and promoting sustainable transportation to mitigate pollution. This study provides the latest insights into heavy metal pollution in Chinese lake sediments, offering references for policy-making and water resource management.

Machine learning-based analysis identifies and validates serum exosomal proteomic signatures for the diagnosis of colorectal cancer

The potential of serum extracellular vesicles (EVs) as non-invasive biomarkers for diagnosing colorectal cancer (CRC) remains elusive. We employed an in-depth 4D-DIA proteomics and machine learning (ML) pipeline to identify key proteins, PF4 and AACT, for CRC diagnosis in serum EV samples from a discovery cohort of 37 cases. PF4 and AACT outperform traditional biomarkers, CEA and CA19-9, detected by ELISA in 912 individuals. Furthermore, we developed an EV-related random forest (RF) model with the highest diagnostic efficiency, achieving AUC values of 0.960 and 0.963 in the train and test sets, respectively. Notably, this model demonstrated reliable diagnostic performance for early-stage CRC and distinguishing CRC from benign colorectal diseases. Additionally, multi-omics approaches were employed to predict the functions and potential sources of serum EV-derived proteins. Collectively, our study identified the crucial proteomic signatures in serum EVs and established a promising EV-related RF model for CRC diagnosis in the clinic.

Differentially used codons among essential genes in bacteria identified by machine learning-based analysis.

Codon usage bias (CUB), the uneven usage of synonymous codons encoding the same amino acid, differs among genes within and across bacteria genomes. CUB is known to be influenced by gene expression and accordingly, CUB differs between the high-expression and low-expression genes in several bacteria. In this article, we have extended codon usage study considering gene essentiality as a feature. Using machine learning (ML) based approaches, we have analysed Relative Synonymous Codon Usage (RSCU) values between essential and non-essential genes in Escherichia coli and thirty-four other bacterial genomes whose gene essentiality features were available in public databases. We observed significant differences in codon usage patterns between essential and non-essential genes for majority of the bacterial genomes and accordingly, ML based classifiers achieved high area under curve (AUC) scores, with a minimum score of 70.0 across twenty-eight organisms. Further, importance of the codons towards classifying genes found to differ among the codons in each genome. Arg codon CGT and Gly codon GGT were observed to be the most preferred codons among essential genes in Escherichia coli. Interestingly, some of the codons like CGT, ATA, GGT and GGG observed to be contributing consistently towards classifying essential genes across thirty-five bacteria genomes studied. In other hand, codons TGY and CAY encoding amino acids Cys and His respectively were among the least contributing codons towards classification among all these bacteria. This study demonstrates the gene essentiality based differences in synonymous codon usage in bacteria genomes and presents a common codon usage pattern across bacteria.

Spectral histology of hair and hair follicle using infrared microspectroscopy.

Today, there is only limited knowledge of the spatial organization of hair chemistry. Infrared microspectroscopy is a well-established tool to provide such information and has significantly contributed to this field. In this study, we present new results combining multiple infrared microspectroscopy methods at different length scales to create a better chemical histology of human hair, including the hair follicle, hair shaft, hair medulla and hair cuticle. We used hyperspectral IR imaging & spectroscopy (HIRIS) and synchrotron-radiation FTIR microspectroscopy (SR-μFTIR) to measure transversal hair sections and SR-μFTIR to obtain high-resolution maps of longitudinal sections from the hair shaft and from the hair follicle. We used optical photothermal IR microspectroscopy (OPTIR) to analyse the cuticle surface of intact hairs. By mapping longitudinal sections of the human hair follicle with confocal SR-μFTIR, we report the first demonstration of glycogen presence in the outer root sheath of the hair follicle by spectroscopy, and its quantification at the micron scale. Spectral maps, combined with machine learning-based analysis, enabled us to differentiate the various layers of the hair follicle and provided insights into the chemical changes that occur during hair formation in the follicle. Using HIRIS and SR-μFTIR to analyse the hair medulla in transversal sections of human hairs, we report here, for the first time by vibrational spectroscopy methods, the detection of unsaturated lipids at very low concentrations in the medulla. By analysing longitudinal sections of the hair shaft with SR-μFTIR, we found that calcium carboxylates are present in large regions of the hair cuticle, and not just in small focal areas as previously thought. We then use OPTIR to analyse the hair cuticle of intact hairs at submicron resolution without sectioning and report the distribution of calcium carboxylates at the surface of intact hair for the first time. These new findings illustrate the potential of infrared microspectroscopy for imaging the chemical composition of human hair and may have implications for biomedical research or cosmetology.

Machine learning-enhanced HRCT analysis for diagnosis and severity assessment in pediatric asthma.

Chest high-resolution computed tomography (HRCT) is conditionally recommended to rule out conditions that mimic or coexist with severe asthma in children. However, it may provide valuable insights into identifying structural airway changes in pediatric patients. This study aims to develop a machine learning-based chest HRCT image analysis model to aid pediatric pulmonologists in identifying features of severe asthma. This retrospective case-control study compared children with severe asthma (as defined by ERS/ATS guidelines) to age- and sex-matched controls without asthma, using chest HRCT scans for detailed imaging analysis. Statistical analysis included classification trees, random forests, and conventional ROC analysis to identify the most significant imaging features that mark severe asthma from controls. Chest HRCT scans differentiated children with severe asthma from controls. Compared to controls (n = 21, mean age 11.4 years), children with severe asthma (n = 20, mean age 10.4 years) showed significantly greater bronchial thickening (BT) scores (p < 0.001), airway wall thickness percentage (AWT%, p < 0.001), bronchiectasis grading (BG) and bronchiectasis severity (BS) scores (p = 0.016), mucus plugging, and centrilobular emphysema (p = 0.009). Using AWT% as the predictor in conventional ROC analysis, an AWT% ≥ 38.6 emerged as the optimal classifier for discriminating severe asthmatics from controls, with 95% sensitivity, specificity, and overall accuracy. Our study demonstrates the potential of machine learning-based analysis of chest HRCT scans to accurately identify features associated with severe asthma in children, enhancing diagnostic evaluation and contributing to the development of more targeted treatment approaches.

Machine learning-based analysis and prediction of meteorological factors and urban heatstroke diseases.

Heatstroke is a serious clinical condition caused by exposure to high temperature and high humidity environment, which leads to a rapid increase of the core temperature of the body to more than 40°C, accompanied by skin burning, consciousness disorders and other organ system damage. This study aims to analyze the effect of meteorological factors on the incidence of heatstroke using machine learning, and to construct a heatstroke forecasting model to provide reference for heatstroke prevention. The data of heatstroke incidence and meteorological factors in a city in South China from May to September 2014-2019 were analyzed in this study. The lagged effect of meteorological factors on heatstroke incidence was analyzed based on the distributed lag non-linear model, and the prediction model was constructed by using regression decision tree, random forest, gradient boosting trees, linear SVRs, LSTMs, and ARIMA algorithm. The cumulative lagged effect found that heat index, dew-point temperature, daily maximum temperature and relative humidity had the greatest influence on heatstroke. When the heat index, dew-point temperature, and daily maximum temperature exceeded certain thresholds, the risk of heatstroke was significantly increased on the same day and within the following 5 days. The lagged effect of relative humidity on the occurrence of heatstroke was different with the change of relative humidity, and both excessively high and low environmental humidity levels exhibited a longer lagged effect on the occurrence of heatstroke. With regard to the prediction model, random forest model had the best performance of 5.28 on RMSE and dropped to 3.77 after being adjusted. The incidence of heatstroke in this city is significantly correlated with heat index, heatwave, dew-point temperature, air temperature and zhongfu, among which the heat index and dew-point temperature have a significant lagged effect on heatstroke incidence. Relevant departments need to closely monitor the data of the correlated factors, and adopt heat prevention measures before the temperature peaks, calling on citizens to reduce outdoor activities.

Differential Serum Peptidomics Reveal Multi-Marker Models That Predict Breast Cancer Progression.

Here, we assess how the differential expression of low molecular weight serum peptides might predict breast cancer progression with high confidence. We apply an LC/MS-MS-based, unbiased 'omics' analysis of serum samples from breast cancer patients to identify molecules that are differentially expressed in stage I and III breast cancer. Results were generated using standard and machine learning-based analytical workflows. With standard workflow, a discovery study yielded 65 circulating biomarker candidates with statistically significant differential expression. A second study confirmed the differential expression of a subset of these markers. Models based on combinations of multiple biomarkers were generated using an exploratory algorithm designed to generate greater diagnostic power and accuracy than any individual markers. Individual biomarkers and the more complex multi-marker models were then tested in a blinded validation study. The multi-marker models retained their predictive power in the validation study, the best of which attained an AUC of 0.84, with a sensitivity of 43% and a specificity of 88%. One of the markers with m/z 761.38, which was downregulated, was identified as a fibrinogen alpha chain. Machine learning-based analysis yielded a classifier that correctly categorizes every subject in the study and demonstrates parameter constraints required for high confidence in classifier output. These results suggest that serum peptide biomarker models could be optimized to assess breast cancer stage in a clinical setting.

Toward automated microstructure characterization of stainless steels through machine learning-based analysis of replication micrographs

Machine learning-driven automated replication micrographs analysis makes possible rapid and unbiased damage assessment of in-service steel components. Although micrographs captured by scanning electron microscopy (SEM) have been analyzed at depth using machine learning, there is no literature available on the technique being attempted on optical replication micrographs. This paper presents a machine-learning approach to segment and quantify carbide precipitates in thermally exposed HP40-Nb stainless-steel microstructures from batches of low-resolution optical images obtained by replication metallography. A dataset of nine micrographs was used to develop a random forest classification model to segment precipitates within the matrix (intragranular) and at grain boundaries (intergranular). The micrographs were preprocessed using background subtraction, denoising, and sharpening to improve quality. The method achieves high segmentation accuracy (91% intergranular, 97% intragranular) compared to human expert classification. Furthermore, segmented micrographs were quantified to obtain carbide size, shape, and density distribution. The correlations in the quantified data aligned with expected carbide evolution mechanisms. Results from this study are promising but necessitate validation of the method on a larger dataset representative of evolution of thermal degradation in steel, given that characterization of the evolution of microstructure components, such as precipitates, applies to broad applications across diverse alloy systems, particularly in extreme service.

Are analysts’ Forecasts Reliable? A Machine Learning-Based Analysis of the Target Price Accuracy

This paper examines the accuracy of target price forecasts made by sell-side analysts, focusing on predicting target price accuracy using machine learning approaches. Utilizing a dataset of target price forecasts for U.S. listed companies from 1999 to 2021, we employ ensemble methods and incorporate market-level, firm-level, and analyst-level information to predict target price accuracy in terms of target price errors and target price achievement. The long-short portfolio constructed based on our predictions significantly outperform the benchmark in terms of cumulative return and Sharpe ratio.

Harmonizing heterogeneous transcriptomics datasets for machine learning-based analysis to identify spaceflown murine liver-specific changes

NASA has employed high-throughput molecular assays to identify sub-cellular changes impacting human physiology during spaceflight. Machine learning (ML) methods hold the promise to improve our ability to identify important signals within highly dimensional molecular data. However, the inherent limitation of study subject numbers within a spaceflight mission minimizes the utility of ML approaches. To overcome the sample power limitations, data from multiple spaceflight missions must be aggregated while appropriately addressing intra- and inter-study variabilities. Here we describe an approach to log transform, scale and normalize data from six heterogeneous, mouse liver-derived transcriptomics datasets (ntotal = 137) which enabled ML-methods to classify spaceflown vs. ground control animals (AUC ≥ 0.87) while mitigating the variability from mission-of-origin. Concordance was found between liver-specific biological processes identified from harmonized ML-based analysis and study-by-study classical omics analysis. This work demonstrates the feasibility of applying ML methods on integrated, heterogeneous datasets of small sample size.

A retrospective machine learning–based analysis of nationwide cancer comprehensive genomic profiling data to identify features associated with recommendation of mutation-based therapy.

e13510 Background: Comprehensive genomic profiling (CGP) has played key roles in cancer precision medicine through optimization of therapeutic interventions based on genomic alterations in cancer cells. However, the probability of discovering mutation-based treatments through CGP remains low. To enhance the effectiveness and efficiency of cancer precision medicine, it is crucial to identify patients who are likely to benefit from CGP tests. This study aims to identify characteristics of patients in which mutation-based treatments are discovered by CGP tests. Methods: We retrospectively analyzed data from 60,655 patients who underwent CGP tests and were registered in the Center for Cancer Genomics and Advanced Therapeutics (C-CAT), the national data center for cancer CGP in Japan. The C-CAT database covers 99.7% of cancer patients who have undergone CGP tests in Japan. Major cancer types included 10,182 cases of bowel cancer, 8,691 pancreas cancer, 5,062 biliary tract cancer, and 3,777 breast cancer. We developed an eXtreme Gradient Boosting (XGBoost) model using machine learning, and used clinical information as input to predict whether one or more Japanese Pharmaceuticals and Medical Devices Agency (PMDA)-approved drugs are discovered through CGP tests or not. Shapley Additive Explanations (SHAP) was employed to extract significant features that contribute to the model prediction. Results: The prediction model achieved an area under the receiver operating characteristic curve of 0.826 for the overall cancer population. Positive SHAP values were observed for patients with breast (mean SHAP in breast cancer patients: 1.38), lung (1.08), bowel (0.75), and pancreas cancers (0.35), while negative SHAP values were associated with head and neck (-1.75), cervical cancers (-1.54), and brain (-1.33). Positive SHAP values were also associated with presence of liver or lymph node metastasis (0.23, 0.08), shorter intervals between diagnosis and CGP testing or specimen collection and CGP testing, and advanced age. Similar trends were observed in cancer-type-specific prediction models, which also identified their own unique features. In the adolescent and young adult (AYA) age group, primary brain tumors were strongly associated with negative SHAP values (-1.37). Conclusions: Our machine learning-based analysis of nationwide CGP data identified features that predict cases in which mutation-based treatments are discovered by CGP tests, both in the overall cancer population and within specific cancer types and the AYA age group. Expedited CGP testing is recommended for patients who match the identified profile to facilitate early targeted therapy interventions.

Protein Adsorption on Solid Surfaces: Data Mining, Database, Molecular Surface-Derived Properties, and Semiempirical Relationships.

Protein adsorption on solid surfaces is a process relevant to biological, medical, industrial, and environmental applications. Despite this wide interest and advancement in measurement techniques, the complexity of protein adsorption has frustrated its accurate prediction. To address this challenge, here, data regarding protein adsorption reported in the last four decades was collected, checked for completeness and correctness, organized, and archived in an upgraded, freely accessible Biomolecular Adsorption Database, which is equivalent to a large-scale, ad hoc, crowd-sourced multifactorial experiment. The shape and physicochemical properties of the proteins present in the database were quantified on their molecular surfaces using an in-house program (ProMS) operating as an add-on to the PyMol software. Machine learning-based analysis indicated that protein adsorption on hydrophobic and hydrophilic surfaces is modulated by different sets of operational, structural, and molecular surface-based physicochemical parameters. Separately, the adsorption data regarding four "benchmark" proteins, i.e., lysozyme, albumin, IgG, and fibrinogen, was processed by piecewise linear regression with the protein monolayer acting as breakpoint, using the linearization of the Langmuir isotherm formalism, resulting in semiempirical relationships predicting protein adsorption. These relationships, derived separately for hydrophilic and hydrophobic surfaces, described well the protein concentration on the surface as a function of the protein concentration in solution, adsorbing surface contact angle, ionic strength, pH, and temperature of the carrying fluid, and the difference between pH and the isoelectric point of the protein. When applying the semiempirical relationships derived for benchmark proteins to two other "test" proteins with known PDB structure, i.e., β-lactoglobulin and α-lactalbumin, the errors of this extrapolation were found to be in a linear relationship with the dissimilarity between the benchmark and the test proteins. The work presented here can be used for the estimation of operational parameters modulating protein adsorption for various applications such as diagnostic devices, pharmaceuticals, biomaterials, or the food industry.

Gradient boosted regression as a tool to reveal key drivers of temporal dynamics in a synthetic yeast community.

Microbial communities are vital to our lives, yet their ecological functioning and dynamics remain poorly understood. This understanding is crucial for assessing threats to these systems and leveraging their biotechnological applications. Given that temporal dynamics are linked to community functioning, this study investigated the drivers of community succession in the wine yeast community. We experimentally generated population dynamics data and used it to create an interpretable model with a gradient-boosted regression tree approach. The model was trained on temporal data of viable species populations in various combinations, including pairs, triplets, and quadruplets, and was evaluated for predictive accuracy and input feature importance. Key findings revealed that the inoculation dosage of non-Saccharomyces species significantly influences their performance in mixed cultures, while Saccharomyces cerevisiae consistently dominates regardless of initial abundance. Additionally, we observed multispecies interactions where the dynamics of Wickerhamomyces anomalus were influenced by Torulaspora delbrueckii in pairwise cultures, but this interaction was altered by the inclusion of S. cerevisiae. This study provides insights into yeast community succession and offers valuable machine learning-based analysis techniques applicable to other microbial communities, opening new avenues for harnessing microbial communities.