Due to delayed symptoms and dependence of behavioral assessment, early diagnosis of autism spectrum disorder remains challenging. Identification of multivariate biomarker for the etiological mechanisms of ASD may enhance diagnostic accuracy. Multivariable logistic regression combines many predictors into a single risk score (linear predictor), resulting in an optimised ROC curve that enhances diagnostic accuracy over individual markers. The method comprises modelling a binary result, determining the likelihood, and visualising ROC based on the projected probabilities, which often improves individual marker AUCs. In the present study a diagnostic performance for a biomarker panel reflecting glutamatergic dysfunction, oxidative stress, and neuroinflammation was evaluated. Plasma levels of glutaminase, 8-isoprostane, and prostaglandin E₂ (PGE₂) obtained from 44 children with ASD and 40 age-matched controls were evaluated using receiver operating characteristic (ROC) analysis, both individually and in combined ROC models. Glutaminase showed significant negative correlations with both 8-isoprostane and PGE₂, whereas a positive correlation was observed between 8-isoprostane and PGE₂. All the three-biomarker showed good diagnostic performance for ASD on its own with statistically significant (p = 0.001) values of AUC of 0.830 for glutaminase, AUC of 0.815 for 8-Isoprostane and AUC of 0.818 for PGE₂. However combined ROC modeling substantially improved diagnostic accuracy by achieving high apparent discriminative performance with AUC value of 0.977 with 92.3% sensitivity and 100.0% specificity. In conclusion, the diagnostic usefulness of independent glutaminase, 8-isoprostane, and prostaglandin E₂ (PGE₂) biomarkers may be enhanced by combining ROC. Combined markers show strong apparent discriminating power in a case-control method, but estimates are biassed towards optimism and are not diagnostic. Comprehensive assay validation, calibration, and clinically representative cohorts (including females and relevant differentials) are required for replication.

Feed efficiency is an economically important but costly trait to measure in pig breeding. Previous studies have shown that integrating metabolomic data, such as proton nuclear magnetic resonance (¹H NMR) - derived metabolomic profiles, into genomic prediction models can improve the accuracy of estimated breeding values (EBVs) - for example, using a univariate metabolomic-genomic best linear unbiased prediction (MGBLUP) model for malting quality traits in barley and for average daily gain (ADG) in pigs using NMR-based metabolomic features (MFs). In this study, we extend this approach to predict feed conversion ratio (FCR) in pigs. We tested two hypotheses: (1) incorporating NMR metabolomic data into a univariate MGBLUP model increases the accuracy of EBVs for FCR compared with a univariate genomic BLUP (GBLUP) model, and (2) a bivariate MGBLUP model that jointly analyses FCR and the correlated trait ADG further improves EBV accuracy compared with a univariate MGBLUP model. We tested these hypotheses using an offspring-validation design, allowing prediction of EBVs for animals lacking individual FCR records. The experimental population comprised 8,174 Duroc pigs (4,027 males from a test station and 4,147 females from breeding herds). To evaluate the accuracy of EBVs for FCR, males with FCR records were used as the training population, and females without FCR records served as the validation population. These validation females had offspring with recorded FCR, and EBV accuracy was assessed by correlating their EBVs with the corrected phenotypes of their offspring. Incorporating metabolomic data into the univariate MGBLUP model generated EBVs for FCR that were 2.3% more accurate than EBVs from the univariate GBLUP model (0.394 vs. 0.385). The bivariate MGBLUP model that jointly analysed FCR and ADG generated EBVs for FCR that were 1.2% more accurate than EBVs from the bivariate GBLUP model (0.432 vs. 0.427). The increases in accuracy were modest and statistically insignificant, reflecting limitations in the current metabolomic data, such as single-time-point sampling. Even so, the observed improvements suggest that metabolomic information may provide complementary information beyond genomic data. With further refinements in sampling strategies and genetic models, metabolomics could contribute to improving prediction accuracy in animal breeding.

Rapid genetic gain was achieved for cooking time in common bean based on multivariate genomic analysis, but forward predictive ability was low and phenotyping remains essential to secure genetic gain. Common beans (Phaseolus vulgaris L.) are a major source of protein and energy in sub-Saharan Africa, but their long cooking time (CKT) imposes social, economic, environmental and health burdens. This study aimed to accelerate genetic gain for shorter CKT while maintaining or improving other seed traits such as seed iron (Fe) and zinc (Zn) content, water absorption capacity (WAC) and 100-seed weight (SW100) across rapid cycles of early-generation genomic selection. Two related founder populations were selected from the African bean panel and intercrossed in 2020 (population A) and 2021 (population B), followed by rapid two-year cycles of augmented S0-derived family selection based on an index and optimal contributions selection. Best linear unbiased predictions (BLUPs) of breeding values were obtained from pedigree(ABLUP), genomic(GBLUP) and single-step (HBLUP) multivariate linear mixed model analysis across two cycles. Realised genetic gain from cycle 1 to cycle 2 was high for CKT (average -8.0miny-1) and favourably correlated with WAC (average + 7.7%y-1), but detrimental genetic correlations limited gain in Fe, Zn and SW100. Genomic and single-step models resulted in high accuracy of BLUPs based on prediction error variance. Forward predictive ability for CKT across cycles was low, but rank correlationof realised and predicted HBLUPs forCKT in cycle 2B S0seedlings was moderate-high (0.734)when phenotypes for Fe, Zn, WAC and SW100 in cycle 2B were included in the analysis. Desired genetic gains in CKT, Fe and Zn in future cycles will require high levels of phenotyping in each cycle, although easy-to-measure correlated traits such as WAC and SW100 may assist genetic gain in all traits.

Genomic prediction (GP) has become an essential tool for accelerating modern plant breeding, particularly for complex traits. We evaluated different GP approaches for pre-breeding in thirteen biparental strawberry families derived from crosses between Fragaria virginiana and Fragaria chiloensis, focusing on resistance to powdery mildew (PM) in both leaves and fruits. Five-fold cross-validation using Genomic Best Linear Unbiased Prediction (GBLUP) for combined-year data yielded mean predictive abilities (PAs) of 0.56 and 0.36 for leaf and fruit resistance, respectively. Family-based GBLUP analyses showed higher PAs for closely related families, ranging from 0.10 to 0.89. Simulations identified key parameters: training sets comprising 40% of the population provided stable predictions for both traits, while ≈8300 SNPs were sufficient for predicting leaf resistance. However, PA for fruit resistance remained consistently low regardless of marker density. We then compared GBLUP with a marker-assisted model that iteratively incorporated the major resistance loci as fixed effects. This strategy increased PA by ≈10-30% for leaf resistance and ≈25-47% for fruit resistance across models. We further applied cross-environment forward prediction in independent greenhouse and separate field validation trials. The GBLUP model maintained substantial PA when trained on either the full population or a 40% subset, demonstrating robustness in predicting new genotypes across distinct environments. Our findings highlight the importance of taking trait genetic architecture into account to enhance PAs for PM resistance in strawberry. Together, these results provide evidence-based thresholds for training population size and marker density, offering a framework for efficient implementation of genomic selection in strawberry pre-breeding populations.

The potential of social chatbots show significant potential in advancing healthy aging, yet their effectiveness is limited by trust deficits among elderly users. Integrating multidimensional trust with the Stimuli-Organism-Response framework, this study examines how perceived chatbot characteristics shape cognitive trust (CT), emotional trust (ET), behavioral trust (BT), and usage behavior. Data from 292 older Chinese social chatbot users were analyzed using PLS-SEM and fsQCA. Results show that accuracy and transparency significantly enhance CT, warmth promotes ET, and CT and ET sequentially strengthen BT, which drives usage behavior. Anthropomorphism directly increases BT but does not significantly affect ET. Total effects identify anthropomorphism, ET, and CT as the strongest linear predictors of BT. FsQCA reveals three equifinal high-trust configurations: a cognitive-transparency-driven path, an emotional-centric synergy path, and a comprehensive compensation path. These findings clarify both net-effect and configurational mechanisms of trust formation and inform trustworthy chatbot design for older adults.

Inhalation of crystalline silica dust causes lung cancer, although evidence assessing risk at low levels of exposure is needed. We sought to estimate the association between cumulative silica exposure and lung cancer risk in a cohort of miners in Ontario, Canada. A cohort of 48 772 hard rock miners enumerated from a medical surveillance programme was followed for lung cancer diagnoses in the Ontario Cancer Registry from 1964 through 2022. Silica dust exposure was estimated using a linear prediction model based on 12 325 personal sampling measurements from Ontario mines. Exposure-response associations were estimated, with adjustments for uranium mining, arsenic dust and screening CXR. Probabilistic bias analysis was used to evaluate differential outcome misclassification due to loss to follow-up using a known bias parameter. 3218 lung cancer cases were diagnosed during a median of 44 years of follow-up. Positive associations between cumulative silica exposure and incident lung cancer rates were observed in all categories of exposure, with 0.5-2.0 mg/m3-years of exposure associated with 1.33 times (95% CI 1.06 to 1.66) the adjusted rate of lung cancer compared with<0.5 mg/m3-years. The positive exposure-response association remained after excluding known silicotics and increased in magnitude when excluding radon-exposed uranium miners. Bias adjustment for outcome misclassification increased the monotonicity and slope of the exposure-response relationship, but attenuated the IRR in the 0.5-2.0 mg/m3-year category to 1.08. Silica exposure was positively associated with lung cancer risk at low levels of cumulative exposure in a universally exposed population of miners. Findings were robust to suspected sources of bias.

As one of the complications of liver cirrhosis, portal vein thrombosis (PVT) brings a significant clinical challenge. To find the predictive clinical variables evaluating the probability of PVT, a prediction model was established and a nomogram was built for visualization. Our study selected 245 cirrhosis patients admitted in First Hospital of Jilin University during August 2020 to June 2024, the majority of whom were admitted during July 2022 to June 2023. All patients were randomly separated into a training set and a validation set. Binary logistic and Lasso regression were used to establish predictive models with the occurrence of PVT events within 24months after discharge as the clinical outcome. In discovery set, 64 (26.1%) of them had events of PVT within 24months after their discharge. A linear model "ASP" was constructed with logarithm of alkaline phosphatase (Log10ALP), splenectomy and portal vein width selected for predicting PVT risk, preformed with its AUC reached 0.793, 0.750 and 0.760 in training, internal validation and external validation set, respectively. We finally screened out three clinical variables identifying high-risk PVT patients in liver cirrhosis beforehand. ALP level might be a potential clinical parameter connected to PVT, but need to be further verified.

BackgroundGenomic prediction is widely used in pig breeding, but phenotypic prediction of complex traits such as disease resilience remains limited because genotypes alone do not capture infection-induced regulatory responses, environmental and management effects, or their interactions. Blood molecular profiles measured in young healthy pigs reflect both genetic and non-genetic influences and may improve prediction of performance under disease challenge. We evaluated whether integrating multiple blood-based omics layers with genomic data improves prediction of production and disease resilience phenotypes in pigs exposed to a polymicrobial disease challenge.ResultsData were from 836 healthy pigs from 15 batches with transcriptomic, proteomic, and metabolomic profiles measured in blood collected at ~27 days of age, before transfer into a natural polymicrobial disease challenge at ~40 days of age. Pigs were also genotyped using a commercial 650 K marker array. We analyzed 21 traits related to growth, health scores, antibiotic treatments, mortality, feed efficiency, and carcass traits using best linear unbiased prediction (BLUP) animal models with random animal effects based on relationship matrices constructed from genomic (G), transcriptomic (T), proteomic (P), and metabolomic (M) data. Across traits, G-BLUP explained the largest proportion of phenotypic variance for most traits. However, T-, P-, or M-BLUP explained similar or greater variance than G-BLUP for several growth and health traits recorded before challenge. Adding T and/or M to G-BLUP generally increased variance explained and improved prediction accuracy for pre-challenge growth rate and health scores, and for mortality and carcass weight after challenge. Models combining G, T, and M often yielded the highest accuracies, whereas adding P did not consistently improve accuracy. For later grow-finish traits, gains from multi-omics were smaller and less consistent.ConclusionsBlood multi-omics profiles from healthy young pigs can improve prediction of performance and disease resilience beyond genomic data alone. Gains were greatest for traits recorded before challenge and for some resilience traits expressed soon after pathogen exposure, suggesting that pre-challenge molecular profiles capture latent resilience potential. These findings support the use of pre-challenge blood multi-omics as biomarkers for precision management and as a basis for breeding and management strategies targeting disease resilience in pigs.Supplementary InformationThe online version contains supplementary material available at 10.1186/s40104-026-01416-9.

Improving end-use quality in bread wheat (Triticum aestivum) requires dissecting the genetic basis of complex processing traits and deploying robust prediction pipelines in breeding. We performed genome-wide association studies (GWASs) using 1767 high-quality single-nucleotide polymorphisms generated by genotyping-by-sequencing in a diverse Canada Western Red Spring panel phenotyped near Swift Current, SK, from 2009 to 2019 for grain protein content, milling yield, mixing energy, water absorption, and doughextensibility. The analysis detected significant marker-trait associations on 13 chromosomes, recovering signals at Rht-B1 and Glu-1 and revealing multiple additional signals that may represent previously unreported loci in this germplasm and environmental context, consistent with polygenic control. We then evaluated genomic selection using GBLUP (Genomic Best Linear Unbiased Predictor) and BayesB with and without including significant GWAS hits as fixed effects; gains in predictive accuracy were generally negligible, although water absorption showed modest improvement, compatible with fewer, larger effect loci. Functional annotation of genes near associated variants implicated stress responses, protein metabolism, and grainfilling. Together, these results refine the genetic architecture of Canadian wheat quality and support integrating GWAS-informed biology with genome-wide prediction to accelerate quality-by-design breeding.

To evaluate the relationship between baseline axial length (AL) and the rate of myopia progression in children and determine whether baseline AL alone predicts rapid myopia progression. This retrospective study included 1,458 patients (<20 years old) who underwent cycloplegic refraction and biometry for at least 2 years between 2011 and 2024. Myopia progression rate was assessed using AL elongation (mm/yr), spherical equivalent (SE) change (diopters/yr), and AL/K radius (AL/corneal radius per year). Partial correlation analysis and multiple linear regression were performed to assess linearity between AL and myopia progression rate. A total of 2,916 eyes were analyzed. Baseline AL exhibited weak partial correlations with progression indicators. After adjusting for age, partial correlation coefficients for the right and left eyes were 0.297 and 0.305 for AL elongation, -0.267 and -0.278 for SE change, and 0.259 and 0.269 for AL/K radius rate, respectively. Multiple regression analyses, adjusting for age and K radius effect revealed that the linear model for the right and left eyes accounted for only 2.4% and 2.0% for AL elongation, 10.0% and 10.5% of SE change, and 1.5% and 1.3% of AL/K radius rate, respectively. Analysis using generalized estimating equations to account for inter-eye correlation revealed that AL had a minimal impact on myopic progression rates. Progression rates decreased with baseline AL >24 mm, suggesting a nonproportional relationship between AL and progression rate of myopia. Baseline AL was not a linear independent predictor of rapid myopia progression. Myopia progression tended to be slow in patients with an AL >24 mm.

Data-driven online SOH estimation of a grid-connected BESS: Accuracy improvements and lifetime prediction

Study of survival of Holstein cattle of Iran using random regression and single-step BLUP.

Anthracnose, caused by Colletotrichum dematium, has emerged as a major foliar disease that threatens spinach (Spinacia oleracea L.) production. In this study, a diverse panel of 266 accessions was evaluated under field conditions to dissect the genetic architecture of anthracnose resistance. Substantial phenotypic variation was observed, with disease severity indices (DSIs) ranging from 1 (highly resistant) to 10 (highly susceptible). A total of 20 accessions showed moderate to high resistance (DSI ≤ 4), representing valuable resistance sources. A genome-wide association study (GWAS) identified 20 significant single-nucleotide polymorphisms (SNPs). The most consistent marker (SOVchr3_19667279) explained up to 66.8% of phenotypic variance, and several associated SNPs were located within or near putative defense-related genes. Genomic prediction (GP) using multiple models demonstrated that predictive accuracy increased when a set with more SNPs was used. The genomic best linear unbiased prediction model achieved the highest accuracy (r=0.92), while the Bayesian ridge regression model attained the best predictive accuracy (r=0.51) when using 20 GWAS-derived SNPs, highlighting the value of integrating association mapping with prediction approaches. Cross-population prediction performed well (r=0.58-0.71), whereas across-population prediction showed reduced accuracy (r=0.10-0.14), indicating the influence of genetic background on model transferability. This integrative genomic study provides novel insights into the genetic basis of anthracnose resistance in spinach and demonstrates the potential of GWAS and GP in accelerating resistance breeding.

Wheat (Triticum aestivum L.), a crucial cereal crop for global food security, faces growing challenges from climate change. Future production requires varieties that are resilient to environmental extremes and fluctuations. The goal of this study was to assess strategies to increase selection response through genomic selection in wheat by integrating genotypic-specific phenology-derived environmental covariates (ECs) and random regression models (RRM) in multi-environment trials. We analyzed phenotypic and genomic data from 1683 genotypes from 2010 to 2020 across 71 environments using 45 ECs derived from vegetative, reproductive, and grain-filling phenological phases. Seven key ECs were selected via partial least squares regression to model genotype by environment interaction (GEI) and evaluate their integration in three different genomic prediction scenarios (CV0, CV1, and CV2). Genomic best linear unbiased prediction models (GBLUP), GBLUP models with GEI (GBLUPG × E) modeled as a factor analytic (FA) model, and RRM were compared for their predictive ability performance. RRM with three ECs outperformed GBLUP achieving 50%-100% higher accuracy in CV1 and CV2. The FA exhibited the highest accuracy overall for CV2 but not for CV1. At least one RRM model improved predictions in >89% of environments when predicting new, un-phenotyped environments. Integrating ECs into the RRM enhances genomic prediction by effectively capturing the GEI with a limited number of covariates.

A nonparametric framework for linear–circular regression: Applications in environmental and biological sciences

The Reflections series takes a look back on historical articles from The Journal of the Acoustical Society of America that have had a significant impact on the science and practice of acoustics.

The evaluation of genotype-by-environment interaction (GEI) through multi-environment trials (METs) is an essential prerequisite in breeding improvement programs targeting wide adaptation. In this study, a panel of newly developed barley genotypes was examined under field conditions at five experimental sites across Iran's warm climatic zone during the 2023-2025 growing seasons. The pooled analysis of variance clearly demonstrated that effects of genotype (G), environment (E), and GEI influenced all measured traits, including the number of days to heading (DH), days to physiological maturity (DM), grain filling period (GFP), plant height (PH), 1000-kernel weight (TKW), and grain yield (GY). Analytical outputs obtained from the additive main effects and multiplicative interaction (AMMI) model, as well as the best linear unbiased prediction (BLUP) approach, highlighted the substantial contribution of genotype × environment interaction to variation in grain yield. Insights from genotype-trait (GT) biplot analysis indicated that TKW and GFP were positively associated with GY. When genotypes were ranked using the multi-trait stability index (MTSI), G5 (1.689), G4 (1.914), G20 (2.380), and G14 (2.508) achieved the highest scores, reflecting superior performance and stability across test environments. A comprehensive evaluation strategy integrating classical AMMI, BLUP, Bayesian AMMI, and genotype-by-environment (GGE) biplot methodologies was employed to identify genotypes that combine productivity with stability. Across all derived stability parameters, including the weighted average of absolute scores (WAAS) and its yield-weighted counterpart (WAASY), genotype G4, derived from the pedigree [Sahra/3/Bda/Rhn-03//ICB-107766], consistently outperformed other candidates. This genotype exhibited a robust yield response, high stability, and broad environmental responsiveness. Independent validation using the stability Mahalanobis distance (SM) index and Y × WAASY biplot visualization further substantiated these findings. In conclusion, genotype G4 represents a strong candidate for subsequent validation trials and potential varietal release in the warm agroecological regions of Iran and other areas with comparable climatic conditions.

BackgroundGenomic prediction (GP) is a central component of modern plant breeding, enabling the early selection of superior genotypes based on genomic marker data. Classical GP models, such as genomic best linear unbiased prediction (GBLUP), operate within the data modeling culture and typically assume additive genetic effects, with extensions required to model non-additive effects such as dominance and epistasis. In contrast, machine learning (ML) models from the algorithmic modeling culture can flexibly model complex, non-additive genetic relationship but often lack direct grounding in quantitative genetic theory and interpretability. To bridge these gaps, we propose 2NPLGBM, a hybrid genomic prediction approach that integrates quantitative genetics with ML. This method introduces a two-matrix (2NP) genotype representation by concatenating additive (Z) and dominance (W) matrices, which are then used as input to a Light Gradient Boosting Machine (LGBM), enabling the simultaneous modeling of additive, dominance, and higher-order genetic interactions (AA, AD, DD).ResultsThe 2NPLGBM model was evaluated using six years of test-cross hybrid maize trial data across four agronomic traits (grain yield, plant height, days to silking, and days to anthesis) under five cross-validation schemes simulating temporal: Leave-One-Year-Out (LOYO), Rolling Window (RW), and genetic generalization: Five-Fold, and tester-based schemes (Tester CV0 and Tester CV00). Compared to GBLUP, 2NPLGBM achieved an average of 5% improvement in predictive accuracy under temporal validations and over 15% gains under tester-based schemes, particularly for flowering traits (days to silking and days to anthesis). Performance was generally comparable to LGBM, with both ML models outperforming GBLUP for most traits. Under Tester CV0, 2NPLGBM showed its strongest relative advantage over LGBM for flowering traits, suggesting improved capture of interaction-related genetic signals, whereas LGBM generally performed best for plant height and grain yield. In five-fold CV and Tester CV00, GBLUP remained competitive for some traits, while both machine learning models showed reduced gains, with LGBM slightly outperforming 2NPLGBM. In addition, 2NPLGBM generally improved selection efficiency over GBLUP and, in most cases, LGBM, indicating enhanced ability to capture complex genetic signals relevant for hybrid ranking, particularly for flowering traits, whereas LGBM tended to achieve the highest selection efficiency for plant height and grain yield. Feature interpretation using Shapley Additive exPlanations (SHAP) confirmed that non-additive interactions contributed substantially to prediction accuracy for highly heritable traits. It also revealed trait-specific architectures, additive effects dominated flowering traits, while dominance effects contributed more to plant height and yield. Classical variance component analysis supported these findings, indicating high dominance contributions of 17.3% for yield and 8.2% for plant height.Conclusion The 2NPLGBM model integrates quantitative genetic theory with machine-learning, bridging classical statistical (data-model) and algorithmic modeling cultures. bridging classical statistical (data model) and algorithmic modeling cultures. By jointly modeling additive and non-additive effects it can improve predictive accuracy, interpretability, and selection efficiency in test-cross hybrids. Future work should explore multi-trait and multi-environment extensions, integration of environmental covariates, and the inclusion of multi-omic data to further strengthen predictive power and interpretability.Supplementary InformationThe online version contains supplementary material available at 10.1186/s13007-026-01545-2.

The growing adoption of crossbred populations by dairy farmers has sparked increased interest in multibreed genomic evaluation, which offers promising opportunities to accelerate genetic progress and refine breeding strategies. This study assessed the feasibility of applying single-step genomic best linear unbiased prediction (ssGBLUP) for multibreed genomic evaluation of cow wellness traits in Holstein (HO), Jersey (JE) and their crosses (HO × JE). The traits considered were abortion (ABRT), cystic ovaries (CYST), displaced abomasum (DA), ketosis (KETO), lameness (LAME), mastitis (MAST), metritis (METR), milk fever (MFEV), respiratory illness (RESP), retained placenta (RETP) and twinning (TWIN). The number of phenotypic records ranged from 1,176,935 for RESP to 7,703,872 for MAST. Traits were analysed using a linear model within a multi-trait framework. Variance components were estimated using Gibbs sampling, incorporating the effects of inbreeding, retained heterosis, trait-specific systematic effects and random effects of additive direct and permanent environmental effects. The algorithm for Proven and Young was used by randomly selecting a core set of 30,000 animals. Two scenarios were developed based on the number of genotyped HO animals in the evaluation: (1) including all available genotyped individuals, and (2) including only a subset of relevant genotyped HOs. Genomic predictions were compared against commercially available single-breed evaluations to assess consistency and potential improvements in predictive performance. Heritability estimates ranged from 0.003 for MFEV and CYST to 0.057 for MAST. Spearman and Pearson correlations between multibreed and single-breed evaluations for cow wellness ranged from 0.36 and 0.33 (METR) to 0.88 and 0.89 (MAST) for JE, and from 0.79 (MFEV) to 0.97 and 0.96 (METR) for HO. These findings suggest that a single-step approach can produce comparable results and accurate genomic predictions, demonstrating the feasibility of ssGBLUP for multibreed genomic evaluation in dairy populations.

In india sunflower, an important crucial oilseed crop, has endured considerable yield declines over the past two decades. One of the factor in the vulnerability of the reduced cultivation to powdery mildew disease. This persistent challenge underscores the pressing necessity for resistant germplasm and robust molecular markers to advance breeding initiatives. In this study, the phenotypic diversity of 20 sunflower inbreds was assessed for traits during two seasons (winter 2022, rainy season 2023). Using three multi trait selection indices Genotype-by-Yield-Trait (GYT) biplot, the Multi-Trait Genotype-Ideotype Distance Index (MGIDI) and the Factor Analysis and Interaction Best Linear Unbiased Prediction (FAI-BLUP) were used to evaluate the inbred lines. The inbreds were screened for powdery mildew resistance under both natural (winter 2022, rainy season 2023) and artificial conditions (summer 2023). Significant variation was observed for all traits, with four inbreds (COSF 6B, COSF 10B, COSF 12B and COSF 15B) exhibiting high oil content, seed yield and oil yield per plant. The inbreds COSF 6B, CSFI-17008 and CSFI-17024 consistently achieved the highest rankings across all indices, demonstrating strong potential for exploitation breeding programmes. SSR marker-based cluster analysis using DARwin software grouped the inbreds into five distinct clusters and STRUCTURE analysis revealed five sub-populations. Out of 141 SSR markers tested, 92 primers were polymorphic, with ORS898, ORS188, ORS1008 and ORS725 identified as the most effective. Marker validation using ORS 684 and ORS1110, both linked to powdery mildew resistance, showed that ORS 684 explained 27.17% of phenotypic variation and reliably distinguished resistant and susceptible inbreds. Notably, CSFI 17018 and CSFI 17020 exhibited resistance under both screening conditions and their resistance was further confirmed by marker validation. Based on diversity analyses and artificial screening, we recommend utilizing the identified diverse inbreds-COSF 6B, COSF 10B, COSF 12B, COSF 15B, CSFI 13023, CSFI 17008, CSFI 17018 and CSFI 17020-as parental lines in future breeding programs. For developing powdery mildew-resistant hybrids, inbreds, or varieties through marker-assisted backcrossing, ORS 684 is proposed as an effective foreground marker to facilitate the transfer of resistance loci.