Manifold-learning is particularly useful to resolve the complex cellular state space from single-cell RNA sequences. While current manifold-learning methods provide insights into cell fate by inferring graph-based trajectory at cell level, challenges remain to retrieve interpretable biology underlying the diverse cellular states. Here, we described MGPfact^XMBD, a model-based manifold-learning framework and capable to factorize complex development trajectories into independent bifurcation processes of gene sets, and thus enables trajectory inference based on relevant features. MGPfact^XMBD offers a more nuanced understanding of the biological processes underlying cellular trajectories with potential determinants. When bench-tested across 239 datasets, MGPfact^XMBD showed advantages in major quantity-control metrics, such as branch division accuracy and trajectory topology, outperforming most established methods. In real datasets, MGPfact^XMBD recovered the critical pathways and cell types in microglia development with experimentally valid regulons and markers. Furthermore, MGPfact^XMBD discovered evolutionary trajectories of tumor-associated CD8⁺ T cells and yielded new subtypes of CD8⁺ T cells with gene expression signatures significantly predictive of the responses to immune checkpoint inhibitor in independent cohorts. In summary, MGPfact^XMBD offers a manifold-learning framework in scRNA-seq data which enables feature selection for specific biological processes and contributing to advance our understanding of biological determination of cell fate.

Apart from ancestry, personal or environmental covariates may contribute to differences in polygenic score (PGS) performance. We analyzed the effects of covariate stratification and interaction on body mass index (BMI) PGS (PGS_BMI) across four cohorts of European (N = 491,111) and African (N = 21,612) ancestry. Stratifying on binary covariates and quintiles for continuous covariates, 18/62 covariates had significant and replicable R² differences among strata. Covariates with the largest differences included age, sex, blood lipids, physical activity, and alcohol consumption, with R² being nearly double between best- and worst-performing quintiles for certain covariates. Twenty-eight covariates had significant PGS_BMI–covariate interaction effects, modifying PGS_BMI effects by nearly 20% per standard deviation change. We observed overlap between covariates that had significant R² differences among strata and interaction effects – across all covariates, their main effects on BMI were correlated with their maximum R² differences and interaction effects (0.56 and 0.58, respectively), suggesting high-PGS_BMI individuals have highest R² and increase in PGS effect. Using quantile regression, we show the effect of PGS_BMI increases as BMI itself increases, and that these differences in effects are directly related to differences in R² when stratifying by different covariates. Given significant and replicable evidence for context-specific PGS_BMI performance and effects, we investigated ways to increase model performance taking into account nonlinear effects. Machine learning models (neural networks) increased relative model R² (mean 23%) across datasets. Finally, creating PGS_BMI directly from GxAge genome-wide association studies effects increased relative R² by 7.8%. These results demonstrate that certain covariates, especially those most associated with BMI, significantly affect both PGS_BMI performance and effects across diverse cohorts and ancestries, and we provide avenues to improve model performance that consider these effects.