While biological age scores have been shown to characterize aging by estimating chronological age based on physiological biomarkers, interactions between different age scores are largely unknown. To study this, large-scale multi-modal data are crucial. However, such data are scarce as population-based cohorts are generally restricted in sharing their data. Here, we employ federated learning to study the relationship between the two types of biological age scores: BrainAge based on brain MRI and MetaboAge based on metabolites. Using three large population-based cohorts, we trained a federated deep learning model to estimate BrainAge and compared its performance to models trained in a single cohort. The federated BrainAge model yielded significantly lower error for age prediction across the cohorts than locally trained models. Harmonizing the age interval between cohorts further improved BrainAge accuracy. Subsequently, we compared BrainAge and MetaboAge by performing association analysis and survival analysis for dementia and mortality prediction to further characterize both scores. The association analysis showed a weak association between BrainAge and MetaboAge, while the survival analysis indicated complementary predictive values for the mortality risk of the two scores. Federated learning has been shown to be a valuable technique for enabling the use of research cohorts that are restricted in data sharing. We conclude that BrainAge and MetaboAge act synergetically for the prediction of time to all-cause mortality, and both aging scores capture different aspects of the aging process.

Background: Single-cell transcriptomics is rapidly advancing our understanding of the cellular composition of complex tissues and organisms. A major limitation in most analysis pipelines is the reliance on manual annotations to determine cell identities, which are time-consuming and irreproducible. The exponential growth in the number of cells and samples has prompted the adaptation and development of supervised classification methods for automatic cell identification. Results: Here, we benchmarked 22 classification methods that automatically assign cell identities including single-cell-specific and general-purpose classifiers. The performance of the methods is evaluated using 27 publicly available single-cell RNA sequencing datasets of different sizes, technologies, species, and levels of complexity. We use 2 experimental setups to evaluate the performance of each method for within dataset predictions (intra-dataset) and across datasets (inter-dataset) based on accuracy, percentage of unclassified cells, and computation time. We further evaluate the methods' sensitivity to the input features, number of cells per population, and their performance across different annotation levels and datasets. We find that most classifiers perform well on a variety of datasets with decreased accuracy for complex datasets with overlapping classes or deep annotations. The general-purpose support vector machine classifier has overall the best performance across the different experiments. Conclusions: We present a comprehensive evaluation of automatic cell identification methods for single-cell RNA sequencing data. All the code used for the evaluation is available on GitHub (https://github.com/tabdelaal/scRNAseq-Benchmark). Additionally, we provide a Snakemake workflow to facilitate the benchmarking and to support the extension of new methods and new datasets.