A Framework Integrating Single-Cell Metallomic Data in Health Effect Analysis via Quantile Features and Machine Learning

Heavy and essential metals coexist in a cell, collectively
participate
in cellular biological processes, and lead to overall health consequences.
However, current analytical strategies cause information loss when
using aggregated features and are unable to establish comprehensive
associations between single-cell metallomic data and health outcomes.
We developed an analytical framework with more accurate characterization
of multimetal distribution features, enabling the integration of single-cell
metallomic data into health-related studies. We used quantiles to
comprehensively extract the features of multimetals in spermatozoa
population. Then, machine learning and factor analysis further consolidated
key features into interpretable indices of multimetal co-occurrence
characteristics. This framework addressed the conservation of the
pseudobulk method and bias of the single-cell method, due to a more
detailed characterization and appropriate integration of multimetal
features at single-cell resolution. We showcase a negative association
between multiple metals and sperm motility in the Bayesian Kernel
Machine Regression (BKMR) model with a 10% increase in the collective
features of 34 metals corresponding to a 3.3% reduction in sperm motility.
Notably, the contribution of the multifaceted effects of lead and
platinum suggests that dynamic changes in metal distribution may hold
biological significance comparable to their total content. Finally,
the generalizability of this analytical framework was validated with
an additional single-cell metallomic data set. This method facilitates
the application of single-cell metallomic measurement techniques in
health effect studies, providing deeper insights for understanding
the collective biological roles of metals in multicell organisms.