Biologically informed machine learning for identifying human placental cellular heterogeneity and preeclampsia discovery

Single-cell transcripts of 20,518 placental cells and identified 12 major placental cell clusters were collected from the European Bioinformatics Institute (EBI; accession no. EGAS00001002449) (29). Based on the same parameters, we clustered and visualized highly similar cells using t-distribution random neighbourhood embedding (T-SNE) to identify 17 cell subpopulations. we selected nine placental cell clusters that have received more attention from biologists for our study according to the literature survey (Table 1). Considering the sample balance, 7178 single-cell transcriptome data were used to identify human placental cell subpopulations. The samples were randomly divided into a 4809-sample training set and a 2369-sample testing set. The same strategy was applied to split single-cell transcriptomic datasets from healthy and preeclampsia patients (EBI; accession no. EGAS00001002449), with 9852 samples (healthy 4705, preeclampsia 5147) in the training set and 5305 samples (healthy 2473, preeclampsia 2832) in the independent test set