Context-Aware Hypergraph Representation of Higher-Order Transcriptional Activation in Single-Cell Antimalarial Resistance Data

Context-Aware Hypergraph Representation of Higher-Order Transcriptional Activation in Single-Cell Antimalarial Resistance Data Single-cell transcriptomic studies of antimalarial resistance commonly represent transcriptional relationships as individual features or pairwise gene associations, potentially overlooking coordinated higher-order relationships involving multiple genes. We developed a context-aware hypergraph framework to represent higher-order transcriptional activation in Plasmodium falciparum single-cell RNA-seq data from two isogenic lines differing at the K13 locus (K13C580 and K13580Y), sampled under untreated, dihydroartemisinin (DHA), and vehicle, dimethylsulfoxide (DMSO) conditions at 2, 4, and 6 hours post-treatment. Following quality control, 161,100 cells from an initial 280,000 across 14 sequencing libraries were retained, with 2,000 highly variable genes identified and 1,286 meeting a minimum detection threshold (≥ 1% of response cells) for hypergraph construction. Transcriptional contexts were defined by combining unsupervised cluster identity, genotype, and timepoint, retaining 122 contexts with at least 100 cells. Within each context, genes with elevated standardized expression (z ≥ 1.5) were identified as candidate members of context-specific hyperedges, with hyperedges restricted to 3-50 genes. This yielded 80 hyperedges spanning 721 genes and 2,127 gene-hyperedge memberships, forming a single connected structure comprising all 801 genes and hyperedges. Validation revealed a strong association between the number of genes passing the activation threshold and context sequencing depth (Pearson (r) = 0.858 for total counts; r = 0.881 for detected genes), indicating that apparent activation breadth was substantially influenced by technical depth. Seven clusters exceeding three times the global median sequencing depth accounted for 38 of the 80 hyperedges. Excluding these depth-outlier clusters yielded a corrected structure of 42 hyperedges spanning 112 genes and 353 memberships, with 100% of the 154 nodes retained in a single connected component. The depth-controlled hypergraph retained treatment, genotype, and temporal structure, with DHA-dominant hyperedges represented across all three timepoints and both K13 genotypes. DHA-associated genes included 29 resistant-only, 25 sensitive-only, and 27 shared genes after depth-outlier exclusion. These findings demonstrate that context-specific transcriptional activation can be represented as a sparse, higher-order hypergraph while explicitly accounting for sequencing-depth effects, providing a validated computational representation for studying coordinated transcriptional patterns in antimalarial resistance. Keywords: Plasmodium falciparum, Artemisinin resistance, Single-cell transcriptomics, hypergraph, Higher-order relationships; machine learning