Having chosen the L1000 landmark transcripts using an entirely data-driven approach optimized for maximal information rather than on biological function, we asked whether the landmarks were enriched in any particular functional class (e.g., transcription factors). We computed hypergeometric overlap statistics and found no substantial enrichment for any particular protein class (see STAR Methods). Similarly, we found no evidence of developmental lineage bias based on an analysis of landmark expression patterns across 30 tissue types (Figure S1B).
