(**A**) We validated hits using an independent experimental assay based on the population-averaged mean value of each phenotype . We carried out this measurement both for hits as well as for a number of non-hits (genes with below-threshold Z-scores). The y-axis shows the fraction of original hits validated (FP = 0.1); the x-axis shows the fraction of sub-threshold genes validated. Each dot gives the result for a single feature type; hits for features G11 and G15 were not included in the secondary measurement. Horizontal and vertical dotted lines show the FP rate. The distinction between hits and sub-threshold genes was based on shape-based scoring alone, but the former are detected at a much higher rate than the latter using the population-average-based assay. This demonstrates a strong correlation between the ability of a gene to influence the mean value of a phenotype and the shape of a phenotypic population distribution. (**B**) Relationships of hit subsets to cell density. We show the cell number per imaging field as a vertical histogram. The median (horizontal line), mean (box), and percentiles (5%, 25%, 75%, 95%) of the cell number distribution are overlaid. Histograms are separately shown for negative control wells, all test wells, and for fluid uptake, Transferrin uptake, and nuclear morphology hits. (**C**) Area-proportional Venn diagram of hits that influence fluid-phase uptake (F), Transferrin-receptor-mediated uptake (T), or nuclear and cell morphology (N). Of the 26 genes that influence cell size, 21 which do not influence other features have been omitted. Numbers give the sizes of non-overlapping subsets. The total number of hits is 1051 (shown) + 21 (not shown). (**D**) Functional enrichment. We annotated genes according to the Gene Ontology (GO) ‘cellular component’ classification system, using only the most specific term for each gene. We used the one-tailed Fisher’s exact test to determine an enrichment p-value for each teach GO term among the 1072 hits, given its background occurrence among the 7216 RNAi probes. To correct p-values for multiple hypotheses, we used 1000 simulated datasets in which GO terms had been randomly permuted. The table shows the seven GO terms with corrected p-values < 0.1, along with the observed and expected number of genes among the seven non-overlapping gene subsets of the Venn diagram.

## PubMed Commons