Extended Data Fig. 5. Validation of the observed GE-CP relationship by GO-terms search analysis.

Landmark genes highly predictable according to morphological features in each specific Cell Painting channel are more likely to have GO annotation related to that channel compared to the rest of CP channels. For each channel in the rows of the table, the first column shows the Odds Ratio (OR) derived from the Fisher’s exact test for associations between the landmark genes being highly predictable (R2>0.6) by CP features in a channel and having GO annotations for that channel. The second column shows the association between the same set of highly predictable genes and having GO annotation for any channel but not the target row channel. Higher values in the first column compared to the second column show that highly predictable genes according to features in a CP channel are more likely to have GO annotations for that channel compared to the rest of the channels. This pattern holds for DNA and ER channels but not for the rest of CP channels. The third and fourth columns show the same associations but for low-predictability genes (R2<0). Lower values in the third column compared to the fourth column show that non-predictable genes according to features in a CP channel are less likely to have GO annotations for that channel compared to the rest of the channels. This pattern holds for all CP channels except for RNA. The CP channel specific predictability map used for this analysis was derived from the result of the experiment and results presented partially in Figure 2d. As we can observe from the map, usually multiple categories of morphological features contribute to the predictability of a gene, which explains the lack of a simple relationship between a given channel’s predictability and GO term associations presented in this table.