Extended Data Fig. 6: Low-dimensional representation of transcriptome data.

(a), (b), Applying principal component analysis to all transcriptomes of all biological replicates reveals that samples from different replicates (indicated by different symbols) are consistent with each other - each data point corresponds to one transcriptome. Furthermore, all samples and all replicates show the same trend when coloured by the sampling time (a) or normalized position (b), which is defined by dividing the sample position by the current swarm radius. These observations show that the gene expression profiles are consistent across all replicates. Both the sampling time and normalized position correlate with principal component 1 (but with a stronger correlation for the normalized position). (c), A spatiotemporal heatmap in which principal component 1 (PC 1, from panels a, b) is colour-coded shows that PC 1 gradually decreases from the swarm front towards the late swarm centre. (d), A spatiotemporal heatmap in which t-SNE 1 is colour-coded shows that t-SNE 1 depends on space and time in the swarm, similar to PC 1, implying that both dimensionality reduction techniques identify consistent gene expression patterns. Taken together, the results presented in (a)-(d) show consistency across independent biological replicates and indicate that while the position of a sample has a strong influence on its gene expression profile, the sampling time also plays an important role and needs to be taken into account.