Thanks for this wonderful post! I always appreciate your careful caveats on interpretation of graphic results. I also appreciate your explanation of why you excluded the Italian samples from your baseline PCA (onto which you subsequently mapped the Italians). Please elaborate a little bit on the “Raveane rule” for regional locality definition, and on how you derived individual regional localities. (Since Raveane et al. used different analytical tools - i.e., fS and CP - to define their regions, I wonder if your PCA results would define regions differently.) Maybe Figure 4 would work better as green, like your PC2 color assignment. Thanks for great work! I am looking forward to your world PCA graphic atlas.
Thank you, these are excellent questions. “Raveane rule” was too compressed a description on my part. It does not refer to using Raveane et al.’s fS or chromosome-painting regions as if they were localities. For 129 individuals, the released population crosswalk supplied an Italian administrative-region label; I placed those samples at the corresponding regional capital as a reproducible representative coordinate. These are therefore regional anchors, not inferred individual sampling locations.
Where better information was available, I used it. Another 51 individuals had locality-level coordinates recoverable from AADR metadata and the associated supplement. Thirteen Sicilians had only four source-defined subregional labels, so I used explicitly declared representative cities for those subregions. The Sardinians share a published population-level coordinate and should not be interpreted as providing within-island resolution.
Most importantly, none of these geographic assignments was derived from my PCA. Geography was assigned independently and then used to map the mean PC scores of samples sharing a locality. You are quite right that clustering the Italian PCA coordinates themselves might produce boundaries different from Raveane et al.’s fS/CP regions. That would be an interesting separate analysis, but it would answer a different question—and defining regions from the same PCs being mapped would introduce some circularity.
In the analysis described in the post, the Italians were held out so they could not determine the baseline axes onto which they were projected. The updated online atlas also provides a population-balanced inclusive analysis in which Italian populations contribute controlled total influence rather than dominating through their larger sample counts.
I also agree about Figure 4: using green for PC2 would make its relationship to the RGB convention much more immediately apparent. I am gonna regenerate the figure and replace it. Thanks for the suggestion! I will keep making upgrades to the PCA Atlas and I hope to be able to ship the world atlas soon!
Thanks for this wonderful post! I always appreciate your careful caveats on interpretation of graphic results. I also appreciate your explanation of why you excluded the Italian samples from your baseline PCA (onto which you subsequently mapped the Italians). Please elaborate a little bit on the “Raveane rule” for regional locality definition, and on how you derived individual regional localities. (Since Raveane et al. used different analytical tools - i.e., fS and CP - to define their regions, I wonder if your PCA results would define regions differently.) Maybe Figure 4 would work better as green, like your PC2 color assignment. Thanks for great work! I am looking forward to your world PCA graphic atlas.
Thank you, these are excellent questions. “Raveane rule” was too compressed a description on my part. It does not refer to using Raveane et al.’s fS or chromosome-painting regions as if they were localities. For 129 individuals, the released population crosswalk supplied an Italian administrative-region label; I placed those samples at the corresponding regional capital as a reproducible representative coordinate. These are therefore regional anchors, not inferred individual sampling locations.
Where better information was available, I used it. Another 51 individuals had locality-level coordinates recoverable from AADR metadata and the associated supplement. Thirteen Sicilians had only four source-defined subregional labels, so I used explicitly declared representative cities for those subregions. The Sardinians share a published population-level coordinate and should not be interpreted as providing within-island resolution.
Most importantly, none of these geographic assignments was derived from my PCA. Geography was assigned independently and then used to map the mean PC scores of samples sharing a locality. You are quite right that clustering the Italian PCA coordinates themselves might produce boundaries different from Raveane et al.’s fS/CP regions. That would be an interesting separate analysis, but it would answer a different question—and defining regions from the same PCs being mapped would introduce some circularity.
In the analysis described in the post, the Italians were held out so they could not determine the baseline axes onto which they were projected. The updated online atlas also provides a population-balanced inclusive analysis in which Italian populations contribute controlled total influence rather than dominating through their larger sample counts.
I also agree about Figure 4: using green for PC2 would make its relationship to the RGB convention much more immediately apparent. I am gonna regenerate the figure and replace it. Thanks for the suggestion! I will keep making upgrades to the PCA Atlas and I hope to be able to ship the world atlas soon!
Thanks for the clarifications! 🙏