How near are modern Greeks and Byzantine Anatolians to Anatolians, Caucasians and Levantines?

Every statement in this article is a count or status returned by the platform's PCA distance tool, which counts ordinal statements on the conventional map and on the entropy-rescaled maps of Taleb et al.¹ over four reference panels, three maps, sixteen origin shifts and 2,000 bootstrap resamples. Present-day Anatolian (Turkish), Caucasian (Adygei, Armenian) and Levantine (Palestinian, Druze, Lebanese) populations are projected like the Greek, Byzantine and Roman groups. At the tool's thresholds (10 of 12 settings, 13 of 16 origin shifts) three nearest-class statements hold: the Levantine class is nearest to Byzantine SE Anatolia in all 12 settings, and the Anatolian class is nearest to South Italians and Sicilians. For the modern Greeks the Anatolian class is nearest in 9 of 12 settings, one short of the threshold. Byzantine West Anatolia, Imperial Roman Turkey, Roman Greece, Imperial Roman Lazio, Iron Age Turkey and the Cypriots lie on the Greek side of the Levantine pool in 10 to 12 settings. Most other statements depend on panel or map. Positions are not ancestry shares.

How near are the modern Greeks, and how near are Byzantine Anatolians and Imperial Roman groups, to present-day Anatolians, Caucasians and Levantines? The analysis is distance-based: no admixture model is fitted and no qpAdm is used. Every number below is a count, status or value returned by the platform's PCA distance tool, and the figures that show distance statements are drawn by the platform from the same tool run (Fig. 1 to Fig. 4).

Tool. The platform tool run_pca_distances reads stored PCA coordinates (no genotypes) and, for the conventional map C and the entropy-rescaled maps A and B of Taleb et al.¹, counts statements per group over every reference panel and map, over origin shifts of the rescaled maps (−1, −0.5, +0.5 and +1 standard deviation along PC1 and PC2 of maps A and B; 16 shifts) and over a bootstrap of individuals (2,000 resamples, seed 20261024, panel SB0). A statement is stable if it holds in at least 10 of 12 settings, in at least 13 of 16 origin shifts and in the bootstrap; panel-stable if it holds in the settings but fails the shift or the bootstrap criterion; and otherwise it depends on panel or map. No distance, coordinate, threshold or bootstrap is computed outside the tool.

Data. Human Origins SNP set (AADR v66), no all-SNPs mode; ancient individuals with at least 100,000 SNPs. Four reference panels construct the axes (Table 1): SB0 (26 present-day reference populations from Europe, North Africa, the Caucasus and the Near East), SB1 and SB2 (SB0 plus East Asian or African references) and SB3 (both). A fifth panel (SB0 with shrinkage) failed the tool's panel check (bridge centroids differ by 0.0024 and 0.0020 against a tolerance of 0.001) and is not used; the statements therefore count 4 × 3 = 12 settings. All groups are projected, so the present-day comparison populations enter at individual level.

Groups and classes. Targets: the modern Greeks (Greek and Greek_WGA, n = 36); Byzantine West Anatolia (29) and SE Anatolia (12); Imperial Roman Turkey (11), Greece (4), Lazio (41), Serbia (19) and Croatia (15); Iron Age Turkey (3); and eight modern comparators (Albanian, Bulgarian, Croatian, Cypriot, South Italian, Maltese, Romanian, Sicilian). Classes of present-day populations: Anatolian = Turkish (n = 48) and Turkish Trabzon (9); Caucasian = Adygei (48) and Armenian (15); Levantine = Palestinian (83), Druze (83), Lebanese Christian (9) and Lebanese Muslim (11). Two statements are read: the class that holds the nearest group to a target, and the contrast Q = d(target, modern Greeks) / d(target, pool), where a pool joins the persons of a class; Q < 1 is the side of the modern Greeks. The design, including these definitions, was written before any value was computed.

For the modern Greeks the Anatolian class is nearest in 9 of 12 settings and in all 16 origin shifts, the Caucasian class in 3, the Levantine class in none (Fig. 1, Table 2). The count is one below the tool's threshold, so the nearest class of the modern Greeks depends on panel or map. Only three nearest-class statements reach the thresholds: the Levantine class for Byzantine SE Anatolia (12 of 12 settings, 15 of 16 shifts) and the Anatolian class for the South Italians (11 and 16) and the Sicilians (10 and 15). Byzantine West Anatolia (settings Anatolian/Caucasian/Levantine 5/4/3), Imperial Roman Turkey (7/3/2), Roman Greece (4/4/4), Imperial Roman Lazio (6/4/2), Iron Age Turkey (6/4/2) and the Cypriots (6/5/1) are split; the Serbian and Croatian Roman groups and the Albanians lean to the Anatolian class (9/3/0), the Maltese to the Levantine class (8 of 12). The Caucasian class holds the nearest group in at most 5 of 12 settings for any target.

Against the Levantine pool, Byzantine West Anatolia, Imperial Roman Turkey, Roman Greece, Imperial Roman Lazio, Iron Age Turkey and the Cypriots are on the side of the modern Greeks in 10 to 12 of 12 settings (panel-stable; Fig. 2). On the rescaled maps A and B the bootstrap intervals of Q lie below 1 for all six, so each of them is nearer to the modern Greeks than to the Levantine pool; on map C the intervals include 1 (Table 3; the forest plot of all targets is Fig. 4).

Against the Anatolian and Caucasian pools these groups depend on panel or map, with one exception (Iron Age Turkey, below). For Byzantine West Anatolia, Q for the Anatolian pool is 1.47 on map C (pool side) and 0.60 on map A (Greek side). Byzantine SE Anatolia is on the side of the Anatolian pool in the bootstrap on all three maps (Q = 1.75, 1.91, 2.05; intervals above 1), yet its settings split 6:6. Against the Levantine pool its settings split 5:7 (Q = 1.67 on map C, 0.77 on map A: pool side on one map, Greek side on the other), although the Levantine class holds its nearest group in every setting. Iron Age Turkey is on the side of the Anatolian pool in 12 of 12 settings (panel-stable), with a bootstrap interval that includes 1 on maps A and B (n = 3).

The Serbian and Croatian Roman groups, the Albanians, Bulgarians, Croatians and Romanians are on the Greek side of all three pools (stable), and so are the South Italians against the Anatolian and Levantine pools and the Sicilians against the Levantine pool.

In an average-linkage tree of the group centroids the modern Greeks and Byzantine West Anatolia fall in one cluster in 6 of 12 settings at k = 6 and in 9 at k = 4, the Greeks and the Turkish in 3 and 4, the Greeks and the Cypriots in 6 and 9. The statements depend on the panel: the Spearman correlation of the distance vectors between panels is 0.63 to 0.87 on map C, 0.41 to 0.75 on map A and 0.42 to 0.76 on map B, and up to 21 side statements flip between two panels; between maps within a panel the correlation is 0.80 to 1.00. In a second tool run that adds an Arabian class (Bedouin A and B) and an Iranian class to the overall comparison, only Iron Age Turkey changes (Anatolian 6, Caucasian 2, Levantine 2, Iranian 2 of 12 settings); the Arabian class is nearest to no target.

Figure 3 shows the group centroids of the tool run (panel SB0). On all three maps the modern Greeks lie above the Turkish groups, the Palestinian and Bedouin populations far to the right and the Caucasian and Iranian groups at the bottom; Byzantine SE Anatolia lies below the other Byzantine and Roman groups. Figure 4 is the forest plot of Q against the Levantine pool (the values of Table 3 and Fig. 2, with the 95% bootstrap interval): for the six groups of the previous section the intervals lie below 1 on maps A and B and include 1 on map C.

Figure 5 is a PCA of the individuals from a new PCA run on panel SB0 (same 501,922 SNPs and reference populations, eigenvalues 5.25 and 3.22; an earlier run of the panel with the Greeks as one population gives identical coordinates for every person both runs contain). The modern Greeks are split by place: the Mainland Greeks are the 31 persons with a recorded locality in mainland Greece (14 from Thessaloniki, 17 from Athens; two sample series), the Island Greeks are the two present-day Cretans of the dataset, and five modern Greeks have no recorded locality. The 36 modern Greeks of the tool run are the Mainland Greeks and the five persons without locality; the two Cretans are not part of that target. All 38 Greeks and the Cypriots (5 persons, a separate population added for comparison) share one colour (orange), so that the place of the Greeks in the plot is seen at once; the means of the Mainland Greeks, the Island Greeks and the Cypriots remain separate marks. The Bulgarians (12 persons) are added for comparison as an outline and a dark mean. Outlines ring 95% of the individuals of each present-day class (Anatolian, Caucasian, Levantine), of the Mainland Greeks and of the Bulgarians; two persons are too few for an outline, so the Island Greeks appear as two points and their mean.

The Mainland outline is a band from the lower left to the upper middle that overlaps the left part of the Anatolian outline; the persons from Thessaloniki form its lower left cluster (mean PC2 0.002) and those from Athens its upper part (0.022), so the Mainland mean (−0.0065, 0.0130) lies between them, inside the Mainland outline and outside the Anatolian one, to its lower left. The Island mean (−0.0019, 0.0157) lies on the right edge of the Mainland outline, 0.0046 to the right of the Mainland mean along PC1 (their standard deviation is 0.0047) and just below the lower left corner of the Anatolian outline; the Cypriot mean (0.0103, 0.0218) lies at the lower edge of the Anatolian outline. The Bulgarian outline (dashed, lower left) overlaps the lower left tip of the Mainland outline; the Bulgarian mean (−0.0169, −0.0033) lies inside it, below and to the left of the Thessaloniki cluster, 0.0062 to the left of and 0.0054 below the Thessaloniki mean (−0.0107, 0.0021); all 14 persons of that cluster are from Thessaloniki. This is a neighbourhood of positions on two axes, not evidence of mixture or of a Slavic share. With two island persons the Island position is a description, not an estimate, and the Thessaloniki and Athens persons come from different sample series, so the figure cannot tell a regional from a series difference.

The mean of Imperial Roman Turkey lies inside the Anatolian outline near its lower right edge, the means of Imperial Roman Lazio and Byzantine West Anatolia just below that edge, and the mean of Byzantine SE Anatolia at the upper right between the Anatolian and the Levantine outline. The values are those of the run record (PC2 in the sign of the plot). PC2 has the opposite sign to Fig. 3 (the sign of a principal component is arbitrary) and the axis scales differ, so the two figures are compared by configuration.

The modern Greeks lean to the Anatolian class without reaching the threshold: 9 of 12 settings and all 16 origin shifts. Of the ancient groups only Byzantine SE Anatolia reaches the thresholds for its nearest class, the Levantine one; for the other Byzantine, Roman and Iron Age groups the nearest class depends on panel or map. On the rescaled maps Byzantine West Anatolia, Imperial Roman Turkey, Roman Greece, Imperial Roman Lazio, Iron Age Turkey and Cyprus are nearer to the modern Greeks than to the Levantine pool, in 10 to 12 of 12 settings (panel-stable), whereas on map C the intervals of Q include 1. Most other statements depend on panel or map, and the choice of reference panel changes them more than the choice of map (correlation between panels 0.41 to 0.87, between maps within a panel 0.80 to 1.00). No qpAdm is used, and no ancestry share follows from any statement.

The design was written before any run (md5 a16753e24eed64d2b2b38ac571678e7a); an addendum records the failed panel check, the missing coordinates (the tool returns statements, not coordinates), the corrections of earlier wording and the changes to the PCA figure. Every value in the text, the tables and the figures is a record of the tool run or of a PCA run it reads; no distance, coordinate, threshold or bootstrap was computed elsewhere. The platform's figures were compared with the tables of the series, which are copied from the tool record: the nearest-class counts, origin-shift counts and statuses of all 17 targets (Fig. 1, Table 2), the 48 side statements (Fig. 2) and the 144 values of Q with their intervals in the forest plot (Fig. 4) agree. An earlier series (SH) with own code and other Caucasian and Levantine clades, which were reference populations there, is superseded and not used. Figure 5 comes from a new PCA run on a copy of the series dataset with two sample-level labels (Greek_Mainland: 14 Thessaloniki and 17 Athens persons; Greek_LocalityUnknown: 5 persons); the tool target of 36 Greeks is unchanged.

Data and methods: Human Origins SNP set (base dataset ho-v66-ext-3, AADR v66; custom dataset of series SB), no all-SNPs mode, ancient individuals with at least 100,000 SNPs. Principal components from four reference panels of 26 to 30 present-day populations (smartpca normalisation, 10 components, no outlier removal, no LD pruning), all other groups projected by least squares; conventional and entropy-rescaled maps A and B after Taleb et al.¹ Distance statements are counted over four panels, three maps, 16 origin shifts and a bootstrap of 2,000 resamples (seed 20261024). No qpAdm, no right panel and no significance level are involved: the series makes ordinal statements about distances on a plane map, not ancestry statements.

Magnus Albertus Papers · @AlbertusMvagnus