The HUGO Pan-Asian SNP Consortium (PASNP) was a successful model of a genomics research network in Asia, which surveyed the genetic make-up of 1928 participants from 73 populations and permitted an unprecedented insight into possible human migration across the Asian continent.1 However, owing to the small number of participants in each population and the use of a low-density genotyping microarray, analyses of the PASNP data were mostly limited to broad analyses of population structure between the populations, and less is known about the cross-population microstructures such as admixture, linkage disequilibrium variation, as well as within-population adaptive forces such as natural selection and population bottlenecks. In fact, the sparse SNP density confounded several attempts to investigate the role of natural selection in driving genetic diversity between these populations, regardless of whether these relied on comparing allele frequencies2 or locating long haplotypes.3

As the largest continent on Earth spanning over 93.5° in latitude and 196.3° in longitude, Asia possesses tremendous diversity in climate and altitude. This geographical variation, coupled with the myriad cultures and languages in Asia, has exerted profound influences on the genetics of Asian populations, allowing these populations to adapt differentially to high altitude,4, 5, 6, 7, 8 diet9 and infectious diseases.3, 10, 11 Given the vast expanse of Asia, populations that are geographically closer are expected to have adapted in a similar fashion, given the likelihood of more similar microclimates and dietary practices, whereas the selection on populations further apart may substantially differ. However, to date no collection of genetic data spanning multiple Asian populations exits at sufficient genomic coverage to allow a comprehensive and systematic evaluation of local adaptation in Asia.2, 12, 13, 14

The Asian Diversity Project (ADP) overcomes this paucity in data by enlisting scientists across Asia to cooperate and share resources for a deeper survey of population genetics in Asia. To date, ADP has assembled genetic data for 37 cosmopolitan and ethnic minority populations from public databases15, 16, 17, 18, 19 and private repositories (Figure 1a), where there are at least half a million SNP markers genotyped in each population and 33 of the 37 populations surveyed at least 10 individuals (Supplementary Table S1).

Figure 1
figure 1

Coverage and population structure of Asian Diversity Project. (a) Illustration of the coverage of the Asian Diversity Project, where each circle highlights the location of origin for a specific population group with the assigned name, chosen using either existing naming conventions or by the country name. The number accompanying each circle in the figure legend indicates the number of individuals surveyed in that particular population. (b) A nearest-neighbour phylogenetic analysis of the 39 populations, where except for CEU and YRI, all the 37 ADP populations have been binned into four groups as shown by the coloured segments. (c) Principal component analysis of the 37 ADP populations using a common set of 30 927 SNPs in the 39 populations.

Here, we aim to characterise the extent and sharing of positive selection in the 37 ADP populations. To do this, we first investigated the degree of genetic similarity between these populations with a comprehensive suite of population structure analyses, which supported the clustering of the Asian populations into four major population groups that corresponded broadly with the geographical distributions of the populations and present-day definitions of Asian populations. The genotype data for each population were subsequently phased with BEAGLE,20 and the haplotype-based methods haploPS14 were employed on the phased data to locate genomic signatures of positive selection. The regions identified to be under selection were summarised according to the degree of sharing across populations in the four groups. We believe this is the first survey of positive selection in Asia that possesses sufficient statistical power to allow a systematic evaluation of the extent of sharing signatures of local adaptation. The genotype data have been deposited at the European Genome-phenome Archive (EGA, under the accession number EGAS00001002100.


Sample collection and genotyping in ADP

The ADP has integrated 37 Asian populations from publicly available databases such as (i) the four populations with Asian ancestries in Phase 3 of the International HapMap Project19 (HapMap: CHB, CHD, GIH, JPT); (ii) three populations from the Singapore Genome Variation Project16 (SGVP: CHS, INS, MAS); and four North Eurasian populations (Kurd, Buryat, Kyrgystani, Mongolian) from the Jorde Lab17, 18 ( DNA samples of Tajik, Uyghur, Kazakh and Kirgiz were collected from the Xinjiang Uyghur Autonomous Region in China, whereas DNA samples of Hui (30) were collected from the Ningxia Hui Autonomous Region in China. All procedures were followed in accordance with the ethical standards of the ethics committee of Fudan University and the Helsinki Declaration of 1975, as revised in 2000. Data and information of the other ethnic populations from China (Dong, Han, Li, Tibetan, Yao, Zhuang) were described elsewhere as indicated.21, 22 The samples from eight Japanese prefectures (Amagasaki, Ehime, Fukuoka, Kita-nagoya, Ohasama, Okinawa,23 Shimane, Tokyo) were coordinated through the National Center for Global Health and Medicine (NCGM) in Tokyo on behalf of the Japanese SNP Consortium. The NCGM was also responsible for coordinating the Sri Lankan data sets that included Sri Lanka Tamils, Indian Tamils and Sinhalese. A genetic data set of 200 Kinh Vietnamese was made available from the Oxford University Clinical Research Unit in Ho Chi Minh City, Vietnam, whereas two independent Thai data sets were contributed, respectively, by the Siriraj Hospital, Mahidol University, and Genome Technology Research Unit, the National Center for Genetic Engineering and Biotechnology (BIOTEC).

The samples from the HapMap and SGVP were genotyped on both the Affymetrix Genome-wide Human SNP array 6.0 and the Illumina Human1M BeadChip. Genotyping of the Chinese and Korean samples was performed only on the Affymetrix 6.0 array. The Japanese samples were genotyped on different platforms, with Ohasama on Affymetrix SNP 5.0, Tokyo and Shimane on Illumina 550k, Okinawa on the Illumina Omni-Express23 and the rest of the populations on Illumina 2.5M. The three Sri Lankan populations were similarly genotyped on the Illumina 2.5M. Both sets of Thai samples were genotyped on the Illumina 610k, whereas the Vietnamese Kinhs were genotyped on the Illumina 660W. All coordinates reported here have been referenced against Build 37 (hg19).

Data sets from the contributing sites have undergone quality checks, which have excluded problematic samples and SNPs (Supplementary Table S2). Specific details of the QC process can be found from the respective publications. However, for the purpose of our intended analysis, we reiterate here that this QC process also excluded samples with a high level of identity-by-state (IBS) genotypes, where only the sample with the higher call rate was retained in each pair of individuals with excessive IBS.

Details of the 37 ADP populations, including the sample size in each population, can be found in Supplementary Table S1.

Data availability

The post-QC genotype data of the 37 populations in the ADP are available upon request to As the ADP includes data sets from public resources, only the post-QC data will be shared for these populations and requests for the raw data set will be directed to the public resources. The raw genotype data for all other populations will however be available.

Population structure analyses

The analyses of the population structure for the 37 ADP populations were performed with four different strategies: (i) calculating the Weir and Hill FST between every pair of populations, defined as

with ni the sample size of the ith population, n a constant determined by the number of samples and populations, and and “ denote the average frequency of the allele in population i and both populations, respectively; (ii) a nearest-neighbour phylogenetic tree constructed with a genetic distance matrix where the (i, j)th element of the matrix is represented by the FST for population i and population j; (iii) a series of principal component analyses (PCAs) with 30 927 SNPs that are present in all 37 ADP populations as well as CEU and YRI from HapMap3 using smartPCA in the EIGENSOFT package;24 and (iv) a series of admixture analyses with ADMIXTURE version 1.2225 with the same 30 927 SNPs, where the number of ancestral populations (K) was allowed to vary from 2 to 12.

Haplotype phasing

The genotype data for the 33 ADP populations with at least 10 individuals were phased with BEAGLE version 3.3.226 to obtain the haplotype data necessary for the analysis of positive selection. Each population was phased independently of the others although internal validation indicated there were no significant differences in the phased haplotypes even if multiple populations were phased simultaneously (results not shown).

Locating genomic signatures of positive selection

The search for genomic signatures of positive selection was performed with the haplotype-based method haploPS,14 which was applied to the phased haplotype data from each of the 33 ADP populations and effectively relied on evidence of uncharacteristically long haplotypes when adjusted for background recombination.

HaploPS searches for the longest possible haplotype form that is present in the population with at least a pre-defined core haplotype frequency. At each of these frequencies, the identified haplotype form is measured by both the genetic distance spanned and the number of SNPs from the data that is present on the haplotype, where an adjusted haploPS score is subsequently derived from these two metrics after scaling against the number of such haplotypes present across the whole genome for the core frequency. In this study, to reduce the possitibility that false-positive signals of natural selection might confound the clustering analysis of the populations based on selection signals, we applied a stringent threshold for identifying positive selection regions by haploPS, where a genomic region is only identified to present evidence of positive selection if the adjusted haploPS score is <0.01 (see Discussion for sensitivity and specificity comparisons at this threshold). We conducted the search across a range of core haplotype frequencies from 5 to 95% in step-sizes of 5%. For a region identified by haploPS to be positively selected, combining the output across the range of core haplotype frequencies additionally allowed the putative frequency of the positively selected allele to be estimated. For a genomic region that is found to be positively selected in multiple populations, the individual selected haplotype forms can also be used to evaluate whether the emergence of the advantageous allele predates the separation of these populations, or is likely to have emerged independently in the different populations, also known as convergent evolution. The haploPS analysis used the software that is publicly available at

HaploPS was chosen as it conferred higher statistical power and lower type I error than conventional methods for detecting positive selection.14 An equivalent analysis with two additional methods for detecting selection (iHS [12], XP-EHH [13]), along with a comparison of their results with those from haploPS can be found in the Supplementary Material.

Haplotype similarity index

Haplotype similarity index (HSI)14 is a metric to infer whether the selection signals shared by multiple populations originated from the same or different mutation events by quantifying the degree of similarity between the extended haplotype forms corresponding to a selection signal identified by haploPS that is concurrently present in multiple populations. A HSI larger or equal to 0.98 indicates high level of similarity and that the haplotype forms are likely to have descended from the same founder haplotype and thus indicative of a single-origin mutation event. HSI smaller than 0.9 indicates different haplotype backgrounds exist due to convergent evolution. Details regarding the formulation and evaluation of the HSI and associated thresholds can be found in Liu et al.14


Population structure analyses

A total of 3933 samples from 37 Asian populations remained after quality control, of which we identified a set of 30 927 SNPs distributed across the genome that were present in all 37 populations. This data set is used for the analysis of population structure, together with the HapMap2 Europeans (CEU) and Africans (YRI) as outgroups. Using the FST values obtained from all possible pairwise comparisons of the 39 populations as inputs for a genetic distance matrix, we constructed a phylogenetic tree where YRI clearly segregated from the rest of the non-African populations, followed by CEU (Figure 1b). Among the Asian populations, the Kurdish and the Tajiks were genetically close to CEU (FST=1.4% and 1.9%, respectively), and all three populations were grouped together with South Asian populations (GIH, INS, Sinhalese, Indian Tamil, Sri Lankan Tamil) in the same clade in the phylogenetic tree. We thus defined this collection of populations as a population group corresponding to the South region. Central Asian samples such as the Kyrgyzstani, Buryat and Mongolians were grouped together with Chinese ethnic minorities that are located in the same geographical region (Uyghur, Kazak, Kirgiz, Tibetan and Hui) in the Central region. The classically defined East Asian samples from all eight Japanese populations, Korea and the Han Chinese samples (CHB, CHD, CHS, Han) were grouped into the East region, whereas the Southeast Asian populations from Thailand, Singapore, Vietnam and the ethnic minority populations of Dong, Li, Yao and Zhuang are grouped as the Southeast region. However, it should be highlighted that the distinction between these regional groupings, in particular between East and Southeast regions, are not as pronounced as those typically observed when comparing between globally distributed ancestry groups, due to the complex nature of migration and admixture between Asian populations (see below and Supplementary Material).

A PCA performed with the same 30 927 SNPs yielded findings that concurred with the phylogenetic analyses, separating the YRI and CEU from the rest of the ADP populations (Supplementary Figure S1A). The first two principal components from the PCA with only the 3933 ADP samples exhibited remarkable correlation with geographical coordinates with the latitude and longitude of the populations (Figure 1b, Supplementary Figure S1, Pearson rlongitude-PC1=0.79, P-value <4.9 × 10−324; rlatitude-PC1=0.58, P-value <4.9 × 10324; Pearson rlongitude−PC2=0.40, P-value=9.9 × 10166; rlatitude−PC2=0.49, P-value=32 × 10272). Owing to the ambiguity between the East Asian and Southeast Asian clades, we performed a separate admixture and PCA with the samples from Thailand, Vietnam, Singapore Malays and the Han Chinese populations (CHB, CHD, CHS). This indicated there is a considerable extent of genetic diversity in Thailand (Supplementary Figure S2), where the Thai people have varying degree of Han Chinese admixture ranging from 10 to 80%27 (Figure 2 and Supplementary Figure S2), with some that are virtually indistinguishable from the Malays despite the Malays and the Han Chinese being genetically distinct (Figure 1b,Supplementary Figure S1). We also observed that samples from Okinawa were genetically distinct from the other seven mainland Japanese groups23, 28 (Figures 1b and 2, Supplementary Figure S1), consistent with the observation from the pairwise FST comparison (Supplementary Figure S3) where the minimum FST between Okinawa and the seven Japanese populations was 0.55%, whereas the maximum FST between any pairs of the seven Japanese populations was only 0.04%.

Figure 2
figure 2

Admixture analysis of ADP populations. Analysis of admixture with the program ADMIXTURE, where the number of ancestry ranges from K=2 to 12. The populations are binned into four population groups, corresponding to the South, Central, East and Southeast regions where each region comprises predominantly but not solely populations located in the corresponding geographical region in Asia.

Detecting genomic signatures of positive selection

To investigate whether local adaptation may have driven the genetic diversity between Asian populations, we selected 33 populations from ADP with more than 10 samples each and analyse each population for evidence of positive selection with haploPS as it exhibits higher statistical power at locating genomic signatures of positive selection across different derived allele frequencies, while possessing lower false-positive rates across the majority of the allelic spectrum, compared with iHS and XP-EHH14 (see Discussion and Supplementary Material). Our analysis with haploPS identified 693 signals in total from the 33 populations, which were the basis for all subsequent analyses.

Quantifying the distribution of selection signals in Asia

We observed that the 693 selection signals (Supplementary Table S3) were located in 140 distinct genomic regions (Supplementary Table S4), where the majority of the signals were actually present in multiple populations but located in the same genomic regions. We thus investigated the pattern of sharing of the 140 selected regions among the 33 populations by calculating a similarity matrix for the 33 populations, where each entry indicates the number of regions out of 140 that are positively selected in the respective population pair. When a PCA was performed on this similarity matrix, the first two leading principal components indicated that population groups that were geographically closer were more likely to share the same selection signals than population groups located further apart (Figure 3). We observed that populations in the South and Central regions can similarly be grouped separately by the extent of sharing of selection signals, whereas there were considerable overlaps between the populations in the East and Southeast regions.

Figure 3
figure 3

Clustering of ADP populations by positive selection signals. A principal component analysis (PCA) on the extent of sharing of the 140 positively selected genomic regions across 33 ADP populations with at least 10 samples each. This effectively relies on the absence and presence of selection signals between the populations to place them in the biplot, where populations are found closer if they possess a greater degree of sharing of the 140 signals.

Subsequently, we investigated whether sharing of selection signals was more likely between specific population groups. Each of the 140 regions was categorised according to whether it was uniquely present in one population group, or whether it was present in specific combinations of two or more population groups (Figure 4a). Of the 140 regions, 76 were only found in populations in one unique population group (Supplementary Table S4). We observed that the greatest number of clade-specific signals was present in the South region that predominantly comprised of South Asian populations. This concurred with the previous observations that the South Asian populations are most distinct from the rest of the Asian populations. The least number of clade-specific signals was in the Southeast region, comprising predominantly of Southeast Asian populations. It is likely due to the greater extent of admixture observed between the populations in East and Southeast regions. For signals that were present in populations from the three population groups, the majority were located outside the South region, in agreement with earlier evidence that South Asian populations are considerably more distinct from the other Asian populations we have considered.

Figure 4
figure 4

Sharing patterns of selection signals in the four population groups.(a) Of the 140 genomic regions that are found to be positively selected in the 33 ADP populations, 53 were present in at least two populations. This illustrates whether each genomic region is selected only in populations from the same group (red bars), or whether the region is selected in populations from two (yellow bars), three (green bars) or all four groups (blue bars). The numbers in brackets correspond to the total number of populations considered in that particular cluster. (b) This illustrates the population frequency spectrum of the advantageous alleles in each of the 140 genomic regions, which were found to be positively selected in the 33 ADP populations, and binned according to whether the region was positively selected in populations from only one group or from two, three or four groups. The frequency was also binned as low (<30%), medium (30–80%) and high (≥80%).

There were also five regions that were positively selected in all the four population groups, which included two regions in the major histocompatibility complex on chromosome 6 that were similarly identified to be selected in CEU. The remaining three regions were only selected in our Asian populations, and none of the five regions were selected in YRI, highlighting the rarity for evolutionary signals to be shared between Africans and populations from all the four Asian groups.

When haploPS identifies a selection signal, it also estimates the frequency of the (unknown) advantageous allele that is driving the selection. By grouping the selected signals according to the estimated frequencies into three categories (low: <30%; medium: 30–80%; and high: ≥80%), we observed that the majority of the signals that were present in all four ancestry clades were high frequency in nature (Figure 4b).

Sharing of selection signals: single-origin selection and convergent evolution across Asian populations

One feature of haploPS is the identification of the selected haplotype form, on which the beneficial allele is expected to reside. For a genomic region that is positively selected in multiple populations, haploPS compares the selected haplotype forms to deduce whether the shared selection signals were likely to have emerged from the same evolutionary event before the populations diverged. Should the selected haplotype forms be considerably different, then this suggests that separate mutation events have occurred independently on different haplotype backgrounds in different populations, a phenomenon defined as convergent evolution. We utilised the HSI to infer whether the selected haplotype forms across multiple populations were likely to have descended from the same founder haplotype (defined as HSI≥0.98) and thus indicative of a single-origin mutation event, or whether fundamentally different haplotype backgrounds exist (HSI<0.90) due to convergent evolution.

Of the 140 regions, 53 were positively selected in at least two populations from any of the four ancestry clades and have more than 20 SNPs in the overlapping regions. The majority (47/53) were inferred to have stemmed from the same evolutionary events given the HSIs for these regions were at least 0.98 (Supplementary Table S5). These included the classical signals: (i) EDAR on chromosome 2, which was positively selected in 11 ADP populations, including the majority of Japanese and Han Chinese populations, and which exhibited perfectly concordant haplotype forms (HSI=1.00) across all 11 populations (Figure 5a); and (ii) the skin pigmentation gene KITLG on chromosome 12, which was selected in 11 populations, and which similarly exhibited perfectly concordant haplotype forms (HSI=1.00, Figure 5b). The KEL gene that encodes the KELL blood group antigens was also observed to be selected in 12 populations from East Asian, Central Asian and Southeast Asian, likely stemming from the same evolutionary event as well given a HSI of 1.00 (Supplementary Figure S4).

Figure 5
figure 5

Selected haplotype forms at EDAR,KITLGandABCC11. HaploPS identified the background haplotypes that corresponded to the identification of positive selection at (a) EDAR in 17 populations, (b) KITLG in 19 populations and (c) ABCC11 in 5 populations. For each of the three regions, the identified haplotypes were found to be identical (HSI of 1.00), thus indicating that there is a single shared advantageous variant that is likely to have arisen in the same mutation event prior to the splitting of the populations.

Of the 53 regions that were positively selected in at least two populations, six regions demonstrated evidence of convergent evolution (Table 1), including a region spanning between 43.4Mb and 44.1Mb on chromosome 2 that encapsulated genes reported to be associated with lipid metabolism (THADA, ABCG5, ABCG8) and were positively selected in Japanese (Amagasaki, Ehime, Nagoya, Fukuoka) and Southeast Asians (Thailand, Vietnam).

Table 1 Genomic regions identified by haploPS to be under convergent evolution in specific populations in the Asian Diversity Project

Evidence of localised adaptation and ancestry-specific selection

In addition to the genomic regions that were positively selected in multiple ancestry clusters, there were also 76 regions that were uniquely selected in just one ancestry cluster in the ADP. ABCC11 was identified to be positively selected in CHB and JPT. This gene carries a nonsynonymous SNP (hg19 chr16:g.48258198C>T, rs17822931) where homozygous carriers of the thymine allele possess dried earwax29 and are more likely to be neutrally nonodorous,30 which has characteristically been found in East Asian populations. The thymine allele at rs17822931 was found on the selected haplotype form that was perfectly identical in the two populations (HSI=1.00, Figure 5c).

Among the selection signals uniquely present in East Asian populations, the region containing the alcohol dehydrogenase (ADH) genes on chromosome 4 was selected in all the Japanese populations (JPT, Amagasaki, Ehime, Shimane, Nagoya, Okinawa, Fukuoka, Tokyo) and CHD. These ADH genes have been shown previously to protect against alcoholism,31 and we observed that the selected haplotype forms from these populations were perfectly identical (HSI=1.00).


This is the first Pan-Asian population genetics survey of 37 cosmopolitan, ethnic minority and indigenous groups in Asia, where at least half a million variants across the human genome have been genotyped. By defining the ADP populations into four population groups, we were able to evaluate how many of the identified genomic regions were positively selected in more than one population groups. Two lines of evidence have also emerged from our analyses to suggest that regions under positive selection in multiple ancestry clades were likely to be older and have emerged prior to the separation of the ancestry clades: (i) the advantageous alleles were estimated to be at higher frequencies; and (ii) the haplotype similarity indices indicated the background haplotype carrying the advantageous alleles were highly concordant even between populations of different ancestry clades. Conversely, for regions that were positively selected uniquely in one ancestry clade, the advantageous alleles were estimated to be at lower frequencies, thus highly suggestive that the selection is very recent and confined to specific populations in the ancestry clade.

Such an unprecedented characterisation of positive selection in Asian populations relied on the confluence of three elements present in this study: (i) the deeper coverage of the genome provided by high-density genotyping arrays; (ii) 33 of the 37 populations surveyed at least 10 individuals each, thus providing greater confidence in estimating haplotype frequencies; and (iii) the availability of a highly sensitive algorithm (haploPS) to locate signatures of positive selection across almost the entire allele frequency spectrum. In addition, the ability to extract the specific haplotype form carrying the advantageous allele allows a direct comparison to be made between different populations that are similarly selected in the same genomic region, thus conferring the opportunity to determine whether the advantageous allele has emerged in a shared evolutionary event or in independent events due to convergent evolution. We highlight the contrast of the current findings with those from a previous analysis of the PASNP data, where only 59 regions were identified to be positively selected across 79 Asian populations due to the sparse data with only 50 000 markers.3 Here, we performed two separate analyses that leveraged on the density of the data: (i) the analysis of population structure relied on a common set of 30 927 SNPs across all the ADP populations; whereas (ii) the analysis of positive selection benefitted from the full set of SNP markers ranging from at least 500 000 SNPs up to 2.5 million in each population. The significantly higher SNP density has previously been shown to improve the detection of positive selection and to reduce false discoveries due to more accurate phasing.3

The clustering of the ADP populations according to the PCA of the positive selection signals yielded a pattern that is remarkably similar to that obtained by the standard PCA of genome-wide SNP data. We have attributed this to the sharing and differentiation of positive selection signals between the populations in the four ancestry clades. However, an alternative explanation is that haploPS may possess a high false-positive rate that resulted in neutrally evolving loci accounting for the majority of the discovered sites under selection. This inevitably meant the discovered regions were but a random sampling of genome-wide data, thus reproducing what has been seen in the conventional PCA of genome-wide SNP data.

To investigate the extent of false positives produced by haploPS (as compared with well-known methods such as iHS, XP-EHH), we performed a series of simulations producing demographies representative of neutral loci under three varying degrees of bottlenecks in Asian populations, and examined how often each of the three methods erroneously report evidence of positive selection. These relied on the same thresholds used to compare the statistical power of the three methods for successfully discovering sites under selection.14 We observed that haploPS reported significantly lower false-positive rates across all three scenarios (Supplementary Table S6). In particular, at stronger bottlenecks corresponding to inbreeding coefficients of 0.2 and 0.3, haploPS reported false-positive rates of 5.8% and 13.3%, respectively, compared with those by iHS (17.2% and 55.3%, respectively) and XP-EHH (4.4% and 8.7%, respectively). Notably, the investigation into the sharing and differentiation of positive selection signals was only made possible with the advent of haploPS,3, 14, 32, 33 given the higher statistical power with accompanying lower type I errors.

Our study also presented multiple lines of evidence to suggest that South Asian populations were considerably different from the rest of the Asian populations, which concurred with previous findings that South Asia is considerably more diverse than other Asian regions.32, 34, 35 This is contrary to the observations between populations in the other three ancestry clades, where the admixture analyses and PCA indicated there is a varying degree of mixing between these clades. The distinction between East and Southeast Asian populations was particularly ambiguous, in part due to Southeast Asian populations such as the Vietnamese Kinh and the Thais that were genetically close to the Han Chinese from East Asia.

Our survey of positive selection in the ADP populations has essentially relied on the detection of classic selective sweeps that produce distinctive imprints of long haplotypes in the genome. Hernandez et al36 reported that such ‘hard’ selective sweeps are likely to be rare in contrast to polygenic selection or selection on standing variation, although we emphasise that our identification of 140 selected regions across 33 Asian populations is in line with their estimate that <10% of nonsynonymous variants are likely to be advantageous. Furthermore, we were able to consistently identify well-characterised signatures of positive selection, such as at EDAR, which has been functionally established to influence physical traits predominantly found in East Asians,37, 38, 39 which is a valuable reassurance providing proof-of-concept for our findings.

The ADP is the second genomics research network to be developed in Asia following the PASNP, and the present coverage of Asia by the ADP is noticeably incomplete. Numerous populations in Asia are rapidly shrinking in size, where an entire community can possess <50 individuals found within a few consanguineous households. Global efforts studying the population genetics of early human populations have focused predominantly on African communities, and surveys of Asian populations, particularly the indigenous groups in many parts of Southeast Asia, remained sparse and fragmented. Given the significant variations of climate, culture and environment in Asia, the population genetics of Asian populations remained surprisingly unexplored. Valuable insights into human migration and possible links between Asian populations and distant hominin cousins such as the Neanderthals and Denisovans are starting to emerge,40, 41, 42, 43, 44 and this may present the much needed motivation to coordinate efforts to study the population genetics of Asia.