PyPop update--a software pipeline for large-scale multilocus population genomics.

Population genetic statistics from multilocus genotype data inform our understanding of the patterns of genetic variation and their implications for evolutionary studies, generally, and human disease studies in particular. In any given population one can estimate haplotype frequencies, identify deviation from Hardy-Weinberg equilibrium, test for balancing or directional selection, and investigate patterns of linkage disequilibrium. Existing software packages are oriented primarily toward the computation of such statistics on a population-by-population basis, not on comparisons among populations and across different statistics. We developed PyPop (Python for Population Genomics) to facilitate the analyses of population genetic statistics across populations and the relationships among different statistics within and across populations. PyPop is an open-source framework for performing large-scale population genetic analyses on multilocus genotype data. It computes the statistics described above, among others. PyPop deploys a standard Extensible Markup Language (XML) output format and can integrate the results of multiple analyses on various populations that were performed at different times into a common output format that can be read into a spreadsheet. The XML output format allows PyPop to be embedded as part of a larger analysis pipeline. Originally developed to analyze the highly polymorphic genetic data of the human leukocyte antigen region of the human genome, PyPop has applicability to any kind of multilocus genetic data. It is the primary analysis platform for analyzing data collected for the Anthropological component of the 13th and 14th International Histocompatibility Workshops. PyPop has also been successfully used in studies by our group, with collaborators, and in publications by several independent research teams.

Evolution of HLA class II molecules: Allelic and amino acid site variability across populations.

Analysis of the highly polymorphic beta1 domains of the HLA class II molecules encoded by the DRB1, DQB1, and DPB1 loci reveals contrasting levels of diversity at the allele and amino acid site levels. Statistics of allele frequency distributions, based on Watterson's homozygosity statistic F, reveal distinct evolutionary patterns for these loci in ethnically diverse samples (26 populations for DQB1 and DRB1 and 14 for DPB1). When examined over all populations, the DQB1 locus allelic variation exhibits striking balanced polymorphism (P < 10(-4)), DRB1 shows some evidence of balancing selection (P < 0.06), and while there is overall very little evidence for selection of DPB1 allele frequencies, there is a trend in the direction of balancing selection (P < 0.08). In contrast, at the amino acid level all three loci show strong evidence of balancing selection at some sites. Averaged over polymorphic amino acid sites, DQB1 and DPB1 show similar deviation from neutrality expectations, and both exhibit more balanced polymorphic amino acid sites than DRB1. Across ethnic groups, polymorphisms at many codons show evidence for balancing selection, yet data consistent with directional selection were observed at other codons. Both antigen-binding pocket- and non-pocket-forming amino acid sites show overall deviation from neutrality for all three loci. Only in the case of DRB1 was there a significant difference between pocket- and non-pocket-forming amino acid sites. Our findings indicate that balancing selection at the MHC occurs at the level of polymorphic amino acid residues, and that in many cases this selection is consistent across populations.

Locus and population specific evolution in HLA class II genes.

The population genetics of the HLA class II loci was studied with reference to variation in the frequency of (a) alleles at a locus and (b) amino acids at specific sites. Variation was surveyed at 4 loci (DRB1, DQA1, DQB1, and DPB1) in 22 populations from the Twelfth International Histocompatibility Workshop (Saint-Malo, 1996). Allele and amino acid variation was measured by computing heterozygosity and the effective number of alleles. Substantial variations in polymorphism were observed among the various populations and loci studied. In the majority of the populations, DRB1 has the highest heterozygosity and effective number of alleles. As previously shown, the Amerindian populations have lower levels of allelic diversity when compared to other populations. At the amino acid level, DRB1 antigen recognition sites (ARS) have the highest heterozygosities and effective number of alleles. For the other loci (DPB1, DQA1, and DQB1) for which there is no crystal structure and for which ARS sites were inferred from DRB1, non-ARS sites were often among the sites with highest levels of variation. It is possible that these putative non-ARS sites do play a role in antigen presentation. The homozygosity test for neutrality was applied to allele and amino acid data. Of the four HLA class II loci studied, only DPB1 failed to show evidence of balancing selection. DQB1 and DQA1 depart significantly from neutrality in the largest number of populations. Genetic distances between populations were computed based on frequency of alleles and amino acids at ARS sites.

Multivariate optical computation for predictive spectroscopy.

A novel optical approach to predicting chemical and physical properties based on principal component analysis (PCA) is proposed and evaluated using a data set from earlier work. In our approach, a regression vector produced by PCA is designed into the structure of a set of paired optical filters. Light passing through the paired filters produces an analog detector signal that is directly proportional to the chemical/physical property for which the regression vector was designed. This simple optical computational method for predictive spectroscopy is evaluated in several ways, using the example data for numeric simulation. First, we evaluate the sensitivity of the method to various types of spectroscopic errors commonly encountered and find the method to have the same susceptibilities toward error as standard methods. Second, we use propagation of errors to determine the effects of detector noise on the predictive power of the method, finding the optical computation approach to have a large multiplex advantage over conventional methods. Third, we use two different design approaches to the construction of the paired filter set for the example measurement to evaluate manufacturability, finding that adequate methods exist to design appropriate optical devices. Fourth, we numerically simulate the predictive errors introduced by design errors in the paired filters, finding that predictive errors are not increased over conventional methods. Fifth, we consider how the performance of the method is affected by light intensities that are not linearly related to chemical composition (as in transmission spectroscopy) and find that the method is only marginally affected. In summary, we conclude that many types of predictive measurements based on use of regression (or other) vectors and linear mathematics can be performed more rapidly, more effectly, and at considerably lower cost by the proposed optical computation method than by traditional dispersive or interferometric instrumentation. Although our simulations have used Raman experimental data, the method is equally applicable to Near-IR, UV-vis, IR, fluorescence, and other spectroscopies.

Association mapping of disease loci, by use of a pooled DNA genomic screen.

Genomic screening to map disease loci by association requires automation, pooling of DNA samples, and 3,000-6,000 highly polymorphic, evenly spaced microsatellite markers. Case-control samples can be used in an initial screen, followed by family-based data to confirm marker associations. Association mapping is relevant to genetic studies of complex diseases in which linkage analysis may be less effective and to cases in which multigenerational data are difficult to obtain, including rare or late-onset conditions and infectious diseases. The method can also be used effectively to follow up and confirm regions identified in linkage studies or to investigate candidate disease loci. Study designs can incorporate disease heterogeneity and interaction effects by appropriate subdivision of samples before screening. Here we report use of pooled DNA amplifications-the accurate determination of marker-disease associations for both case-control and nuclear family-based data-including application of correction methods for stutter artifact and preferential amplification. These issues, combined with a discussion of both statistical power and experimental design to define the necessary requirements for detecting of disease loci while virtually eliminating false positives, suggest the feasibility and efficiency of association mapping using pooled DNA screening.