Abstract:
Population genetics is the study of spatio-temporal genetic variants among individuals. Its purpose is to understand evolution: the change in frequency of alleles over time. The effects of these alleles are expressed on different levels of biological organization, from molecular com-plexes to entire organisms. Eventually, they will affect traits that can influence the survival and reproduction of organisms. Fitness is a probability of transferring alleles to subsequent genera-tions with respect to successful survival and reproduction. Due to differential fitness, any phe-notypic properties that confer beneficial effects on survival and reproduction may presumably become prevalent in a population.
Random mutations introduce new alleles in a population. The underlying changes in DNA sequences can be caused by replication errors, failures in DNA repair processes, or insertion and deletion of transposable elements. For sexual organisms, genetic recombination randomly mixes up the alleles in chromosomes, in turn, yielding a new composition of alleles though it does not change the allele frequencies. On the molecular level, mutations on a set of loci may cause a gain or loss of function resulting in totally different phenotypes, hereby influencing the survival of an organism. Despite the dominance of neutral mutations, the accumulation of small changes over time may affect the fitness, and further contribute to evolution.
The goal of this study is to provide a framework for a comparative analysis on large-scale genomic datasets, especially, of a population within a species such as the 1001 Genomes Project of Arabidopsis thaliana, the 1000 Genomes Project of humans, or metagenomics datasets. Algo-rithms have been developed to provide following features: 1) denoising and improving the ef-fective coverage of raw genomic datasets (Trowel), 2) performing multiple whole genome alignments (WGAs) and detecting small variations in a population (Kairos), 3) identifying struc-tural variants (SVs) (Apollo), and 4) classifying microorganisms in metagenomics datasets (Po-seidon). The algorithms do not furnish any interpretation of raw genomic data but provide anal-yses as basis for biological hypotheses.
With the advances in distributed and parallel computing, many modern bioinformatics algo-rithms have come to utilize multi-core processing on CPUs or GPUs. Having increased computa-tional capacity allows us to solve bigger and more complex problems. However, such hardware advances do not spontaneously give rise to the improved utilization of large-size datasets and do not bring insights by themselves to biological questions. Smart data structures and algorithms are required in order to exploit the enhanced computing power and to extract high quality infor-mation. For population genetics, an efficient representation for a pan genome and relevant for-mulas should be manifested. On top of such representation, sequence alignments play pivotal roles in solving biological problems such that one may calculate allele frequencies, detect rare variants, associate genotypes to phenotypes, and infer causality of certain diseases. To detect mutations in a population, the conventional alignment method is enhanced as multiple genomes are simultaneously aligned.
The number of complete genome sequences has steadily increased, but the analysis of large, complex datasets remains challenging. Next Generation Sequencing (NGS) technology is consid-ered one of the great advances in modern biology, and has led to a dramatically more precise and detailed understanding of genomes and their activities. The contiguity and accuracy of se-quencing reads have been improving so that a complete genome sequence of a single cell may become obtainable from a sequencing library in the future. Though chemical and optical engi-neering are main drivers to advance sequencing technology, informatics and computer engineer-ing have significantly influenced the quality of sequences. Genomic sequencing data contain errors in forms of substitution, insertion, and deletion of nucleotides. The read length is far shorter than a given genome. These problems can be alleviated by means of error corrections and genome assemblies, leading to more accurate downstream analyses.
Short read aligners have been the key ingredient for measuring and observing genetic muta-tions using Illumina sequencing technology, the dominant technology in the last decade. As long reads from newer methods or assembled contigs become accessible, mapping schemes capturing long-range context, but not lingering in local matches should be devised. Parameters for short read aligners such as the number of mismatches, gap-opening and -extending penalty are not directly applicable to long read alignments. At the other end of the spectrum, whole genome aligners (WGA) attempt to solve the alignment problem in a much longer context, providing es-sential data for comparative studies. However, available WGA algorithms are not yet optimized concerning practical uses in population genetics due to high computing demands. Moreover, too little attention has been paid to define an ideal data format for applications in comparative ge-nomics.
To deal with datasets representing a large population of diverse individuals, multiple se-quence alignment (MSA) algorithms should be combined with WGA methods, known as multi-ple whole genome alignment (MWGA). Though several MWGA algorithms have been proposed, the accuracy of algorithms has not been clearly measured. In fact, known quality assessment tools have yielded highly fluctuating results dependent on the selection of organisms, and se-quencing profiles. Of even more serious concern, experiments to measure the performance of MWGA methods have been only ambiguously described. In turn, it has been difficult to inter-pret the multiple alignment results. With known precise locations of variants from simulations and standardized statistics, I present a far more comprehensive method to measure the accuracy of a MWGA algorithm.
Metagenomics is a study of the genetic composition in a given community (often, predomi-nantly microbial). It overcomes the limitation of having to culture each organism for genome sequencing and also provides quantitative information on the composition of a community. Though an environmental sample provides more natural genetic material, the complexity of analyses is greatly increased. The number of species can be very large and only small portions of a genome may be sampled. I provide an algorithm, Poseidon, classifying sequencing reads to taxonomy identifiers at a species resolution and helping to quantify their relative abundances in the samples. The interactions among individual bacteria in a certain population can result in both conflict and cooperation. Thus, a mixture of diverse bacteria species shows a set of functional adaptations to a particular environment. The composition of species would be changed by dis-tinct biotic or abiotic factors, which may lead to a successive alteration in susceptibility of a host to a certain disease. In turn, basic concerns for a metagenomics study are an accurate quantifica-tion of species and deciphering their functional role in a given environment.
In summary, this work presents advanced bioinformatics methods: Trowel, Kairos, Apollo, and Poseidon. Trowel corrects sequencing errors in reads by utilizing a piece of high-quality k-mer information. Kairos aligns query sequences against multiple genomes in a population of a single species. Apollo characterizes genome-wide genetic variants from point mutations to large structural variants on top of the alignments of Kairos. Poseidon classifies metagenomics datasets to taxonomy identifiers. Though the work does not directly address any specific biological ques-tions, it would provide preliminary materials for further downstream analyses.