High-Throughput Sequencing And Natural Selection: Studies Of Recent Sweep Inferences And A New Computational Approach For Transcription Identification
Short-read high-throughput sequencing is the most popular approach to collect massive amount of DNA sequence data at declining cost in nearly all fields of current biological studies. Its many varieties have been employed for different research purposes, e.g. genomic sequencing for variant detection, RNA-seq for transcriptome profiling, etc. However, the individual reads and the resulting called sequences frequently have missing and errorprone base calls, and appropriate corrections and evaluations are necessary for drawing conclusions. I examined how missing data and sequence errors affect the power and prediction accuracy of two frequently used methods for the inference of recent positive selection from such datasets. I showed that variant-frequency based method, SweepFinder, is very sensitive to data quality and its sensitivity and prediction accuracy are greatly compromised by missing data or sequence errors. In contrast, the haplotype-based method, iHS, is very robust to missing data and sequence errors and is able to efficiently detect signals of recent selective sweeps with very low false discovery rate. I then applied four different computational approaches on the high-throughput resequencing data of a 2.1 Mbp segment of Drosophila melanogaster X chromosome to compare and discuss their performances. The study emphasized the relative advantages of linkage disequilibrium-based methods in detecting recent sweeps relative to site frequency-based approaches when applied on incomplete data. There are also many challenges in other applications of high-throughput sequencing, including discoveries of novel transcription active regions (TARs) in RNA-seq analysis. Here, I present a flexible statistical program, HPIBD (HMM-based Peak Identification and Boundary Definition) for de novo analysis of RNA-seq datasets. It avoids the use of arbitrary read-depth cutoffs and has built-in tolerance to read gaps. It is able to statistically make TARs predictions, estimate peak boundaries and evaluate the confidence in the prediction. I implemented the model and showed that HPIBD has robust performance under various validations and with benchmark to Cufflinks.
Population Genetics; Natural Selection; High-throughput Sequencing
Aquadro, Charles F
Clark, Andrew; Keinan, Alon
Ph. D., Genetics
Doctor of Philosophy
dissertation or thesis