Next Generation Sequencing (NGS) is a powerful tool to gain new insights in molecular biology. With the introduction of the first bench top NGS sequencing machines (e.g. Ion Torrent, MiSeq), this technology became even more versatile in its applications and the amount of data that are produced in a short time is ever increasing. The demand for new and more efficient sequence analysis tools increases at the same rate as the throughput of sequencing technologies. New methods and algorithms not only need to be more efficient but also need to account for a higher genetic variability between the sequenced and annotated data. To obtain reliable results, information about errors and limitations of NGS technologies should also be investigated. Furthermore, methods need to be able to cope with contamination in the data. In this thesis we present methods and algorithms for NGS analysis. Firstly, we present a fast and precise method to align NGS reads to a reference genome. This method, called NextGenMap, was designed to work with data from Illumina, 454 and Ion Torrent technologies, and is easily extendable to new upcoming technologies. We use a pairwise sequence alignment in combination with an exact match filter approach to maximize the number of correctly mapped reads. To reduce runtime (mapping a 16x coverage human genome data set within hours) we developed an optimized banded pairwise alignment algorithm for NGS data. We implemented this algorithm using high performance programing interfaces for central processing units using SSE (Streaming SIMD Extensions) and OpenCL as well as for graphic processing units using OpenCL and CUDA. Thus, NextGenMap can make maximal use of all existing hardware no matter whether it is a high end compute cluster or a standard desktop computer or even a laptop. We demonstrated the advantages of NextGenMap based on real and simulated data over other mapping methods and showed that NextGenMap outperforms current methods with respect to the number of correctly mapped reads. The second part of the thesis is an analysis of limitations and errors of Ion Torrent and MiSeq. Sequencing errors were defined as the percentage of mismatches, insertion and deletions per position given a semi-global alignment mapping between read and reference sequence. We measured a mean error rate for MiSeq of 0.8\% and for Ion Torrent of 1.5\%. Moreover we identified for both technologies a non-uniform distribution of errors and even more severe of the corresponding nucleotide frequencies given a difference in the alignment. This is an important result since it reveals that some differences (e.g. mismatches) are more likely to occur than others and thus lead to a biased analysis. When looking at the distribution of the reads accross the sample carrier of the sequencing machine we discovered a clustering of reads that have a high difference ($> 30\%$) compared to the reference sequence. This is unexpected since reads with a high difference are believed to origin either from contamination or errors in the library preparation, and should therefore be uniformly distributed on the sample carrier of the sequencing machine. Finally, we present a method called DeFenSe (Detection of Falsely Aligned Sequences) to detect and reduce contamination in NGS data. DeFenSe computes a pairwise alignment score threshold based on the alignment of randomly sampled reads to the reference genome. This threshold is then used to filter the mapped reads. It was applied in combination with two widely used mapping programs to real data resulting in a reduction of contamination of up to 99.8\%. In contrast to previous methods DeFenSe works independently of the number of differences between the reference and the targeted genome. Moreover, DeFenSe neither relies on ad hoc decisions like identity threshold or mapping quality thresholds nor does it require prior knowledge of the sequenced organism. The combination of these methods may lead to the possibility of transferring knowledge from model organisms to non model organisms by the usage of NGS. In addition, it enables to study biological mechanisms even in high polymorphic regions.

Next Generation Sequencing (NGS) is a powerful tool to gain new insights in molecular biology. With the introduction of the first bench top NGS sequencing machines (e.g. Ion Torrent, MiSeq), this technology became even more versatile in its applications and the amount of data that are produced in a short time is ever increasing. The demand for new and more efficient sequence analysis tools increases at the same rate as the throughput of sequencing technologies. New methods and algorithms not only need to be more efficient but also need to account for a higher genetic variability between the sequenced and annotated data. To obtain reliable results, information about errors and limitations of NGS technologies should also be investigated. Furthermore, methods need to be able to cope with contamination in the data. In this thesis we present methods and algorithms for NGS analysis. Firstly, we present a fast and precise method to align NGS reads to a reference genome. This method, called NextGenMap, was designed to work with data from Illumina, 454 and Ion Torrent technologies, and is easily extendable to new upcoming technologies. We use a pairwise sequence alignment in combination with an exact match filter approach to maximize the number of correctly mapped reads. To reduce runtime (mapping a 16x coverage human genome data set within hours) we developed an optimized banded pairwise alignment algorithm for NGS data. We implemented this algorithm using high performance programing interfaces for central processing units using SSE (Streaming SIMD Extensions) and OpenCL as well as for graphic processing units using OpenCL and CUDA. Thus, NextGenMap can make maximal use of all existing hardware no matter whether it is a high end compute cluster or a standard desktop computer or even a laptop. We demonstrated the advantages of NextGenMap based on real and simulated data over other mapping methods and showed that NextGenMap outperforms current methods with respect to the number of correctly mapped reads. The second part of the thesis is an analysis of limitations and errors of Ion Torrent and MiSeq. Sequencing errors were defined as the percentage of mismatches, insertion and deletions per position given a semi-global alignment mapping between read and reference sequence. We measured a mean error rate for MiSeq of 0.8\% and for Ion Torrent of 1.5\%. Moreover we identified for both technologies a non-uniform distribution of errors and even more severe of the corresponding nucleotide frequencies given a difference in the alignment. This is an important result since it reveals that some differences (e.g. mismatches) are more likely to occur than others and thus lead to a biased analysis. When looking at the distribution of the reads accross the sample carrier of the sequencing machine we discovered a clustering of reads that have a high difference ($> 30\%$) compared to the reference sequence. This is unexpected since reads with a high difference are believed to origin either from contamination or errors in the library preparation, and should therefore be uniformly distributed on the sample carrier of the sequencing machine. Finally, we present a method called DeFenSe (Detection of Falsely Aligned Sequences) to detect and reduce contamination in NGS data. DeFenSe computes a pairwise alignment score threshold based on the alignment of randomly sampled reads to the reference genome. This threshold is then used to filter the mapped reads. It was applied in combination with two widely used mapping programs to real data resulting in a reduction of contamination of up to 99.8\%. In contrast to previous methods DeFenSe works independently of the number of differences between the reference and the targeted genome. Moreover, DeFenSe neither relies on ad hoc decisions like identity threshold or mapping quality thresholds nor does it require prior knowledge of the sequenced organism. The combination of these methods may lead to the possibility of transferring knowledge from model organisms to non model organisms by the usage of NGS. In addition, it enables to study biological mechanisms even in high polymorphic regions.