To navigate this site, please click the ☰ symbol to the left.
My research focuses on the development of computational data analysis methods for genome biology and small molecule discovery. This includes discovery-oriented data mining projects, as well as algorithm and software development projects for data types from a variety of high-throughput technologies such as next generation sequencing (NGS), genome-wide profiling approaches and chemical genomics. As part of the multidisciplinary nature of my field, I frequently collaborate with experimental scientists on data analysis projects of complex biological networks. Another important activity is the development of integrated data analysis systems for the open source software projects R and Bioconductor. The following gives a short summary of a few selected projects in my group.
Human longevity is heritable, and statistically and biologically compelling genetic associations with longevity and age-related traits have been identified. The translation of these genetic associations into insights that can lead to pharmacological interventions designed to promote healthy aging requires an approach and infrastructure that integrates many genomic resources.
To address this challenge, the Longevity Genomics research group has been established, an NIA funded research project to develop an integrative genomic resource and infrastructure to develop translational strategies to promote human longevity. The infrastructure will include data from longitudinal cohort studies with genome-wide genotype and sequence data, computational methods for annotating genetic variants, information from tissue-specific expression quantitative trait locus (eQTL) studies, and datasets of chemical properties and protein targets of small molecule compounds.
More details about this project can be found here: longevitygenomics.org.
Tools for Analyzing Next Generation Sequence Data
systemPipeR: NGS workflow and report generation environment
systemPipeR is an R/Bioconductor package for building and running automated analysis workflows for a wide range of next generation sequence (NGS) applications. Important features include a uniform workflow interface across different NGS applications, automated report generation, and support for running both R and command-line software, such as NGS aligners or peak/variant callers, on local computers or compute clusters. Efficient handling of complex sample sets and experimental designs is facilitated by a consistently implemented sample annotation infrastructure.
Figure 1: Workflow design structure of systemPipeR.
Reference-Assisted Genome Assembly
De novo assemblies of genomes remain one of the most challenging applications in next-generation sequencing. Usually, their results are incomplete and fragmented into hundreds of contigs. Repeats in genomes and sequencing errors are the main reasons for these complications. With the rapidly growing number of sequenced genomes, it is now feasible to improve assemblies by guiding them with genomes from related species. This project introduces AlignGraph, an algorithm for extending and joining de novo-assembled contigs or scaffolds guided by closely related reference genomes (Bao et al., 2014). It aligns paired-end (PE) reads and preassembled contigs or scaffolds to a close reference. From the obtained alignments, it builds a novel data structure, called the PE multipositional de Bruijn graph. The incorporated positional information from the alignments and PE reads allows us to extend the initial assemblies, while avoiding incorrect extensions and early terminations. In our performance tests, AlignGraph was able to substantially improve the contigs and scaffolds from several assemblers.
Figure 2: Overview of AlignGraph algorithm. (A) shows AlignGraph in the context of common genome assembly workflows, and the one on the bottom (B) illustrates its three main processing steps.
Reference-Assisted Transcriptome Assembly
Owing to the complexity and often incomplete representation of transcripts in RNA-Seq libraries, the assembly of high-quality transcriptomes can be extremely challenging. To improve this, my group is developing algorithms for guiding these assemblies with genomic sequences of related organisms as well as reducing the complexity in NGS libraries. The software tools we have published for this purpose so far include SEED (Bao et al., 2011) and BRANCH (Bao et al., 2013). BRANCH is a reference assisted post-processing method for enhancing de novo transcriptome assemblies (Figure 3). It can be used in combination with most de novo transcriptome assembly software tools. The assembly improvements are achieved with help from partial or complete genomic sequence information. They can be obtained by sequencing and assembling a genomic DNA sample in addition to the RNA samples required for a transcriptome assembly project. This approach is practical because it requires only preliminary genome assembly results in form of contigs. Nowadays, the latter can be generated with very reasonable cost and time investments. In case the genome sequence of a closely related organism is available, one can skip the genome assembly step and use the related gene sequences instead. This type of reference assisted assembly approach provides many attractive opportunities for improving de novo NGS assemblies in the future by making use of the rapidly growing number of reference genome information available to us.
Figure 3: Outline of *BRANCH* algorithm published in Bao et al. 2013. (a) Read alignments against preassembled transcripts and closely related genomic reference. (b) Junction graph based on this alignment result. (c) Assembly of extended transcripts.
Modeling Gene Expression Networks from RNA-Seq and ChIP-Seq Data
As part of several collaborative research projects, my group has developed a variety of data analysis pipelines for profiling data from next generation sequencing projects (e.g. RNA-Seq and ChIP-Seq), microarray experiments and high-throughput small molecule screens. Most of the data analysis resources developed by these projects are described in the associated online manuals for next generation data analysis. Recent research publications of these projects include: Yang et al., 2013; Zou et al., 2013; Yadav et al., 2013; Yadav et al., 2011; Mustroph et al., 2009.
Software Resources for Small Molecule Discovery and Chemical Genomics
Software tools for modeling the similarities among drug-like small molecules and high-throughput screening data are important for many applications in drug discovery and chemical genomics. In this area we are working on the development of the ChemmineR environment (Cao et al., 2008; Backman et al., 2011). This modular software infrastructure consists currently of five R/Bioconductor packages along with a user-friendly web interface, named ChemMine Tools that is intended for non-expert users (Figures 4-5). The integration of cheminformatic tools with the R programming environment has many advantages for small molecule discovery, such as easy access to a wide spectrum of statistical methods, machine learning algorithms and graphic utilities. Currently, the ChemmineR toolkit provides utilities for processing large numbers of molecules, physicochemical/structural property predictions, structural similarity searching, classification and clustering of compound libraries and screening results with a wide spectrum of algorithms. More recently, we have developed for this infrastructure the fmcsR algorithm which is the first mismatch tolerant maximum common substructure search tool in the field (Wang et al., 2013). In our comparisons with related structure similarity search tools, fmcsR showed the best virtual screening (VS) performance.
Figure 4: ChemmineR small molecule modeling environment with its add-on packages and selected functionalities.
Figure 5: Selectivity Analysis with ChemmineR and bioassayR
Function Prediction of Gene and Protein Sequences
Computational methods for characterizing the functions of protein sequences play an important role in the discovery of novel molecular, biochemical and regulatory activities. To facilitate this process, we have developed the sub-HMM algorithm that extends the application spectrum of profile HMMs to motif discovery and active site prediction in protein sequences (Horan et al. 2010). Its most interesting utility is the identification of the functionally relevant residues in proteins of known and unknown function (Figure 6). Additionally, sub-HMMs can be used for highly localized sequence similarity searches that focus on shorter conserved features rather than entire domains or global similarities. As part of this study we have predicted a comprehensive set of putative active sites for all protein families available in the Pfam database which has become a valuable knowledge resource for characterizing protein functions in the future.
Figure 6: Illustration of the sub-HMM extraction process from conserved protein domains, here Pfam desaturase domain (PF00487).
- see Publication List