I am a postdoctoral scholar at Department of Statistics, Stanford University. I received my B.S degree in Biological Sciences from Tsinghua University (Beijing) in 2004. I received my doctorate degree in 2009 on Structural Biology, Oxford University, under Prof. David I. Stuart and Dr. Robert Esnouf's guidance. At the same year, I received my master degree in Statistics at Stanford University and continued my research on biostatistics/bioinformatics in Prof. Wing H. Wong's Lab till present. My current research interests focus on the following areas:
- RNA-seq (Alternative Splicing, isoform detection)
- Third Generation Sequencing
- Stem cell transcriptome analysis
- Proteomics (Protein identification and novel splice detections from high-throughput Mass Spec)
- Gene regulatory network
- RNA editing
Characterization of the human ESC transcriptome by hybrid sequencing
Au, Kin Fai, Sebastiano V., Afshar P.T., Durruthy J.D., Lee L., Williams B.A., Bakel H.V., Schadt E., Pera R.A.R., Underwood J., Wong W.H.
In press [preprint]
Oct4-Sall4-Nanog network controls developmental progression in the preimplantation mouse embryo.
Tan, M.H.*, Au, Kin Fai*, Leong, D. E., Foygel K., Wong, W.H., Yao, M.W.M.
Molecular Systems Biology 2012. 632 doi:10.1038/msb.2012.65.[preprint]
* These authors contributed equally to this work
RNA sequencing reveals diverse and dynamic repertoire of the Xenopus tropicalis transcriptome over development.
Tan, M.H.*, Au, Kin Fai*, Yablonovitch, A.L.*, Wills, A.E., Baker J.C., Wong, W.H., Li, J.B. Genome Research, 2012. doi:10.1101/gr.141424.112 [preprint]
* These authors contributed equally to this work
Activation of Innate Immunity is Required for Efficient Nuclear Reprogramming
Lee, J., Sayed, N., Hunter, A., Au, Kin Fai, Wong, W.H., Mocarski, E., Pera, R.R., Yakubov, E., Cooke, J.P. Cell, 2012 Oct; 151(3): 547-58. [preprint]
Detection of splice junctions from paired-end RNA-seq data by SpliceMap
Au, Kin Fai, Jiang, H., Lin, L., Xing, Y., Wong, W.H. Nucleic Acids Research, Aug;38(14):4570-8. 2010. [preprint]
12-01-2013: Faster and much less memory-required LSC 1.alpha is released
In the LSC 0.3.0 or 0.3.1, we optimized the setting of bowtie2 and BWA to get much more short read alignment, which improve the the accuracy of error correction a lot/ However, the increase of alignments also requires much more running time (on both alignment and the following error correction step) and memory usage. Therefore, a few users met difficulty of running LSC 0.3.0 or 0.3.1.
In LSC 1.alpha, we apply probabilistic algorithm ("SCD" option) to select ""enough" short read alignment for error correction. LSC 1.alpha does NOT sacrifice the error correction performace (sensitivity and specificity). Please see http://www.stanford.edu/~kinfai/LSC/LSC_manual.html#aligner Thus, we save running time and memory usage significantly. The running time is 30-50% of LSC 0.3.1. The peak memory usage decreases to ~10G regardless of the data size.
- Added probabilistic algorithm ("SCD" option) to pre-select SR alignments results based on LR-SR alignment coverage depth (Significant improvement in running time and memory usage)
- Removed requirement for loading SR dataset in memory to generate LR-SR mapping file (Significant improvement in memory usage)
- Added option "sort_max_mem" in run.cfg to control maximum memory used by unix sort command to avoid unexpected Mem crash
- Fixed a bug in generating FASTQ file (it affected some of QualityValue computation results)
11-26-2013 - IDP 0.1 and the manual and a tutorial are released
IDP integrates short reads (e.g. Illumina data) and long reads (e.g. PacBio data) to identify gene isoforms (transcripts) from transcriptome (see Figure above).
- One input of IDP is the short-read RNA-seq results: junctions (bed file) AND alignments of short reads (sam file).
Most RNA-seq tools, such as SpliceMap and Tophat can output these two files.
- The other input is the long reads: raw sequences (FASTA file) OR alignment of long reads (PSL file by BLAT or GPD file)
The error-corrected long reads from PacBio data is perferred. LSC is our default error-correction tool.
- The IDP output are the gene isoform identifications and quantification of genes and gene isoforms. hESC transcriptome (H1 cell line) is the first one identified by this methods. For more details of this transcriptome, please see its homepage http://www.stanford.edu/~kinfai/IDP/hESC.html and our paper Characterization of the human ESC transcriptome by hybrid sequencing [preprint].
11-26-2013: Hompage of hESC transcriptome identified by SpliceMap-LSC-IDP pipline is released.The homepage of hESC transcriptome (H1 cell line) is released. You can also find novel genes, novel isoforms of existing genes (including pluripency markers) and novel ncRNA in this website:
The details of this hESC transcriptome can be in our publication: Characterization of the human ESC transcriptome by hybrid sequencing [preprint]
11-26-2013: IDP and hESC transcriptome paper is releasedKin Fai Au, Vittorio Sebastiano, Pegah Tootoonchi Afshar, Jens Durruthy Durruthy, Lawrence Lee, Brian A. Williams, Honoratus Van Bakel, Eric Schadt, Renee A. Reijo Pera, Jason Underwood, Wing Hung Wong
Characterization of the human ESC transcriptome by hybrid sequencing [preprint]
09-30-2013: More robust and faster LSC 0.3.1
In LSC 0.3.1, we don't have pseudo chromosome, the alignment time reduced to ~10% (in Bowtie2 mode). And you can re-run some crashed jobs easily now.
- Remove pseudo-chr processing
- Accept compressed SR as input (should be named SR.fa.cps/SR.fa.cps.idx in any folder)
- Added "runLSC -cleanup" option to remove redundant files (per thread split, remaining _tmp files) if the run was successful at the end.
- Changed convertNav to sort reads and then generate LR_SR.map (memory optimization instead of loading all alignments in memory)
- Changed "print" to system.echo (messages were not printed out in qsub output files)
- Changed a little bit "cleanup" option to keep per thread data (*.aa, *.ab, ..). It was useful when one thread was crashed and we wanted to just re-run that at the end
08-07-2013: Big changes in LSC 0.3
In LSC 0.3, we have a few updates. They are very IMPORTANT updates, new features and small fixes
Very IMPORTANT updates:
- Support for Bowtie2 and RazerS3 as initial aligners. Now, BWA, Bowtie2, RazerS3 and Novoalign work in LSC. Please see the comparison details of aligners in the "Short read - Long read aligner#manual".
Added SR length coverage percentage on LR (SR-covered length/full length of corrected LR) to corrected_LR output file. Here is an example, where the last number 0.82 is the SR length coverage percentage on LR:
- Added support for three modes for step-wise runs:
- Generating FASTQ output format based on correction probability given short read coverage. Please refer to LSC paper and manual page for more details. You can select well-corrected reads for downstream analyses by using the quality in FASTQ output or SR length coverage percentage above. Please the the filtering in the "Output#manual".
- mode 0: end-to-end
- mode 1: generating LR_SR.map file
- mode 2: correction step
- Used the python path in the cfg file instead of default user/bin path
- Added option (-clean_up) to remove intermediate files or not (Note: important/useful ones will still be there in temp folder)
- Support for input fastq format for LR (long reads) and/or SR (short reads)
- Updated default BWA and novoalign commands options
- Printing out original LR names in the output file
- Support for printing out version number using -v/-version option
Small bug fixed
- Fixed in removing XZ pattern printed out at the end of some uncorrected_LR sequences
- Fixed samParser bug (which was ignoring some valid alignments in BWA output)