====== 454 Presentation by Teri Mueller ======
This talk was hosted by Roche.  It was a general overview of some of the capabilities of the 454 platform and its bundled software.  [[http://www.my454.com|454 website]] - Access to FLX updates and documentation.  Register with UCSC account only.
  * Pre-talk comment: The GUI is accessible by X11 or VNC.
===== 454 Characteristics =====
  * 1 bead/1 DNA fragment:  filter will try to remove beads with >1 fragment (~20%)
  * 200 cycles per FLX Titanium run.  New machine does 400.
  * 1 flow:  single base.  Unit of measure for pyrosequencing, as bases get added by flow.
  * Amount of light ~ # of bases.
  * First 4 bases of each sequence is the key sequence.  The first 3 are used to normalize the amount of light for 1 base.
  * Library sequences:  TCAG standard / GATC rapid.  Do not mix them!  There are also control sequences (did not catch them)
  * 1 mil reads / 2 well plate.  Lane masks will decrease number of reads.
  * Quality statistics can be viewed in the BaseQualityMetrics and QualityFilterMetrics files.
  * 10 hr runs.  Processing can take up to 80 hrs running on the default computer.  This is mitigated by processing with a cluster.
  * 37 gb per run (~28 gb of raw images)
===== Software =====
  * Amplicon Variant Analyzer: for specific region analysis
  * GS Assembler: de Novo or Reference Mapper
  * GS Reporter
  * GS RunProcessor - Image/signal processing
    * Formats:
    * .cwf: raw image
    * .fna: fasta sequence file, header has a specific format.
    * .qual: quality scores (like Phred, but offset)
    * .sff: standard 454 format.  <sfffile -s> will split plate into runs.
===== Quality Filtering =====
  * Shotgun vs. Amplicon pipeline defaults
  * Keypass (read rejecting):
    * Checks for key sequence
    * ~20% rejection expected
  * Dot (read rejecting):
    * Checks for too many negative flows.
    * 3 successive negative flows or N>5% of last positive flow.
  * Mix (read rejecting):
    * Checks for too many positive flows.
    * Indication of more than one sequence in bead.
    * >70% positive reads.
  * Signal Intensity (read trimming):
    * Reduces size of read until <3% borderline reads.
  * Primer (read rejecting):
    * Discard overamplified short sequences.
  * Valley (read rejecting):
    * Discard scaled sum scores that are too close to the valleys between base count decision points.
    * Amplicon default: 4/700 0.57%
    * Shotgun default: 4/320 1.25%
  * Trim back (read trimming):
    * Like valley, except trims instead of discarding until ratio is acceptable.
    * Amplicon default: off
    * Shotgun default: on
  * Quality score trim (read trimming):
    * 40 base window: if error rate >1%, trim a base.
    * <40 bases, throws sequence away.
    * (Even unfiltered, quality scores will reflect low quality areas)
  * SFFtools can also perform screen-trimming with a screening db for known contaminants.
A good run: expect a read length mode ~500 and mean >300.  ~50% should pass filters.
===== de Novo Assembly =====
  * All assemblers use .sff files.
  * 3, 8 & 20kb paired end libraries.
  * Can accept fasta/qual reads.
  * 15-25X coverage best.
    * Low coverage: poor contig building
    * High coverage: may cause contig breaks
  * 4 bytes per __read__ base in RAM.