NGSAlignerToDo < BETA

---++Specifications

---+++File Formats
   * Reads input as FASTQ
   * Alignments output as SAM

---+++General Needs
   * Works in letter space with RNA-Seq data
   * Works in colour space with <nop>SOLiD data
   * Fast
   * Small memory footprint
   * Align against reference with dynamic splice graph
   * Minimize the size of the driver (i.e. abstract functions into classes)
   * Feedback loop of discovered SNPs back into reference

---+++Read Specific Needs
   * Works with 75bp reads, up to 128bp
   * Rescue transcriptome reads with huge chunks of mismatched introns (e.g. =NNNIIIII=; =I= = intron)
   * Split transcriptome reads to align across an intron junction
   * Handle residual introns within exome reads (e.g. =NNNIINNN=)
   * Handle resitual introns flanking exome reads (i.e. =NNNNNNII=)
   * Prevent losing SNPs at the end of reads from clipping (e.g. =NNNNNXN=; =N= = base; =X= = SNP)

---+++Alignment Specific Needs
   * User-specified method of handling multimapped reads (i.e. read mapping equally well to multiple positions)
   * Supports extended CIGAR encoding (including P for Padding)
   * Confirm if traceback ever needs to return multiple best alignments

---+++Quality Specific Needs
   * Alignment Score (from alignment)
   * Mapping Quality (as defined by MAQ, using Mosaik's implementation)
   * Uniqueness (out of 100%, divided by number best mapped positions)
   * Fragment Quality (PE specific value, probability that mapped fragment (flanked by PE) location is correct/incorrect)

---+++Build and Complication
   * Have CMake detect CPU architecture and link appropriate libraries
   * Build as "Release" in ccmake for full optimization

---+++Miscellaneous Questions
   * Do we ever have to assume that the reference is circular (i.e. how to handle negative and out of range positions)
   * Do we ever have to support positive penalty values?
This topic: BETA > TipsAndTricks > WebHome > NGSAlignerProject > NGSAlignerToDo
Topic revision: r5 - 2010-05-20 - jujubix