====== Retention Time Alignment ======



===== Major steps followed by the Retention Time Alignment algorithm =====

{{ :userguide:rtalignmentconcept2.png|}}
  - Collect species from the reference User context and predict their retention time using an external utility and store them in reference species properties
  - For each identification gathered under the reference User context do:
    - Collect species to align from the identification 
    - Constitute several pair groups between collected reference species & species to align. A pair group contains 2 groups of species having same sequence & calculated mass (group1 has species from reference User context and group2 has species to align from identifications).
    - Compute one (or several) representative value(s) for group1 & group2 for each pair group
    - Compute linear regression between representative values 
    - Store linear regression coefficients & reference context name in identification properties

===== In more details... =====

{{ RtAlignmentConcept1.PNG}}
  - Species Retention Time of the reference UserContext are predicted with **NETPrediction v2.2.3378** utility using Kangas method (click [[http://omics.pnl.gov/software/NETPredictionUtility.php|here]] for more details). NETPrediction utility only uses species **sequences** to predict a Normalized Elution Time (NET) value. 
    - First, a list of '//reference//' species is built
         * The reference species list doesn't contain any species with PTMs
         * If several species exist with redundant sequences, the best score species is retained
    - Then, the corresponding sequences are exported in order to be used by the NETPrediction utility
  - The predicted NET are converted to retention time using user-defined parameters (duration & delay). 
  - The user can decide to exclude predicted values too far from the others: 
    * The average absolute deviation (between RT & predicted RT) is computed and all predicted RT far from this average value about a given user-defined threshold are excluded.
  - The predicted RT values are then stored as properties in the reference species.
  - For each identification existing under the reference User context (not necessary directly under the reference user context):
    -  All the final child species to align are collected
    -  Several pair groups are created using reference species & species to align. A pair group contains 2 groups of species having same sequence & calculated mass (group1 has species from reference User context and group2 has species to align from identifications).
    - For each group, according the statistical method choosen by the user, one or several representative RT value(s) is(are) calculated.
      * Each 'Group1' always contains one species because [[userguide:proteingroups|species/protein grouping]] has been executed on the reference context and suppressed sequence/calculated mass redundancy.
      * 'Group2' may contain one or more species. One or several representative value(s) is(are) calculated from one of following statistical methods:
        * Average {{ :userguide:rtalignmentconcept3.png|}}
        * Median
        * Cross-product
        * Combined
        * Best score
    - Compute linear regression between pair groups
    - Store linear regression coefficients (slope/intercept) + reference context name in the identification properties