Building your own organism models

Building your own organism models

If the organism of the sequence you are studying is not available on the initial web page and if no available organism model produces satisfactory results, you can directly build a model for your organism from know coding sequences. From the original web page, click on the text link below the organism selection and you will be redirected to the Markov model building page.

In the model building page, the user must first choose a name for the organism model he/she intend to build (you can choose any simple name that is not already in use, non standard characters will be replaced by underscores). Then the FASTA file containing the coding sequences should be either uploaded or pasted in the text area. The sequences should not contain internal in-frame STOP codons (final STOP codons are tolerated),

Since FrameD uses extended interpolated Markov models for building probabilistic models of coding sequences, the amount of sequence provided is not as crucial as in classical Markov models which may be prone to over-fitting. Five to ten kilobases of CDS provide a reasonable start but the more the better. All sequences submitted are deleted once the model is built. If the model building process runs smoothly, the new organism will be automatically available on the FrameD web site in the list of organism (you may have to reload FrameD main page to update this list).

Users that require that a temporary model becomes available as a permanent model should contact FrameD@ossau.toulouse.inra.fr.

Two options are available in the model building process:

• Build N-tolerant Markov model: will build an extended interpolated Markov model over the A,T,G,C,N alphabet. This is especially useful if you intend to use the model to predict coding regions in unfinished sequences such as EST or EST clusters. Even if this option is not selected, FrameD will still use an extended Markov model for sequences with ambiguous IUPAC codes, but this extension is simply done by averaging and is not as precise as if the option is selected.
• TGA is not STOP: will inform the Markov model builder that in-frame TGA should not be considered as STOP when sequences are filtered. This is not stored in the model file and you should still select the option when using FrameD.

Building your own organism models