Welcome to GRAIL (Gene Recognition and Analysis Internet Link)
Grail is an interface to a system which will ultimately provide
automated gene assembly from DNA sequence data. Currently the
system provides analysis of protein coding potential of a DNA
sequence. The coding recognition module (CRM) uses a multiple-
sensor neural network approach to identify coding exons than are
at least 100 bases long. In its current configuration the CRM
identifies 90% of such regions with less than 1 false positive
coding exon per 5 coding exons indicated. Your success rate will
depend on a number of parameters including the G/C content of
your sequence. In general, coding regions in sequences of low
G/C content are not as well recognized as those in higher G/C.
Investigation is underway to try and improve the performance
for low G/C sequences.
This part of the system is specifically designed to locate
regions of DNA sequence with protein encoding potential. The
system has been trained to recognize coding regions in Human DNA
but seems to work well on DNA sequences from other mammals.
Because the system has not been tested extensively on species
other than human, no claims are made for the predictions of
coding potential on DNA's from other species.
To use GRAIL you must first register and get a user ID.
To become a registered user please send the following
e-mail message to:
grail@ornl.gov
Register
Your Name
Your address
Your phone number
your E-mail address
To have sequences analyzed send e-mail to:
grail@ornl.gov
The message will start with the word "sequences" followed by the
number of sequences you are sending followed by your user ID
followed by the sequences you wish to have analyzed in the
following format:
Sequences number_of_sequences your_user_ID
>seq1name
AAAAAAAA........
>seq2name
TTTTTTTT..........
etc.
For the system to return any interpretation the sequence to be
analyzed must be at least 100 bases long (and not more than
100kb). For each sequence the following information will be
returned:
1. The score for the coding potential for each position analyzed
on each strand (the f-(forward) strand represents the sequence as
received, and the r-(reverse) strand is the reverse compliment).
These scores range from 0.0 to 1.0 and a score greater than 0.5
identifies a region with protein encoding potential. Non-coding
regions often have a score of 0.000. To reduce the output, only
regions with scores of at least 0.01 are reported.
2. frame. In calculating the coding potential, the system
calculates the reading-frame which is "preferred" in the window
over which the calculation is done and this information is
returned for regions with scores over 0.5.
3. orf. The limits between which the preferred frame is open is
returned for windows with scores over 0.5.
The second part of the output is the system's interpretation of
the raw data. This output gives the limits (in general a minimum)
of the extent of the coding exon, the most likely strand for the
exon with a probability for the correctness of the strand
assignment, the preferred reading frame for the exon and a
quality assessment. An interesting phenomenon we have noted
is that some exons seem to have coding character on both strands
or even more coding character on the wrong strand. be aware that strand
assignments are not always correct, and it is sometimes useful to
consider both strands as possible. Any exon with a quality score of
"excellent" is worth further consideration. Please remember that
the system is designed to find coding exon of 100 or more bases,
so small coding exons may well be missed.
This implementation of the CRM has been tested on a set of human
genes containing 102kb of sequence. This set contained 70 coding
exons and the system identified 62 (89%) and assigned them all to
the correct strand. (Though in a larger test set strand assignment
was 90-95% correct). The preferred reading frame assignment was
correct for 60 (96%) of these exons while the frame assignment
for the other two had some ambiguity. Of the eight missed 6 were
less than 100 bases long. Of 43 predicted exons with a quality
score of "excellent" all were actual coding exons. Of predicted
exons scoring "good" 11 of 16 (69%) were expected and of 49
predicted exons with a score of "marginal" only 8 (16%) were
"real". Though this is a rather limited test set, the results
of this analysis give some guidance for interpreting CRM output.
N.B. This is an alpha+ version so we are open to feed-back.
We have a new e-mail address called GRAILMAIL@ORNL.GOV
for user feedback to the GRAIL staff. Or communication can be
addressed specifically to us:
Direct questions to: Richard J. Mural, e-mail:
m9l@stc10.ctd.ornl.gov
Phone: 615-576-2938
or
Edward C. Uberbacher, e-mail:
uber@msr.epm.ornl.gov
Phone: 615-574-6134
or
GRAIL staff, e-mail:
grailmail@ornl.gov
To receive a copy of this help file send the message "help" to
grail@ornl.gov.
-------------------------------------------------------------
Appendix A: GRAIL updates
-------------------------------------------------------------
Modifications to the GRAIL rule base for constructing the exon
table from the coding probability information have been made as
of Feb. 19, 1992. These changes have been designed to recognize
situations where a single real exon, usually with significant
extent, is recognized by GRAIL as multiple peaks or multiple exons.
These additional rules interconnect predicated peaks under
circumstances where consecutive predicated regions have the same
preferred reading frame, the frame is open between them, and they
are relatively close together. The result is generally a beneficial
simplification of the exon table and a more accurate representation
of exon structure. This also better adapts GRAIL for use with cDNAs.
Feedback or questions can be addressed to GRAILMAIL@ornl.gov.
The GRAIL staff
________________________________________