<s>
HMMER	B-Language
is	O
a	O
free	B-Application
and	O
commonly	O
used	O
software	O
package	O
for	O
sequence	O
analysis	O
written	O
by	O
Sean	O
Eddy	O
.	O
</s>
<s>
Profile-HMMs	O
are	O
constructed	O
from	O
a	O
multiple	O
sequence	O
alignment	O
in	O
the	O
HMMER	B-Language
package	O
using	O
the	O
hmmbuild	O
program	O
.	O
</s>
<s>
The	O
profile-HMM	O
implementation	O
used	O
in	O
the	O
HMMER	B-Language
software	O
was	O
based	O
on	O
the	O
work	O
of	O
Krogh	O
and	O
colleagues	O
.	O
</s>
<s>
HMMER	B-Language
is	O
a	O
console	B-Application
utility	O
ported	O
to	O
every	O
major	O
operating	B-General_Concept
system	I-General_Concept
,	O
including	O
different	O
versions	O
of	O
Linux	B-Application
,	O
Windows	B-Application
,	O
and	O
Mac	B-Application
OS	I-Application
.	O
</s>
<s>
HMMER	B-Language
is	O
the	O
core	O
utility	O
that	O
protein	O
family	O
databases	O
such	O
as	O
Pfam	O
and	O
InterPro	O
are	O
based	O
upon	O
.	O
</s>
<s>
Some	O
other	O
bioinformatics	O
tools	O
such	O
as	O
UGENE	B-Language
also	O
use	O
HMMER	B-Language
.	O
</s>
<s>
HMMER3	O
also	O
makes	O
extensive	O
use	O
of	O
vector	B-Device
instructions	I-Device
for	O
increasing	O
computational	O
speed	O
.	O
</s>
<s>
This	O
work	O
is	O
based	O
upon	O
earlier	O
publication	O
showing	O
a	O
significant	O
acceleration	O
of	O
the	O
Smith-Waterman	B-Algorithm
algorithm	I-Algorithm
for	O
aligning	O
two	O
sequences	O
.	O
</s>
<s>
Capturing	O
this	O
information	O
gives	O
them	O
a	O
better	O
ability	O
to	O
detect	O
true	O
homologs	O
than	O
traditional	O
BLAST-based	O
approaches	O
,	O
which	O
penalise	O
substitutions	O
,	O
insertions	O
and	O
deletions	O
equally	O
,	O
regardless	O
of	O
where	O
in	O
an	O
alignment	O
they	O
occur	O
.	O
</s>
<s>
The	O
HMMER	B-Language
package	O
consists	O
of	O
a	O
collection	O
of	O
programs	O
for	O
performing	O
functions	O
using	O
profile	O
hidden	O
Markov	O
models	O
.	O
</s>
<s>
In	O
addition	O
to	O
the	O
software	O
package	O
,	O
the	O
HMMER	B-Language
search	O
function	O
is	O
available	O
in	O
the	O
form	O
of	O
a	O
web	O
server	O
.	O
</s>
<s>
Major	O
reasons	O
for	O
relocating	O
the	O
web	O
service	O
were	O
to	O
leverage	O
the	O
computing	O
infrastructure	O
at	O
the	O
EBI	O
,	O
and	O
to	O
cross-link	O
HMMER	B-Language
searches	O
with	O
relevant	O
databases	O
that	O
are	O
also	O
maintained	O
by	O
the	O
EBI	O
.	O
</s>
<s>
The	O
latest	O
stable	O
release	O
of	O
HMMER	B-Language
is	O
version	O
3.0	O
.	O
</s>
<s>
A	O
major	O
aim	O
of	O
the	O
HMMER3	O
project	O
,	O
started	O
in	O
2004	O
was	O
to	O
improve	O
the	O
speed	O
of	O
HMMER	B-Language
searches	O
.	O
</s>
<s>
While	O
profile	O
HMM-based	O
homology	O
searches	O
were	O
more	O
accurate	O
than	O
BLAST-based	O
approaches	O
,	O
their	O
slower	O
speed	O
limited	O
their	O
applicability	O
.	O
</s>
<s>
The	O
main	O
performance	O
gain	O
is	O
due	O
to	O
a	O
heuristic	B-Algorithm
filter	I-Algorithm
that	O
finds	O
high-scoring	O
un-gapped	O
matches	O
within	O
database	O
sequences	O
to	O
a	O
query	O
profile	O
.	O
</s>
<s>
This	O
heuristic	B-Algorithm
results	O
in	O
a	O
computation	O
time	O
comparable	O
to	O
BLAST	B-Application
with	O
little	O
impact	O
on	O
accuracy	O
.	O
</s>
<s>
Further	O
gains	O
in	O
performance	O
are	O
due	O
to	O
a	O
log-likelihood	B-General_Concept
model	O
that	O
requires	O
no	O
calibration	O
for	O
estimating	O
E-values	B-General_Concept
,	O
and	O
allows	O
the	O
more	O
accurate	O
forward	O
scores	O
to	O
be	O
used	O
for	O
computing	O
the	O
significance	O
of	O
a	O
homologous	O
sequence	O
.	O
</s>
<s>
HMMER	B-Language
still	O
lags	O
behind	O
BLAST	B-Application
in	O
speed	O
of	O
DNA-based	O
searches	O
;	O
however	O
,	O
DNA-based	O
searches	O
can	O
be	O
tuned	O
such	O
that	O
an	O
improvement	O
in	O
speed	O
comes	O
at	O
the	O
expense	O
of	O
accuracy	O
.	O
</s>
<s>
HMMER	B-Language
sequence	O
alignments	O
are	O
accompanied	O
by	O
posterior	O
probability	O
annotations	O
,	O
indicating	O
which	O
portions	O
of	O
the	O
alignment	O
have	O
been	O
assigned	O
high	O
confidence	O
and	O
which	O
are	O
more	O
uncertain	O
.	O
</s>
<s>
While	O
HMMER2	O
could	O
perform	O
local	B-Algorithm
alignment	I-Algorithm
(	O
align	O
a	O
complete	O
model	O
to	O
a	O
subsequence	O
of	O
the	O
target	O
)	O
and	O
global	O
alignment	O
(	O
align	O
a	O
complete	O
model	O
to	O
a	O
complete	O
target	O
sequence	O
)	O
,	O
HMMER3	O
only	O
performs	O
local	B-Algorithm
alignment	I-Algorithm
.	O
</s>
