The answers suggesting using bio{awk,perl,python} are correct; doing even simple things like this yourself as a one-off is surprisingly fraught with danger (what's the right regex for an accession number? Well, it depends.)

Still, as a learning exercise this can be done pretty simply using bash+sed:

#!/bin/bash
APIBASE="http://eutils.ncbi.nlm.nih.gov/entrez/eutils/efetch.fcgi?db=nucleotide&rettype=gi"
NOMATCH="ID list is empty"
while read line
do
    acc=$( echo $line | sed  -E -e '/^[^>]/d' -e 's/>([[:alnum:]_]+\.[[:digit:]]+).*/\1/' )
    if [ -z "$acc" ]
    then
        echo $line
    else
        gi=$( curl "${APIBASE}&id=${acc}" 2> /dev/null )
        if [[ $gi =~ $NOMATCH ]]
        then
            echo $line
        else
            restofline=$( echo $line | sed -e 's/^>\(.*\)/\1/' )
            echo ">gi|${gi}|${restofline}"
        fi
    fi
done < "${1:-/dev/stdin}" > "${2:-/dev/stdout}"​

This will work with either BSD or GNU sed (e.g. mac or linux). Playing with a made-up fasta file:

$ cat sample.fasta
>NC_001477.1 foo bar baz
ATGAACAACCAACGGAAAAAGACGGGTCGACCGTCTTTCAATATGCTGAAACGCGCGAGAAACCGCGTGT
CAACTGTTTCACAGTTGGCGAAGAGATTCTCAAAAGGATTGCTTTCAGGCCAAGGACCCATGAAATTGGT
GATGGCTTTTATAGCATTCCTAAGATTTCTAGCCATACCTCCAACAGCAGGAATTTTGGCTAGATGGGGC
TCATTCAAGAAGAATGGAGCGATCAAAGTGTTACGGGGTTTCAAGAAAGAAATCTCAAACATGTTGAACA
>U49845.1 something somethingelse quux
GATCCTCCATATACAACGGTATCTCCACCTCAGGTTTAGATCTCAACAACGGAACCATTGCCGACATGAG
ACAGTTAGGTATCGTCGAGAGTTACAAGCTAAAACGAGCAGTAGTCAGCTCTGCATCTGAAGCCGCTGAA
GTTCTACTAAGGGTGGATAACATCATCCGTGCAAGACCAAGAACCGCCAATAGACAACATATGTAACATA
TTTAGGATATACCTCGAAAATAATAAACCGCCACACTGTCATTATTATAATTAGAAACAGAACGCAAAAA
TTATCCACTATATAATTCAAAGACGCGAAAAAAAAAGAACAACGCGTCATAGAACTTTTGGCAATTCGCG
>EASYAS123.45 not valid at all
GATCCTCCATATACAACGGTATCTCCACCTCAGGTTTAGATCTCAACAACGGAACCATTGCCGACATGAG
ACAGTTAGGTATCGTCGAGAGTTACAAGCTAAAACGAGCAGTAGTCAGCTCTGCATCTGAAGCCGCTGAA
GTTCTACTAAGGGTGGATAACATCATCCGTGCAAGACCAAGAACCGCCAATAGACAACATATGTAACATA
TTTAGGATATACCTCGAAAATAATAAACCGCCACACTGTCATTATTATAATTAGAAACAGAACGCAAAAA
TTATCCACTATATAATTCAAAGACGCGAAAAAAAAAGAACAACGCGTCATAGAACTTTTGGCAATTCGCG​

You can run this as

$ ./gi-from-accession.sh sample.fasta new.fasta

or

$ cat sample.fasta | ./gi-from-accession.sh > newer.fasta

and get

$ cat newer.fasta
>gi|9626685|NC_001477.1 foo bar baz
ATGAACAACCAACGGAAAAAGACGGGTCGACCGTCTTTCAATATGCTGAAACGCGCGAGAAACCGCGTGT
CAACTGTTTCACAGTTGGCGAAGAGATTCTCAAAAGGATTGCTTTCAGGCCAAGGACCCATGAAATTGGT
GATGGCTTTTATAGCATTCCTAAGATTTCTAGCCATACCTCCAACAGCAGGAATTTTGGCTAGATGGGGC
TCATTCAAGAAGAATGGAGCGATCAAAGTGTTACGGGGTTTCAAGAAAGAAATCTCAAACATGTTGAACA
>gi|1293613|U49845.1 something somethingelse quux
GATCCTCCATATACAACGGTATCTCCACCTCAGGTTTAGATCTCAACAACGGAACCATTGCCGACATGAG
ACAGTTAGGTATCGTCGAGAGTTACAAGCTAAAACGAGCAGTAGTCAGCTCTGCATCTGAAGCCGCTGAA
GTTCTACTAAGGGTGGATAACATCATCCGTGCAAGACCAAGAACCGCCAATAGACAACATATGTAACATA
TTTAGGATATACCTCGAAAATAATAAACCGCCACACTGTCATTATTATAATTAGAAACAGAACGCAAAAA
TTATCCACTATATAATTCAAAGACGCGAAAAAAAAAGAACAACGCGTCATAGAACTTTTGGCAATTCGCG
>EASYAS123.45 not valid at all
GATCCTCCATATACAACGGTATCTCCACCTCAGGTTTAGATCTCAACAACGGAACCATTGCCGACATGAG
ACAGTTAGGTATCGTCGAGAGTTACAAGCTAAAACGAGCAGTAGTCAGCTCTGCATCTGAAGCCGCTGAA
GTTCTACTAAGGGTGGATAACATCATCCGTGCAAGACCAAGAACCGCCAATAGACAACATATGTAACATA
TTTAGGATATACCTCGAAAATAATAAACCGCCACACTGTCATTATTATAATTAGAAACAGAACGCAAAAA
TTATCCACTATATAATTCAAAGACGCGAAAAAAAAAGAACAACGCGTCATAGAACTTTTGGCAATTCGCG

Hello,

I just had a similar issue with a large multi fasta file, and I was able to solve it using FaBox online.

Not a very good exercise in coding, but great for those of us that are perhaps a bit more lazy!

http://users-birc.au.dk/biopv/php/fabox/

-Alice

As Devon Ryan said, use BioPython or BioPerl.

Workflow:

For each sequence in input file,

1. Pick up the header
2. Query the URL with the accession num from the header and add GI to create new header string.
3. Write out above created header string
4. Write out sequence

As mentioned, there are easier ways to do this using Bioperl / Biopython - take a look at them, it is definitely worth your time.

Using bash alone, this should work in principle (i.e., not tested):

#!/bin/sh
FASTA=~/test/sequences.fa # original file containing sequences or accession ids
ACCESSION=$(grep -o -E "^>\w+" $FASTA | tr -d ">") # extract accession id
echo "accession is $ACCESSION" # for visual inspection (optional)
curl http://eutils.ncbi.nlm.nih.gov/entrez/eutils/efetch.fcgi?db=nucleotide&id=$ACCESSION&rettype=fasta > ~/test/$ACCESSION.fasta