NB. There are disadvantages to using Latin transliteration to input Greek -- especially in searching corpora that mix Latin and Greek texts, where Diogenes may not be able to guess which language you mean. Unicode input is preferred.
If you work with the transliteration employed natively by the TLG, the so-called BETA code, your searches must stipulate accents.
In this scheme, you must indicate accents and breathings in the normalized form used by the TLG word list: do not indicate upper-case letters, and do not use barytone accents. You may use lower-case Latin letters. Thus the Attic nominative singular of Herakles would be " h(raklh=s ", and the oblique and variant forms would be found with " h(rakl". Remember that words appearing in all upper-case Greek will not usually have dicritics, so to pick up titles, etc., you will also need to try " hraklhs " (note the leading and trailing spaces to indicate the start and end of a word). You may not use Perl regular expression metacharacters, because many of them are significant in Beta code; your pattern will be turned into a regular expression that allows hyphenation to intervene.
Here are the full details of the Perseus and Beta transliteration schemes: