Spaced seeds have been extensively studied in the homology search field. A spaced seed can be regarded as a very special type of hash function on k-mers, where two k-mers have the same hash value if and only if they are identical at the w (w <k) positions designated by the seed. Spaced seeds substantially increased the homology search sensitivity. It is then a natural question to ask whether there is a better hash function (called hash seed) that provides better sensitivity than the spaced seed. We study this question in the paper. We propose a strategy to classify amino acids, which leads to a better hash seed. Our results raise a new question about how to design the best hash seed.