The top row shows all the possible symbols. Each subsequent row contains a score for the alignment of the symbol in that sequence position. eg: For position 1 symbol A scores 0, B scores -30, C scores -60, and symbol M scores 150; for position 3 symbol A scores -4, L scores 54 etc.
The highest scoring symbol in each row is taken as a representative symbol for that position in the sequence, and is copied into the first column. This 'consensus' shows the symbol with the smallest evolutionary distance from all the residues known to exist at that position.
The values under 'Gap' and 'Len' are percentages, and show how the application of gap penalties should vary at each position depending on whether the region has been observed to have gaps. For our example of three sequences, one of the sequences has a gap at the end, hence the 10% gap penalties in the last position (consensus symbol S).
The last row (starting *) shows the composition of the multiple alignment. In the first column the total number of A's in all the sequences of the alignment is shown.
The profile is constructed from a symbol comparison table (chapter 2-8, 2-9) using the multiple alignment to determine a weighting favouring some symbols. The result is a position-specific symbol comparison table. There are two different programs available to create a profile: PROFILEMAKE uses the PAM250 table, and PROFILEWEIGHT (a non-GCG program) uses the BLOSUM62 table.
Using PROFILEMAKE the L symbol scores 150 in position 5, but only 54 in position 2. The original alignment shows all three sequences with an L at position 5, but only one has L in position 3. Note that the L-M match in the PAM250 table (2-8) is a relatively good one (1.3, where a perfect match is 1.5).
The "consensus" sequence is defined very differently to the consensus produced by PRETTY. (So different, in fact, that a different term might be more suitable.)