So what this did was formulate the questions of the code for the first time in a really serious manner, and in this one… that is the code was in this case postulated to be degenerate, as we say. That isn't something bad, it just means that there are several instances… several triplets correspond to a given amino acid. And the question then arose is whether you could decode the gene, now, from the existing small database of amino acid sequences that was available. At that time we had fragments… we had the whole of insulin which Fred Sanger had done, and we had a database of small pieces of other proteins that various people had done, sometimes really only two or three amino acids in a row. Now I had… if you look at the Gamow Code, then you could actually show… and in fact Francis so far as I know was the first to have done this, you could actually show that the particular code that Gamow had proposed could not correspond to the known amino acid sequences, but then it became realised that we could assign these triplets in any way that we pleased, and so how would you show that any code of this form was not impossible? In fact, a little later in the thing a computer specialist wrote a paper which showed that, in order to eliminate all overlapping triplet codes, we should have started with a computer about a million times more powerful than the one we had now and we should have started working in the last days of Rome, that is 2000 years ago in order to have accomplished the problem, you see… just so many possibilities. But I had already seen that there was a way actually to… to eliminate all overlapping triplet codes and I first had a statistical proof of this. Because you can see if it's an overlapping triplet, that is, you have the first three letters of the gene correspond to an amino acid, so let us say ABB, and then you... you move one ahead and then it would say BBC would correspond to the second, you can see that its four letters correspond to a dipeptide, two amino acids. Now if you look, the number of possibilities for four letters is 256, 4⁴, but the number of possibilities for dipeptides is 400. This meant that if it was an overlapping triplet code of any form we could not have all the dipeptides. Dipeptides would be excluded. All right. So what you have to show is that the present data would allow all dipeptides, then you could eliminate us. Now of course there wasn't enough data to do this. However, there was enough data to do a Poisson analysis, which I did, and I had this big Poisson sheet in Cold Spring Harbor when I arrived there, and I had already shown that if you just did this… saw the present sequences as a statistical sample of everything, it was unlikely.
