A hidden variable problem?

Hi everyone, I have a new problem in my hands and dont know the best way to approach it. I have the mRNA expression levels for different genes of interest: a, b, c, d, e, f... my hypothesis is that at the genes 'team up' at the level of protein function so that [A+B+C] vs [D+E+F] have a higher R-squared than any of the isolated comparisons [X] vs [Y], where X:A,B,or C and Y:D,E, or F (observed: a,b,c,d,e,f; unknown: A,B,C,D,E,F). Given the fact that I dont know how to transform x->X, Im proposing that this could be done by using a logistic function with 3 unknown free parameters. The problem is then that the parameter space would have a total of 5x10^204 combinations. Yet, using a random sub-sample of these combinations, Im finding that the hypothesis holds true. A phycist friend of mine just told me that the problem might be solved by using a Hidden Markov model, but Ive never heard about this before. So... any hints for me for where to start searching for????

thanks in advance :)

[Image Analyst]

Regarding your main question "how to transform x->X", what is little x?

[Star Strider]

Start here for Hidden Markov Models.