Combining bins for chi-sq GoF [Q]

As per the title, my question is about combining non-adjacent bins for a chi-sq GoF test. I understand why combined bins generally ought to be adjacent, but I have a bit of an unusual situation. I have a distribution which is a candidate for a B(7, 0.5) distribution. The observed frequencies for X=0...7 over 52 weeks are [2, 5, 12, 10, 10, 5, 2, 6]. Expected are [0.4, 2.8, 8.5, 14.2, 14.2, 8.5, 2.8, 0.4]. Clearly the binomial distribution is not a good fit. But the problems are in the tails whose expected values are so low they need to be combined. Combining X=0,1 and X=6,7 still only give exp