Invalid background frequencies

Hi,

I'm trying to create PSSMs for proteins with background frequencies.

I make extra sure that my background distribution sums up to 1.0 with:

```python
background_freq = {
"A": 0.0825,
"R": 0.0553,
"N": 0.0406,
"D": 0.0545,
"C": 0.0137,
"Q": 0.0393,
"E": 0.0675,
"G": 0.0707,
"H": 0.0227,
"I": 0.0596,
"L": 0.0966,
"K": 0.0584,
"M": 0.0242,
"F": 0.0386,
"P": 0.0470,
"S": 0.0656,
"T": 0.0534,
"W": 0.0108,
"Y": 0.0292,
"V": 0.0687,
"X": 0.0
}
s = sum(background_freq.values())
background_freq = {aa: f / s for aa, f in background_freq.items()}
background_freq["A"] += 1. - sum(background_freq.values())
assert sum(background_freq.values()) == 1.0
```

Still, when I try to get a PSSM with this I get a `ValueError` every time:
```
>>> pssm = pcm.normalize(pseudocount=0.25).log_odds(background=background_freq)
ValueError: Invalid background frequencies
```

I've traversed the documentation and looked for usage examples in the unit tests, but I couldn't find anything to help me understand how to correctly pass the background distribution.

Many thanks for the help.