steepness, hill - different parameters
- problem/goal/basics
- steepness
- log(size_99th) - log(size_95th)
- the percentiles are parameters, these are the defaults, they were used in the article
- steepness and hill is highly correlated
- why isn't lkb1 the best?
- steepness will be high if there is a high probability of a big tumor
- we saw on the CDFs that lkb1 has a "bump" (dúsulás)
- so we expect lkb1 to have a high steepness value
- => is it because of the parameters?
- is there correlation between the fitting error and the steepness? (default and new parameters)
- we expect correlation (with "good" parameters) because we saw that all the badly fitting ones have that "bump"
- original size cdf:
- (red is lkb1, green is pten)
- pten's high steepness:
- 99th percentile size is the highest
- 95th percentile is not that different from the others
- => log(size_99th) - log(size_95th) will be big
- lkb1's "low" steepness
- 99th percentile size is high
- 95th percentile is the highest by a big margin!
- => log(size_99th) - log(size_95th) won't be big!
- lkb1 is "steep" as well, but we don't get a high value, because the steepness ("the linear part") starts lower
- new parameters (hand-picked, to get a higher score for the lkb1)
- lower percentile: where the size is still close to the other ones => 90th percentile
- higher percentile: where the size is the highest => anywhere between 93-98 => 95th
- changes in steepness with new parameters
- lkb1 is now the highest
- correlation between fitting error
- similar results with hill
- instead of the default 95th percentile, use the 90th percentile, because we think that lkb1's steep part starts there
- new idea
- for every mutant calculate the hill estimate for several percentiles (1-99)
- we normalize to the inert in every percentile
- plot these (x - percentile, y - normalized hill estimate, every line is a genotype)
- we calculate these values for every genotype and then get the percentile corresponding to the maximum hill estimate, then we plot the percentile and the fitting error:
- the outlier with an error of 0.021 is p53;rb1
- the other 3 outliers: 'Keap1', 'Rasa1', 'Apc_Rnf43'
- if we plot the percentile-hill estimate for these, we can see that it has a maximum at ~0.8, but the line is mostly flat
- p53;rb1 is red, others are grey
Megjegyzések
Megjegyzés küldése