Need for Transformation?

I'm running an OLS regression. My predictors are continuous, but they are proportions that are bounded by 0 and 1.

Because there are many 0s in my predictors, they are heavily skewed to the right.

1. Do I need to transform my predictors? (I know that when the DV is a proportion, you would use logit transformations, etc.)

2. I've tried sqrt, fourth sqrt, arcsin, and log (with adding a constant to deal with 0s) but it doesn't seem to help the shape of the distribution. What to do next?

Thanks in advance for your help!


Ambassador to the humans
Typically there aren't any assumptions made on the distribution of the predictors. Are the predictors giving you any other troubles that causes you to want to transform?
Thanks! Nope, no other seems like a common misunderstanding (which I had myself) that the predictors and outcomes both need to meet a normal distribution. Really is about the normality of residuals.