"cheating" on non-linear regression


Recently I have been struggling with multiple linear regression with gene expression analysis. With my methods I have found a few interesting transcripts that correlate with the data I'm studying. However, I know there are transcripts that may increase exponentially and those fit poorly in my analysis. Therefore, to fit these genes better in my model, I considered transforming the variables into log2, log4, log8 and so on, which could transform exponentials into straight lines. I borrowed this old trick from analytical chemistry, where this has been long used to make formulas and numbers easier to work with in a daily basis.

Would you say this is a valid method?