You should leave them out of your analysis.
You must transform your data before you analyse them.
You can use a non-parametric test to analyse the data.
You should repeat the experiment and collect a new data set.
You can repeat the analysis both including and excluding the outliers, and include the outliers if the results are consistent.