When I ran Cuffdiff, I had 92 genes that had the problems with the "test_stat" column.
These genes had "-nan" for their test_stat value. All of these had log2(fold_change) values equal to either infinity or -infinity.
1. What does mean?
2. Can I trust the significance tests for these genes even if their test_stat = -nan?? P-values were still calculated and were different amongst the various genes that had a test_stat = -nan.