As per USFDA guidelines for Bioanalytical Method validation for human studies, one of the acceptance criteria for linearity is that four out of six non zero standards meet the criteria of <= 20 % deviation at LOQ and <= 15 % deviation at other concentrations. Thus if one of the level which does not meet the above criteria, should be omitted for the calculation of regression line ?

If I understand you correctly then the answer is: Yes
If you start off with 6 calibration samples and one of those calibration samples fails the above criteria then the regression line should be re-calculated using the remaining 5 calibration samples.