Meriem, unfortunately your question is a bit vague, because the reliability of the genotype calls depends greatly on the technology used to generate them. For example, an exome SNP-chip will probably have a much lower false positive rate than Whole Exome Sequencing data.

If your data is from a large sample size, and the variants are common enough to have plenty of homozygotes in the set, calculating the deviation from Hardy Weinburg equilibrium could give an indication of unreliability for a specific variant, but only in control cohorts - otherwise it could indicate association with disease.

For WES data, a minimum of 10 reads covering a position is a good cutoff, and most people also filter the data at GQ > 20, but I've found that any GQ < 90 could be indicative of misaligned reads (especially in genes with close paralogues, and the HLA cluster is going to be very problematic). Having access to the BAM files to view the genomic alignments is extremely useful, but only feasible for very rare variants where you don't have to check many samples

I agree with Simon. Also to add that genotyping can be difficult with WES as the futher apart your variants are the harder they are to phase (i.e. is a compound het on the same allele or not).