I would like to use low pass sequencing to replace a genotyping chip to be able to detect variants up to 0.1 % allele frequency in available population data. What is the minimum depth I can opt for to retain the confidence of base calls? Can GATK be used to analyse this data?
This is a good idea - low pass sequencing has shown to be better than genotype arrays for things such as GWAS or QTL mapping.
The primary way of processing low-pass sequencing data is to use imputation - partciularly methods such as GLIMPSE or QUILT and a large reference panel such as the HRC or TopMed in order to refine the genotype calls.
This is a figure from the GLIMPSE paper showing how you can retain variants of different frequencies at different levels of sequencing depth.
And a similar image from the QUILT paper:
You might also want to check out some of the methodology in this paper http://m.genome.cshlp.org/content/31/4/529