Hi, I have some candidate genomic regions of interest and would like to see if they are differentially expressed between two conditions. Each condition has three biological replicates.

Because I cannot define the boundaries of the candidate genomic regions exactly (they are not annotated with gene model), I thought about comparison read counts at the single nucleotide level (therefore comparison of sequencing depth) and applying some thresholds (p-value or test statistics) to define differential (continuous) regions. Is it possible for me to provide DESeq2 with read counts at single nucleotide resolution ("depth") to perform statistical tests generating statistic or p-value so that I can get significant at each base ?

I would not recommend running DESeq2 on read depth in regions. The dispersion estimation may be thrown off and the multiple test correction needs to be considered as you have very high correlations for each position with neighbors.

See the derfinder package.