Companies such as 10x have official recommendations for depth at their websites. Low-abundance detection and scRNA-seq is not the best combination as scRNA-seq often only detects a fraction of all genes. I would either consider a type of scRNA-seq that focuses on depth rather than many cells (so plate-based and not droplet-based) or alternatively do an initial scRNA-seq to define heterogeneity in your model system and identify per-population marker genes and then sort these populations you want by FACS and do a bulk RNA-seq which is still the best in terms of detecting even lowly-abundant genes.