Marker-Centered Evaluation of De Novo Assemblers Using Validated Microsatellite Loci in the Endangered Korean Fish Pseudopungtungia nigra

The recovery of experimentally validated microsatellite loci from short-read assemblies may provide an application-specific measure of assembly quality that complements conventional genome-wide statistics. In this study, four de novo assemblers—SPAdes, ABySS, SOAPdenovo2, and MaSuRCA—were evaluated using genomic reads from two individuals of the endangered Korean freshwater fish Pseudopungtungia nigra. Assembler performance was assessed based on assembly contiguity, BUSCO completeness, computational requirements, and the recovery of 20 PCR-validated microsatellite loci and their flanking regions. SPAdes produced the most contiguous assemblies and the highest BUSCO completeness among the tested assemblers, with N50 values of 5.5–7.0 kb and BUSCO completeness of 30.2–35.2%, and recovered all 20 loci and their flanking sequences at high confidence in both individuals. Marker recovery by ABySS, SOAPdenovo2, and MaSuRCA varied according to k-mer size and sample, with high-confidence recovery decreasing at larger k-mer sizes in several configurations and reaching 9 of 20 loci in the least favorable condition. SPAdes used 32.1–34.2 GB of peak memory, compared with up to 135.0 GB in the most memory-intensive SOAPdenovo2 runs. Reference-guided comparisons retained substantially greater total aligned sequence with the two Pseudorasbora references than with the more divergent Cyprinus carpio reference, although the mean identity of the smaller retained C. carpio alignment set was slightly higher. All 20 validated marker sequences produced high-confidence provisional placements in both Pseudorasbora reference genomes. As an exploratory downstream application, the SUC6832-derived pseudo-reference enabled joint genotyping of the two samples and yielded 2,012,643 filtered candidate variants across 25 guide-assigned scaffolds. These results indicate that SPAdes provided the most favorable balance of validated marker recovery, relative assembly quality, and memory efficiency among the tested configurations, although all assemblies remained fragmented at the genome-wide level. Marker-centered benchmarking can therefore complement conventional assembly statistics when short-read assemblies are generated primarily for microsatellite development, while pseudo-reference-based genomic analyses should remain provisional because of assembly fragmentation, reference bias, and limited biological sampling.