Current algorithms have used profile Hidden Markov Models (pHMMs) for the automated detection of biosynthetic gene clusters (BGCs) in microbial genomes. However, these algorithms rely on the availability of well-assembled genomic input. Moreover, they are unable to tolerate unassembled or mixed genomes. Here, we evaluated the performance of available pHMMs to identify the Condensation domain in non-ribosomal protein synthetases (NRPSs) in the metagenomic sequence data of healthy American patients. We found that sensitivity increased from 56.22% using pHMMs to >70% in several segments of the pHMM (spHMM). The spHMM models performed better compared to the original pHMM model that they were built from and showed that there are several conserved region of residues that are more essential in detecting Condensation domains than others within a model. Our results provide a rapid detection method that is less computationally expensive for profiling the biosynthetic capacity of a large scale cohort.