A metagenomics-based technological framework has been proposed for evaluating the potential and utility of FEAST as an ARG profile-based source apportionment tool. To this end, a large panel of metagenomic data sets was analyzed, associating with eight source types of ARGs in environments. Totally, 1089 different ARGs were found in the 604 source metagenomes, and 396 ARG indicators were identified as the source-specific fingerprints to characterize each of the source types. With the source fingerprints, predictive performance of FEAST was checked using "leave-one-out" cross-validation strategy. Furthermore, artificial sink communities were simulated to evaluate the FEAST for source apportionment of ARGs. The prediction of FEAST showed high accuracy values (0.933±0.046) and specificity values (0.959±0.041), confirming its suitability to discriminate samples from different source types. The apportionment results reflected well the expected output of artificial communities which were generated with different ratios of source types to simulate various contamination levels. Finally, the validated FEAST was applied to track the sources of ARGs in river sediments. Results showed STP effluents were the main contributor of ARGs, with an average contribution of 76 %, followed by sludge (10 %) and aquaculture effluent (2.7 %), which were basically consistent with the actual environment in the area.