Typing of real-world clinical specimens at the species level (A) and strain level (B). (A) Comparison of human reads percentage, read count of total bacteria, read count of Top1 bacterial species and read count of the cultured species between culture-positive and culture-negative specimens. The culture-positive specimens are classified into two groups: Culture⁺_Y: the culture-positive species and the Top1 species by read count are the same; Culture⁺_N: cultured species and Top1 species in mNGS are different. (B) Strain-level statistics of four species in the specimens. For each of the four species, the culture-positive specimens containing them are classified into two types: Culture⁺_Y, the species is the cultured one and meanwhile is the Top1 by read count (hence, the species is probably the causative pathogen in the specimen); Culture⁺_N, the species is not the cultured one or the Top1 by read count (hence, the species is probably not the causative pathogen). The culture-negative specimens are also classified into two types: Culture⁻_Y, the species ranks Top1 by read count; Culture⁻_N, the species does not rank Top1 by read count. Top1/Top2 indicates a ratio of strain-level Top1/Top2 based on the abundance inferred by MIST. TPM indicates read counts per million. (C) ANOSIM correlation R at different Top1/Top2 values, which reflect the significance of divergence between the specimens with Top1/Top2 below the value and the specimens with Top1/Top2 above the value. The value with the highest R is used as the cutoff that decides whether there is a predominant strain in the specimen. (D) Pielou’s evenness index shows the diversity of strain composition between culture-positive and culture-negative specimens. The higher the value, the more diverse the strain level of the specimen tends to be.