Work

Quantitative microbiota profiling: Comparison of computationally predicted profiles against quantitative PCR-based reference

Year: 2026
Type: article
Abstract: Microbiota profiles generated by high-throughput sequencing are compositional and do not capture bacterial load. Studies using quantitative microbiota profiling (QMP) have shown that variation in microbial load is a major source of microbiota variation and can confound disease associations. This thesis evaluated whether sequencing-based computational load proxies and differential abundance approaches can reproduce qPCR-based QMP in a longitudinal cohort. Fecal samples from 95 Finnish healthy adults traveling to Benin, West Africa, were collected before travel, immediately after travel, and one month later (287 samples in total). Shotgun metagenomic sequencing was combined with universal 16S rRNA gene qPCR for QMP. Two machine-learning models and the bacterial-to-host read count ratio derived from the metagenomic data were compared with qPCR-based total bacterial load. Family-level differential abundance results of bacterial taxa were compared across absolute, relative, and several common differential abundance testing methods using qPCR-based quantitative profiles as the reference. Both machine-learning models correlated weakly with qPCR-based bacterial load, and the bacterial-to-host ratio showed no significant correlation. Agreement in differential abundance results was highest when bacterial load was stable and lowest for the contrast with the largest qPCR-detected load decrease. Machine-learning-predicted absolute abundance performed better than most relative-abundance approaches but failed to reproduce the qPCR-based QMP reference. Among methods without measured bacterial load, total-sum-scaled log-transformed relative abundance with a linear mixed model showed the closest agreement. Methods such as centered log-ratio, LinDA, ANCOM-BC2, and DESeq2 showed more missed findings, method-specific findings, or direction errors. In longitudinal studies, or where substantial shifts in microbial load are expected, experimental load measurement is important for quantitative profiling. More complex normalization or statistical testing methods did not consistently improve agreement with quantitative microbiota profiles, and some produced misleading biological interpretations when bacterial load changed.
Language: English
Open Access status: green
OpenAlex ID: W7212416681
FWCI:
Cites:
Cited by:
Related to: 10