Information Scale Correction for Varying Length Amplicons Improves Eukaryotic Microbiome Data Integration

Abstract

The integration and reanalysis of big data provide valuable insights into microbiome studies. However, the significant difference in information scale between amplicon data poses a key challenge in data analysis. Therefore, reducing batch effects is crucial to enhance data integration for large-scale molecular ecology data. To achieve this, the information scale correction (ISC) step, involving cutting different length amplicons into the same sub-region, is essential. In this study, we used the Hidden Markov model (HMM) method to extract 11 different 18S rRNA gene v4 region amplicon datasets with 578 samples in total. The length of the amplicons ranged from 344 bp to 720 bp, depending on the primer position. By comparing the information scale correction of amplicons with varying lengths, we explored the extent to which the comparability between samples decreases with increasing amplicon length. Our method was shown to be more sensitive than V-Xtractor, the most popular tool for performing ISC. We found that near-scale amplicons exhibited no significant change after ISC, while larger-scale amplicons exhibited significant changes. After the ISC treatment, the similarity among the data sets improved, especially for long amplicons. Therefore, we recommend adding ISC processing when integrating big data, which is crucial for unlocking the full potential of microbial community studies and advancing our knowledge of microbial ecology.

Primer selection impacts the evaluation of microecological patterns in environmental microbiomes

“…These primer biases could be attributed to differential taxa specificity and amplification efficiency. Moreover, the regional information scale, for example, the longer region and higher regional hypervariability, can affect taxonomic resolution [ 34 , 35 , 36 , 37 ], potentially explaining the improved taxa detection at the species level in V5–V7. These findings underscore the bias inherent in marker gene‐based metagenomics, resulting from the imbalances among primer sequences, amplification systems, and reaction condition optimization.…”

“…Specifically, V5–V7 data set exhibited stronger phylogenetic signals than V4. These disparities may arise from the differential taxonomic composition yielded by differential resolution [ 34 , 36 ]. The longer length and higher entropy of the V5–V7 region provide greater resolution than V4 [ 34 ], leading to divergent estimation of species relatedness/phylogenetic structure (lower βNTI) and community assembly process partitioning.…”

“…Our results are consistent with recent findings that demonstrated the strong effects of primer selection on CR‐based community profiling using machine learning [ 27 ]. Although amplicon sequences from different primers can be combined and aligned through representative sequence mapping, CR cannot fully address the biases between primers, such as information scale during sequence mapping and the over or underrepresentation of taxa due to taxa specificity and amplification bias [ 36 ]. Moreover, CR per se can influence the results and does not significantly reduce primer bias.…”

“…Moreover, CR per se can influence the results and does not significantly reduce primer bias. These findings underscore the importance of caution when integrating data sets from different primers and suggest potential mitigation strategies, such as abundance calibration information scale correction and sample pairing [ 36 , 52 ]. For example, a recent meta‐analysis integrated paired samples of bulk soils and rhizosphere from different studies, thereby mitigating the impact of primer bias [ 52 ].…”