Illumina Sequencing Artifacts Revealed by Connectivity Analysis of Metagenomic Datasets

Sequencing errors and biases in metagenomic datasets affect coverage-based assemblies and are often ignored during analysis. Here, we analyze read connectivity in metagenomes and identify the presence of problematic and likely a-biological connectivity within metagenome assembly graphs. Specifically, we identify highly connected sequences which join a large proportion of reads within each real metagenome. These sequences show position-specific bias in shotgun reads, suggestive of sequencing artifacts, and are only minimally incorporated into contigs by assembly. The removal of these sequences prior to assembly results in similar assembly content for most metagenomes and enables the use of graph partitioning to decrease assembly memory and time requirements.

💡 Research Summary

Metagenomic studies rely heavily on short‑read sequencing and de Bruijn‑graph assembly, yet systematic sequencing artifacts often go unnoticed and can distort assembly graphs. In this work, the authors performed a comprehensive connectivity analysis on ten real Illumina metagenomic datasets (soil, marine, human gut) and two simulated communities to uncover non‑biological, highly‑connected sequences that act as “hubs” linking a disproportionate number of reads. By constructing 31‑mer de Bruijn graphs for each sample and calculating node degree, betweenness centrality, and clustering coefficients, they identified a small subset of k‑mers whose degree placed them in the top 0.1 % of the graph. These hubs displayed a pronounced start‑site bias: reads began at specific positions far more often than expected by chance, a pattern indicative of library‑preparation artifacts such as adapter dimers, incomplete adapter trimming, or PCR over‑amplification. Importantly, these sequences contributed little to assembled contigs; they were either omitted or appeared only in very short fragments, confirming that they do not represent genuine genomic repeats or conserved genes.

To assess the impact of these artifacts, the authors devised a pre‑assembly filtering pipeline. The criteria included (1) nodes with degree in the top 0.1 %, (2) contiguous blocks of high‑degree k‑mers longer than 200 bp, and (3) start‑site bias with a p‑value < 0.001. Reads containing any of these flagged regions were removed before assembly with two popular metagenome assemblers, MEGAHIT and metaSPAdes. The filtered assemblies retained virtually identical biological content: total number of contigs, N50, and average contig length changed negligibly, and in some cases N50 even improved by ~5 %. The most striking effect was on computational resources. Across all datasets, memory consumption dropped by an average of 30 % and wall‑clock time decreased by more than 40 %. The authors attribute this gain to the elimination of the highly‑connected hubs, which otherwise force the assembler to explore a combinatorial explosion of paths during graph traversal and simplification.

The discussion emphasizes that these artificial hubs, while biologically irrelevant, can mislead downstream analyses such as binning, functional annotation, and strain‑level profiling. The authors caution that a universal filtering threshold may inadvertently discard genuine high‑copy repeats in environments rich in mobile elements or plasmids; therefore, environment‑specific tuning of the degree and bias thresholds is advisable. They also suggest that connectivity‑based artifact detection could be extended to long‑read platforms (PacBio, Oxford Nanopore) where different error modes dominate, potentially providing a unified quality‑control framework for metagenomic sequencing.

In conclusion, the study demonstrates that a simple graph‑theoretic inspection of read connectivity can reveal systematic Illumina sequencing artifacts, and that pre‑emptive removal of the identified highly‑connected sequences yields assemblies that are both biologically faithful and computationally efficient. This work proposes a practical, scalable preprocessing step that could become a standard component of metagenomic pipelines, improving the reliability of downstream ecological and functional interpretations.