Salgado, A. | Search bioRxiv

bioRxiv2026

Ground-truth in silico spike-ins reveal limits of microbiome biomarker recovery in colorectal cancer

Microbiome-based biomarkers have been proposed for colorectal cancer (CRC), yet candidate taxa are often interpreted without knowing whether taxonomic profiling workflows can reliably detect and quantify them in human samples. Existing ground-truth studies commonly rely on simplified communities that do not preserve the biological and technical complexity of clinical stool metagenomes. We hypothesized that weak CRC-associated signals, particularly those relevant to early-stage disease, may be missed through analytical non-recovery rather than biological absence. We developed an in silico spike-in framework that embeds CRC-associated signals into clinical stool metagenomes. Ten taxa were introduced individually at six fractions (0.01-5%) or as an equally weighted community at seven total fractions (0.01-10%; effective per-taxon fractions, 0.001-1%), generating 5,770 spike-in metagenomes from 310 samples. The resulting metagenomes were profiled with Kraken2/Bracken and MetaPhlAn 4 to quantify detection, abundance accuracy, false-positive signals, biomarker recovery, and calibration against a known ground truth. Recovery depended strongly on workflow, taxon, abundance, and clinical background. At 0.01%, four taxa- F. nucleatum, P. micra, P. stomatis, and P. intermedia-showed good recovery in 85-90% of samples under Kraken2/Bracken, whereas none achieved good recovery in at least 50% under MetaPhlAn 4. Greater low-abundance recovery was accompanied by a broader artefact-prone background (54.9% versus 0.5% of non-target taxa). Artefact-prone taxa accounted for 96.3% and 100% of enriched off-target differential-abundance calls, respectively. Spike-in-derived artefact exclusion substantially reduced off-target detections where present, while abundance-response modelling provided proof-of-principle correction of systematic abundance distortions in both evaluated configurations. Overall, the evaluated profiling configurations demonstrate that known low-abundance CRC-associated signals can be missed or distorted across a complete biomarker-discovery pipeline. Analytical non-recovery may cause early-detection biomarkers to be missed rather than indicate biological absence. Importantly, artefact-aware filtering and abundance calibration show that these limitations can be partially overcome. Improvements in taxonomic profiling may help bring reliable microbiome-based CRC diagnostics closer to clinical application.

bioinformatics↗

bioRxiv2021

Machine learning models exploring characteristic single-nucleotide signatures in Yellow Fever Virus

Yellow fever virus (YFV) is the agent of the most severe mosquito-borne disease in the tropics. Recently, Brazil suffered major YFV outbreaks with a high fatality rate affecting areas where the virus has not been reported for decades, consisting of urban areas where a large number of unvaccinated people live. We developed a machine learning framework combining three different algorithms (XGBoost, random forest and regularized logistic regression). This method was applied to 56 YFV sequences from human infections and 27 from non-human primate (NHPs) infections to investigate the presence of genetic signatures possibly related to disease severity (in human related sequences) and differences in the PCR cycle threshold (Ct) values (in NHP related sequences). Our analyses reveal four non-synonymous single nucleotide variations (SNVs) on sequences from human infections, in proteins NS3 (E614D), NS4a (I69V), NS5 (R727G, V643A) and six non-synonymous SNVs on NHP sequences, in proteins E (L385F), NS1 (A171V), NS3 (I184V) and NS5 (N11S, I374V, E641D). We performed comparative protein structural analysis on these SNVs, describing possible impacts on protein function. Despite the fact that the dataset is limited in size and that this study does not consider virus-host interactions, our work highlights the use of machine learning as a versatile and fast initial approach to genomic data exploration. ImportanceYellow fever is responsible for 29-60 thousand deaths annually in South America and Africa and is the most severe mosquito-borne disease in the tropics. Given the range of clinical outcomes and the availability of YFV genomic data, the use of machine learning analysis promises to be a powerful tool in the investigation of genetic signatures that could impact disease severity and its potential of being reintroduced in an urban transmission cycle. This can assist in the search for biomarkers of severity as well as help elucidating variations in hosts Ct value. This work aims to propose a relatively fast and inexpensive computational analysis framework, which can be used as a real-time, innitial strategy associated with genomic surveillance to identify a set of single nucleotide variants putatively related to biological and clinical characteristics being observed.

microbiology↗