Biology subjects

Safaei, N.

Publications and source records attributed to Safaei, N..

4 recordsLinked to original sources

bioRxiv2025

Metax: A Coverage-Informed Probabilistic Framework for Accurate Cross-Domain Taxon Profiling

Metagenomic taxonomic profiling is essential for characterizing microbial community composition in both environmental and clinical contexts. Existing profilers have greatly advanced community characterization; however, achieving accurate profiling across all domains of life - especially for archaea, fungi, and viruses - and for low-biomass, host-dominated samples, remains challenging. We describe Metax, a cross-domain taxonomic profiler that employs probabilistic modeling of genome coverage to distinguish true community members from artifactual signals arising from reference contamination, local genomic similarity, or reagent-derived DNA fragments. In comprehensive benchmarks across more than 500 samples, Metax demonstrated accurate species-level profiling, with consistent performance for bacteria, archaea, eukaryotes and viruses, and robustness to shallow sequencing. Applied to an oral microbiome cohort, Metax identified differentially abundant viral taxa distinguishing peri-implantitis from healthy sites, while analyses of tumor microbiome data revealed reagent-borne contaminants and potential reference misassemblies. By integrating coverage-informed statistics, Metax delivers accurate, robust, and interpretable cross-domain taxonomic profiles, maintaining stable performance across diverse sequencing depths and sample types.

bioinformatics↗

bioRxiv2025

Comparative Assessment of Large Language Models for Microbial Phenotype Annotation

Large language models (LLMs) are increasingly used to extract knowledge from text, yet their coverage and reliability in biology remain unclear. Microbial phenotypes are especially important to assess, as comprehensive data remain sparse except for well-studied organisms and they underpin our understanding of microbial characteristics, functional roles, and applications. Here, we systematically assessed the biological knowledge encoded in publicly available LLMs for structured phenotype annotation of microbial species. We evaluated the performance of over 50 LLMs, including state-of-the-art models such as Claude Sonnet 4 and the GPT-5 family of models. Across phenotypes, LLMs reached accurate assignments for many species, but performance varied widely by model and trait, and no single model dominated. Model self-reported confidence is informative, with higher confidence aligning with higher accuracy, and can be used to prioritize phenotype assignment, effectively distinguishing between high-and low-confidence inferences. Overall, our study outlines the utility and limitations of text-based LLMs for phenotype characterization in microbiology.

bioinformatics↗

bioRxiv2025

Detection of Paracoccus yeei in Spontaneous Bacterial Peritonitis using Rapid, Long Read Sequencing

Ascites is a common complication in patients with decompensated liver cirrhosis. Spontaneous bacterial peritonitis (SBP) is the most frequent infection, affecting up to 30% of hospitalized patients with ascites. Multidrug-resistant bacteria in patients with liver cirrhosis are becoming more common, particularly in nosocomial infections. With early detection and adequate antibiotic therapy, the mortality rate of SBP can be reduced from over 90% to 15-20%. However, the etiologic agent remains undetected by conventional pathogen diagnostics from ascites in more than half of the patients. Here, we describe a robust workflow combining long read metagenome sequencing with best practices for low microbial biomass sample processing and bioinformatic analytics, to detect and characterize pathogens from the cirrhotic ascites of SBP patients, as a complement to routine microbiological diagnostics. This approach identified Paracoccus yeei as a likely etiologic agent for a patient who developed a fever after successful treatment of an Escherichia coli infection. Analysis of the genome of P. yeei recovered directly from ascites delineated pathogenicity-related features and an antimicrobial resistance profile consistent with the patients treatment. Pangenome phylogenetics placed the P. yeei strain closest to isolates associated with abdominal infections. Our study underscores the utility of metagenomics for a rapid and comprehensive assessment of infectious agents, suggesting a novel and rarely reported pathogen contributing to SBP with potential implications for diagnostics and therapeutic strategies. ImportanceOur study demonstrates the clinical utility of real-time, long-read metagenomics in identifying elusive pathogens in Spontaneous Bacterial Peritonitis (SBP), a condition with high mortality and frequently inconclusive conventional diagnostics. By uncovering Paracoccus yeei as a likely secondary etiologic agent and characterizing its pathogenicity and AMR profile, our findings expand the known etiology of SBP and emphasize the potential of metagenomics to improve diagnostic accuracy, guide targeted treatment, and ultimately reduce mortality rates in infectious diseases. This approach could shape future clinical practice for managing complex infections.

microbiology↗

bioRxiv2024

Volitional stopping is preceded by a transient beta oscillation

Human motor cortex EEG beta (15-30 Hz) oscillations undergo transient power modulations (bursts) during volitional control of movements. They are a potential control signal for brain-machine interfaces and are a therapeutic target in Parkinsons disease. The prevailing view is that EEG beta bursts increase during stopping and immobility, but do not precede stopping. In contrast to prior work in humans and animals that used a latent and unobservable stopping time in the stop-signal task, we developed a translational animal model to align EEG with overt action stopping. We recorded 32-electrode EEG along with the angular velocity of a treadmill while head-fixed rats stopped in-progress running on a freely-rotating, non-motorized treadmill. Contrasting prior work, motor cortex beta bursts increased before stopping and not during stopping or immobility. Using information theoretic measures, we show that beta power was informative about treadmill velocity 200 msec in the future, but only during planning to stop. By introducing artificial temporal jitter to mimic the estimation of stopping time used in prior work, we show that this predictive brain-action relationship fails with even small jitter. Finally, we use a variety of machine learning methods to show that, despite EEG beta oscillations being a clear neural correlate preceding stopping, it has limited utility for real-time action decoding. Our work suggests a new conceptual model for neural control of action stopping.

neuroscience↗