Clinical Chemistry

IntroductionDespite its promise, accuracy of microbial cell-free DNA (mDNA) in plasma as a diagnostic tool is hindered by its low abundance and process contaminants. We have previously shown that combining size selection with single-stranded DNA (ssDNA) library preparation increased mDNA yield by 200-fold but also decreased sensitivity for pathogen detection due to higher background noise. A recent study showed that pathogen-derived DNA was enriched for CC dinucleotide at 5 ends compared to contaminants. Since ssDNA libraries preserve sequence motifs at both ends (5 and 3), we hypothesized that analysis of nucleotide motifs at microbial fragment ends in size-selected ssDNA libraries could help differentiate pathogen DNA from background noise. MethodsWe performed deep sequencing on size-selected ssDNA libraries (<110 bp) generated from longitudinal plasma samples of 11 critically-ill patients (5 with culture-proven infections, 20 samples; 6 without infections, 18 samples) and 6 no-template controls (NTCs). For each 2-mer and 1-mer motif, we calculated the ratio between its frequency observed at 5 and 3 fragment ends in sequencing data and its expected frequency in the corresponding reference genome (O/E ratio). We compared enrichment of motifs in pathogen DNA and contaminant DNA fragments. ResultsPathogen-derived mDNA fragments were more biased in O/E end motif ratios compared to contaminants across all 3 groups (NTCs, no-infections and culture-proven infections), at both 5 and 3 fragment ends. Notably, the GG dinucleotide was enriched at the 3 end in pathogens compared to contaminants (P < 0.0001). Combining O/E ratios for C and G nucleotides at the 3 end achieved areas under the receiver operating characteristic curve of >0.98 for distinguishing common contaminants from culture-proven pathogens. ConclusionsPathogen-derived mDNA in size-selected ssDNA libraries is biased at 5 and 3 fragment end compared to contaminants. Incorporating microbial fragment end motif analysis can enhance signal-to-noise ratio and improve pathogen detection and identification in plasma metagenomic sequencing.

BackgroundCirculating tumor DNA (ctDNA) is a non-invasive biomarker that can be used as a tool to detect minimal residual disease (MRD). MRD can provide important prognostic information in diffuse large B-cell lymphomas (DLBCL). Here, we present an MRD assay with an improved detection method for ctDNA, Phased Variant Enrichment and Detection Sequencing (PhasED-Seq) which leverages phased variants (PVs) to detect ctDNA. MethodsPlasma samples from non-cancer controls were used to assess assay specificity. A limiting dilution series using DLBCL clinical-contrived samples was performed to assess assay sensitivity and precision. The accuracy of the PhasED-Seq-based assay was assessed using plasma samples from individuals with DLBCL and for whom MRD comparator assay results were also available. All samples were sourced from commercial vendors or academic studies. ResultsThe analytical and clinical performance of the MRD assay was evaluated using clinical and clinical-contrived DLBCL samples. The assays false positive rate was 0.24% and the background error rate was 1.95E-08. The limit of detection at 95% detection rate (LoD95) at 120 ng was 0.7 parts in 1,000,000 and precision was >96%. Clinical accuracy was 90.62% PPA and 77.78% NPA. ConclusionsThe PhasED-Seq-based MRD assay has strong analytical and clinical performance in B-cell dyscrasias. Through the development of improved ctDNA detection methods such as that presented here, patient outcomes may be improved through the detection of residual disease or early relapse which may be used to guide treatment decisions. Brief SummaryHere we present the analytical validation of a non-invasive minimal residual disease (MRD) assay which uses Phased Variant Enrichment and Detection Sequencing (PhasED-Seq) to improve the error profile and sensitivity of circulating tumor DNA (ctDNA) detection. The assays performance included a false positive rate of 0.24% and a background error rate of 1.95E-08. The limit of detection at 120 ng was 0.7 parts in 1,000,000 (6.61E-07 PVAF) with precision >96%. Positive and negative agreement were 90.62% and 77.78%, respectively. This suggests that the PhasED-Seq-based MRD assay is accurate and reproducible, thus appropriate for clinical use for individuals with B-cell malignancies.

BackgroundLiquid biopsy based on cell-free DNA (cfDNA) is an established approach in clinical diagnostics. In recent years, a fraction of cfDNA comprising short fragments has been discovered, that is enriched at gene promoters and binding sites of DNA-binding proteins. However, the diagnostic potential of such short double-stranded cell-free DNA (footprint DNA) remains to be fully explored. Therefore, we characterized the clinical utility of footprint DNA in septic patients. MethodsWe enriched for footprint DNA based on size selection and subsequent high-throughput DNA sequencing to receive an unbiased, genome-wide picture of the host response to the infection. Footprint DNA occupancies were analyzed for correlation with clinical metrics including urea, hemoglobin, or alanine aminotransferase (ALT). Additionally, footprint DNA markers were benchmarked by read and receiver operating curve (ROC) analysis against procalcitonin (PCT) as an established marker for infection status as well as against clinical parameters for early death prediction. FindingsWe found that levels of occupancy of footprint DNA at defined genomic loci semi-quantitatively correlated with physiological markers like ALT or urea from major organ systems including liver or kidney. In a small proof-of-concept cohort, differential signatures of DNA footprints distinguished between patient groups with bacterial and viral infections with an area under the ROC (AUROC) of 1.0, which is considerably better than PCT with an AUROC of 0.75. Likewise, footprint DNA could also predict early death in septic patients with an AUROC of 0.983, compared to the SOFA (Sepsis-related organ failure assessment) score with an AUROC of 0.76. InterpretationOur findings show that footprint DNA delivers quantitative information on physiology at the DNA level, demonstrating its diagnostic and prognostic potential. Identified footprint biomarker regions could be helpful in the clinical assessment of septic patients and other complex diseases outperforming current state-of-the-art clinical diagnostics. FundingThis study was financed with internal funds from Fraunhofer society.

IntroductionFree light chain kappa (FLC{kappa}) are a newer sensitive biomarker to detect intrathecal immunoglobulin synthesis. Here we report the study protocol of the ORCAS study which will evaluate a novel laboratory work flow recently proposed including FLC{kappa} analysis simplifying cerebrospinal fluid (CSF) analysis. MethodsThe ORCAS study is a prospective multicenter diagnostic biomarker study across at least 4 study sites. The study protocol does not specify which assays should be applied in the participating laboratories to detect oligoclonal bands (OCB) or measure FLC{kappa} or Ig synthesis to represent real world data. Primary outcome parameter is the sensitivity and negative predictive value of the absence of local FLC{kappa} synthesis for the absence of intrathecal synthesis according to OCB and/or intrathecal IgG, IgA, IgM synthesis in quotient diagrams. The reference range of FLC{kappa} will be indicated by the FLC{kappa} quotient diagram. This study was designed according to the STARD criteria. PerspectiveThe establishment of a newer biomarker in routine practice is associated with significant difficulties, such as the inconsistent comparability of different measurement platforms, the establishment of suitable cut-offs or insufficient knowledge regarding sensitivity and specificity of the biomarker. If the ORCAS study objective is achieved, the use of the proposed workflow integrating FLC{kappa} analysis in routine practice in CSF diagnostics could help to better characterize the intraindividual and disease-specific intrathecal humoral immune response in a resource-efficient manner.

Patients who have radiographically detectable lesions in their brain or other symptoms compatible with brain tumors pose challenges for diagnosis. The only definitive way to diagnose such patients is through brain biopsy, an obviously invasive and dangerous procedure. Here we present a new workflow termed "CSF-BAM" that simultaneously identifies B cell or T cell receptor rearrangements, Aneuploidy, and Mutations using PCR-mediated amplification of both strands of the DNA from CSF samples. We first describe the details of the molecular genetic assessments and then establish thresholds for positivity using training sets of libraries from patients with or without cancer. We then applied CSF-BAM to an independent set of 206 DNA samples from patients with common, aggressive cancer types as well as other forms of brain cancers. Among the 126 samples from patients with the most common aggressive cancer types (high grade gliomas, medulloblastomas, or metastatic cancers to the brain), the sensitivity of detection was >81%. None of 33 CSF-BAM assays (100% specificity, 90% to 100% credible interval) were positive in CSF samples from patients without brain cancers. The sensitivity of CSF-BAM was considerably higher than that achieved with cytology. CSF-BAM provides an integrated multi-analyte approach to identify neoplasia in the central nervous system, provides information about the immune environment in patients with or without cancer, and has the potential to inform the subsequent management of such patients. Statement of significanceThere is a paucity of technologies beyond surgical biopsy that can accurately diagnose central nervous system neoplasms. We developed a novel, sensitive and highly specific assay that can detect brain cancers by comprehensively identifying somatic mutations, chromosomal copy number changes, and adaptive immunoreceptor repertoires from samples of cerebrospinal fluid.

Rapid classification and detection of SARS-CoV-2 variants have been critical in comprehending the viruss transmission dynamics. Clinical manifestation of the infection is influenced by comorbidities such as age, immune status, diabetes, and the infecting variant. Thus, clinical management may differ for new variants. For example, some monoclonal antibody treatments are variant-specific. Yet, an FDA-approved test for detecting the SARS-CoV-2 variant is unavailable. A laboratory-developed test (LDT) remains a viable option for reporting the infecting variant for clinical intervention or epidemiological purposes. Accordingly, we have validated the Illumina COVID-Seq assay as an LDT according to the guidelines prescribed by the College of American Pathologists (CAP) and Clinical Laboratory Improvement Amendments (CLIA). The limit of detection (LOD) of this test is Ct<30 ([~]15 viral copies) and >200X genomic coverage, and the test is 100% specific in the detection of existing variants. The test demonstrated 100% precision in inter-day, intra-day, and intra-laboratory reproducibility studies. It is also 100% accurate, defined by reference strain testing and split sample testing with other CLIA laboratories. Advanta Genetics LDT COVID Seq has been reviewed by CAP inspectors and is under review by FDA for Emergency Use Authorization.

BackgroundCurrent next generation sequencing (NGS) and microarray based Non-Invasive Prenatal Tests (NIPT), used for the detection of common fetal trisomies, are still expensive, time consuming and need to be performed in centralized laboratories. To improve NIPT in clinical routine practice as universal prenatal screening, we have developed a digital droplet PCR (ddPCR) based assay called iSAFE NIPT using cell free fetal DNA (cffDNA) for detection of fetal trisomies 13, 18 and 21 in a single reaction with advantage of high diagnostic accuracy and reduced cost. Materials and MethodsWe first used artificial DNA samples to evaluate analytical sensitivity and specificity of the iSAFE NIPT. Next, we analysed 269 plasma samples for the clinical validation of iSAFE NIPT. Fifty-eight of these, including five trisomies 21, two trisomies 18 and one trisomy 13 were utilised to establish the assay cut-off values based on ratios between chromosome counts. The remaining 211 plasma samples, including 10 trisomies 21, were analysed to evaluate iSAFE NIPT clinical performance. ResultsiSAFE NIPT achieved a 100% analytical sensitivity (95% CI 94.9-100% trisomy 21; 79.4-100% trisomy 18; 73.5-100% trisomy 13) and 100% specificity (95% CI 96.3-100% trisomy 21; 97.6-100% trisomy 18; 97.6-100% trisomy 13). It also achieved a 100% clinical sensitivity and specificity for trisomy 21 detection in the 211 clinical samples (95% CI for sensitivity is 69.1-100%, and 95% CI for specificity is 98.2-100%). ConclusionsThe iSAFE NIPT is a highly multiplexed ddPCR based assay for detection of fetal trisomies from maternal blood. Based on clinical validation, the iSAFE NIPT has high diagnostic sensitivity and specificity. It can be decentralized in routine clinical laboratories, is fast, easy to use and economical comparing to current NIPT.

Circulating cell-free DNA (cfDNA) from blood plasma of cancer patients can be used to interrogate somatic tumor alterations non-invasively or when adequate tissue is unavailable. We have developed and clinically implemented MSK-ACCESS (Analysis of Circulating cfDNA to Evaluate Somatic Status), an NGS assay for detection of very low frequency somatic alterations in select exons and introns of 129 genes. Analytical validation demonstrated 92% sensitivity in de-novo mutation calling down to 0.5% allele frequency and 98% for a priori mutation profiling. To evaluate the performance and utility of MSK-ACCESS, we report results from the first 681 prospective blood samples (617 patients) that underwent clinical analysis to guide patient management. Somatic mutations, copy number, and/or structural variants were detected in 73% of the samples, and 56% of these circulating-tumor DNA (ctDNA) positive samples had clinically actionable alterations. The utilization of matched white blood cell sequencing allowed retention of somatic alterations while filtering out over 10,000 germline and clonal hematopoiesis variants, thereby greatly enhancing the specificity of the assay. Taken together, our experience illustrates the importance of analyzing a matched normal sample when interpreting cfDNA results and highlights the potential of cfDNA profiling to guide treatment selection, monitor treatment response, and identify mechanisms of treatment resistance.

Structural variations (SVs) play a key role in the pathogenicity of hematological malignancies. Standard-of-care (SOC) methods such as karyotyping and fluorescence in situ hybridization (FISH), employed globally for the past three decades have significant limitations in the resolution or the number of recurrent aberrations that can be simultaneously assessed, respectively. Next-generation sequencing (NGS) based technologies are now widely used to detect clinically significant sequence variants but are limited in their ability to accurately detect SVs. Optical genome mapping (OGM) is an emerging technology enabling the genome-wide detection of all classes of SVs at a significantly higher resolution than karyotyping and FISH. OGM neither requires cultured cells nor amplification of DNA and hence addresses the limitations of culture and amplification biases. This study reports the clinical validation of OGM as a laboratory developed test (LDT), according to CLIA guidelines, for genome-wide SV detection in different hematological malignancies. In total, 68 cases with hematological malignancies (of various subtypes), 27 controls and two cancer cell lines were used for this study. Ultra-high molecular weight DNA was extracted from the samples, fluorescently labeled, and run on the Bionano Genomics Saphyr system. A total of 207 datasets, including replicates, were generated and 100% could be analyzed successfully. Sample data were then analyzed using either disease specific or pan-cancer specific BED files to prioritize calls that are known to be diagnostically or prognostically relevant. Accuracy, precision, PPV and NPV were all 100% against standard of care results. Sensitivity, specificity, and reproducibility were 100%, 100% and 96%, respectively. Following the validation, 11 cases were run and analyzed using OGM at three additional sites. OGM found more clinically relevant SVs compared to SOC testing due to its ability to detect all classes of SVs at much higher resolution. The results of this validation study demonstrate OGMs superiority over traditional SOC methods for the detection of SVs for the accurate diagnosis of various hematological malignancies.

BACKGROUNDCirculating cell-free DNA (cfDNA) has become a valuable analyte for molecular testing, but requires specialized collection tubes or immediate processing. We investigated the feasibility of using residual plasma from heparin separators, which are routinely used in clinical chemistry, as an accessible and underutilized source for cfDNA biobanking and testing. METHODSWe analyzed matched plasma samples from healthy volunteers in two experiments: an immediate-processing tube comparison across EDTA, Streck, and heparin separators (n = 5) and a clinical-handling simulation that paired EDTA and heparin separator tubes and delayed processing at room temperature versus 4{degrees}C (n = 6). We also analyzed matched EDTA and heparin separator plasma samples from viral PCR-positive patients (Hospital Cohort; n =38). Whole-genome sequencing and genome-wide enriched methylation sequencing were performed to evaluate concordance across multiple benchmarks, including metagenomics, chromosomal copy number, methylome, and fragmentomics. RESULTSUnder immediate processing, heparin separator plasma showed high concordance with EDTA and Streck plasma for methylation patterns (Pearsons r = 0.92-0.93, Spearmans {rho}=0.65-0.70) and fragmentation features (n = 5). In the clinical-handling simulation, cfDNA integrity in heparin separators was comparable to that in EDTA at 4{degrees}C (n=6). In the Hospital Cohort, heparin separators showed a strong concordance with matched EDTA tubes for viral detection (n=38, Pearsons r=0.96), copy number alteration profiling (n=6, Pearsons r=0.96-1.00), and methylation patterns (n=12, r=0.83-0.93). CONCLUSIONHospital residual plasma from routine clinical chemistry tests that are processed within a short pre-centrifugation window and refrigerated can provide a vast, untapped resource for cfDNA biobanking and potential testing.

The COVID-19 pandemic is challenging the global supply chain and equipment needed for mass testing with RT-qPCR, the gold standard for SARS-CoV-2 diagnosis. Here, we propose the RT-LAMP assay as an additional strategy for rapid virus diagnosis. However, its validation as a diagnostic method remains uncertain. In this work, we validated the RT-LAMP assay in 1,266 nasopharyngeal swab samples with confirmed diagnosis by CDC 2019-nCoV RT-qPCR. Our cohort was divided, the first (n=984) was used to evaluate two sets of oligonucleotides (S1 and S3) and the second (n=281) to determine whether RT-LAMP could detect samples with several types of variants. This assay can identify positive samples by color change or fluorescence within 40 minutes and shows high concordance with RT-qPCR in samples with CT [&le;]35. Also, S1 and S3 are able to detect SARS-CoV-2 with a sensitivity of 68.4% and 65.8%, and a specificity of 98.9% and 97.1%, respectively. Furthermore, RT-LAMP assay identified 279 sequenced samples as positive (99.3% sensitivity) corresponding to the Alpha, Beta, Gamma, Delta, Epsilon, Iota, Kappa, Lambda, Mu and Omicron variants. In conclusion, RT-LAMP is able to identify SARS-CoV-2 with good sensitivity and excellent specificity, including all VOC, VOI, VUM and FMV variants.

INTRODUCTIONDementia is commonly caused by underlying pathologies driven by misfolded protein aggregates. Although dementia subtypes have distinct mechanisms, overlapping symptoms make diagnosis without biomarkers difficult. Misdiagnosis has previously hindered drug development by enrolling patients non-specifically in trials. METHODSWe developed a digital Seed amplification assay (dSAA) that isolates individual aggregates in nanoliter compartments, enabling precise quantification of TDP-43 seeds in cerebrospinal fluid (CSF). RESULTSTesting 40 CSF samples from patients with genetic and sporadic FTLD-TDP, as well as healthy controls, we found elevated seed concentrations in FTLD-TDP patients that correlated with disease severity, demonstrating the potential of dSAA as a sensitive diagnostic tool. DISCUSSIONThis study demonstrates a new quantitative, high-sensitivity digital assay for TDP-43 seeds in CSF. The platforms single-aggregate resolution and low LOD establish a technical foundation for developing a diagnostic and monitoring tool for FTLD-TDP and other TDP-43-related diseases.

Morphology-based classification of cells in the bone marrow aspirate (BMA) is a key step in the diagnosis and management of hematologic malignancies. However, it is time-intensive and must be performed by expert hematopathologists and laboratory professionals. We curated a large, high-quality dataset of 41,595 hematopathologist consensus-annotated single-cell images extracted from BMA whole slide images (WSIs) containing 23 morphologic classes from the clinical archives of the University of California, San Francisco. We trained a convolutional neural network, DeepHeme, to classify images in this dataset, achieving a mean area under the curve (AUC) of 0.99. DeepHeme was then externally validated on WSIs from Memorial Sloan Kettering Cancer Center, with a similar AUC of 0.98, demonstrating robust generalization. When compared to individual hematopathologists from three different top academic medical centers, the algorithm outperformed all three. Finally, DeepHeme reliably identified cell states such as mitosis, paving the way for image-based quantification of mitotic index in a cell-specific manner, which may have important clinical applications.

IntroductionAccess to accurate, non-invasive diagnostics remains a critical unmet need in womens health. Menstrual effluence, containing endometrial tissue, immune cells, and microbial communities, represents a clinically relevant specimen for genomic and molecular pathology applications, yet has historically been underutilized due to concerns about sample integrity and variability. MethodsWe developed and validated a standardized, at-home tampon-based collection system designed to preserve nucleic acids at ambient temperature for clinical-grade analyses. 1,067 tampon samples from 328 participants underwent, RNA sequencing and metatranscriptomic profiling to assess specimen transcript integrity, diagnostic fidelity, and microbial composition over time. 12 patients were exome sequenced using matched menstrual effluence and whole blood to assess assay concordance between sample types. ResultsRNA extracted from menstrual effluence maintained stability for up to 14 days without refrigeration, achieving sufficient yield and quality for sequencing in >97% of samples. Variant detection via exome sequencing demonstrated 100% concordance among overlapping single nucleotide variants between menstrual fluid and matched venous blood, confirming clinical equivalency for genetic testing. Transcriptomic analyses revealed cycle-dependent variation in key reproductive and immune markers, while metatranscriptomic profiling identified shifts in microbial communities consistent with known reproductive tract dysbiosis. ConclusionsStandardized at-home collection of menstrual effluence provides a clinically actionable platform that supports remote specimen acquisition without compromising molecular assay fidelity, offering a scalable solution to improve access to carrier screening, reproductive health assessment, and infectious disease monitoring in clinical practice.

Accurate diagnosis and risk stratification of hematological malignancies require disease-specific laboratory testing procedures involving the use of hematopathology, flow cytometry, molecular, and cytogenetic testing. While individual laboratories develop unique workflows to accommodate volume, clinical needs, and staffing, cytogenetic laboratories generally require a multitude of targeted and genome-wide tests that detect clinically relevant aberrations in hematologic malignancies. Specifically, the frequent use of multiple FISH panels coupled with concurrent chromosome analysis, can be both labor, and resource intensive. Optical Genome Mapping (OGM) is a comprehensive cytogenetic solution for detecting structural variants with high resolution and increased accuracy for hematological malignancy subtypes at the DNA level without need of any cell culture regimens. A new software tool for analysis of OGM data called VIA (Variant Intelligence Applications), provides an integrative analysis, interpretation, and reporting solution for OGM and other datatypes. In this study, we performed retrospective review of 56 datasets, representing 10 unique myeloid cases to assess multi-user (technologist and laboratory director) analyses and classification. Interpretation and reporting of OGM results were 100% concordant between reviewers for four cases with negative results by standard of care (SOC) testing. For the other six cases, five pathognomonic gene fusions identified by SOC assays were unanimously reported as Tier 1A classification was unanimous for five sentinel gene fusion rearrangements identified by SOC. OGM also found additional structural variants of clinical relevance in five of the six cases that were not found by SOC methods. Leveraging automatic pre-classification of variants and a custom decision tree, the VIA software enabled complete analysis with a mean technologist review time (variant analysis and initial tier determination) of 30.7 minutes. The analysis, interpretation, and reporting workflow described in this pilot study provides a framework for standardized and streamlined reporting of clinically significant variant in myeloid malignancies using VIA.

Assays that account for the biological properties and fragmentation of cell-free DNA (cfDNA) can improve the performance of liquid biopsy. However, pre-analytic and physiological differences between individuals on fragmentomic analysis are poorly defined. We analyzed the impact of collection tube, plasma processing time and physiology on the size distribution of cfDNA, their genome-wide representation and sequence diversity at the cfDNA fragment-ends using shallow Whole Genome Sequencing. We observed that using different stabilizing collection tubes, or processing times does not affect the cfDNA fragment sizes, but can impact the genome-wide fragmentation patterns and fragment-end sequences of cfDNA. In addition, beyond differences depending on the gender, the physiological conditions tested between 63 individuals (age, body mass index, use of medication and chronic conditions) minimally influenced the outcome of fragmentomic methods. Our results highlight that fragmentomic approaches have potential for implementation in the clinic, pending clear traceability of analytical and physiological factors.

Early diagnosis and identification of causative pathogens using blood culture in patients suspected of Blood Stream Infection (BSI) and sepsis are critical for improving patient outcomes through early and more targeted treatment. There is a need for tools that can guide the use of microbiologic diagnostics, especially where resources are limited, such as in lower and middle income countries (LMICs), pandemic and mass-casualty scenarios, and prolonged field care settings during military operations. MethodsPost-hoc retrospective analysis of individual patient data from three prospective clinical studies, conducted in North America, Europe and Africa, to investigate the association between SeptiCyte RAPID test results (SeptiScores) and blood culture (BC) results. Hypothesisthat a significant correlation exists between elevated SeptiScores and positive blood culture results, and between low SeptiScores and negative blood culture results. ResultsThe area under the receiver operating characteristic curve (ROC AUC) was 0.91 for 85 BC(+) versus 257 SIRS, and was 0.80 for 164 BC(-) versus 257 SIRS. As the SeptiScore increases, the relative probability of a septic patient being BC(+) as opposed to BC(-) also increases. A non-linear positive correlation is observed. Below a crossover point at SeptiScore 10, the ratio of probabilities of BC(+) sepsis / BC(-) sepsis is <1 while above the crossover point this ratio is >1. Thus, septic patients with SeptiScores >10 have a higher probability of being BC(+) compared to BC(-). ConclusionsElevated SeptiScores, obtained before blood culture results, are indicative of increased blood culture positivity. This may have clinical utility, particularly in resource limited settings, as an aid for improving the efficiency of blood culture practice, for instance by informing patient selection and interpretation of blood culture results.

BackgroundCirculating tumor associated cell (CTAC) detection-based multi-cancer early detection (MCED) strategies may be hindered by the rarity of CTACs among millions of peripheral blood nucleated cells (PBNCs). We developed an advanced U-Net-based encoder-decoder model for pixel-level CTAC discrimination that integrates attention-gated skip connections to preserve morphological and fluorescence details. MethodsModel suitability was explored in an initial cohort of asymptomatic individuals (n = 428) and patients with advanced solid tumors (n = 354). A case-control study assessed clinical performance in therapy-naive stage I/II cancer patients (n = 185), individuals with benign conditions (n = 129), and asymptomatic individuals (n = 111). The model was then validated across four prospective studies on distinct populations: recurrent cancer cases with low tumor burden (n = 224); patients with solid tumors in the peri-operative setting (n = 17); suspected cancer cases (n = 259); and asymptomatic individuals (n = 7,183), respectively. All studies used blinded peripheral blood specimens from which PBNCs were isolated, stained for EpCAM / Hoechst 33342, and imaged. Ground truth annotations were established via pathologist review. The U-Net pipeline encoded spatial information in the images via convolutional and pooling layers and generated pixel-wise segmentation masks to identify CTACs. In all studies, sensitivity was based on CTAC detection rate in cancer specimens and CTAC undetectability rate in specimens from healthy asymptomatic individuals or those with benign conditions ResultsIn the exploratory study, the model had 90.68% (95% CI: 87.16%, 93.50%) sensitivity and 99.53% (95% CI: 98.32%, 99.94%) specificity. In the case-control cohort, the model had 88.65% sensitivity (95% CI: 83.17%, 92.83%), 78.95% (95% CI: 71.03%, 85.53%) specificity in benign conditions, and >99.9% specificity in asymptomatic individuals. Among the four prospective studies, the model had: (a) 91.96% (95% CI: 87.60%, 95.17%) sensitivity in pretreated patients with low tumor burden; (b) 100% sensitivity in pre-surgery specimens, and 29.41% sensitivity in post-surgery specimens; (c) 96.34% PPV (95% CI: 93.22%, 98.05%) and a 32.35% NPV (95% CI: 25.58%, 39.95%) for diagnostic triaging; and, (d)11% PPV (95% CI: 31.72%, 53.24%) and 99.97% NPV (95% CI: 99.90%, 99.99%) for MCED in healthy asymptomatic individuals. ConclusionsThe attention-enhanced U-Net achieved robust, generalizable performance for CTAC-detection in case-control and prospective cohorts, supporting its clinical utility for accurate cancer detection.

While liquid biopsy has potential to transform cancer diagnostics through minimally-invasive detection and monitoring of tumors, the impact of preanalytical factors such as the timing and anatomical location of blood draw is not well understood. To address this gap, we leveraged pet dogs with spontaneous cancer as a model system, as their compressed disease timeline facilitates rapid diagnostic benchmarking. Key liquid biopsy metrics from dogs were consistent with existing reports from human patients. The tumor content of samples was higher from venipuncture sites closer to the tumor and from a central vein. Metrics also differed between lymphoma and non-hematopoietic cancers, urging cancer-type-specific interpretation. Liquid biopsy was highly sensitive to disease status, with changes identified soon after post chemotherapy administration, and trends of increased tumor fraction and other metrics observed prior to clinical relapse in dogs with lymphoma or osteosarcoma. These data support the utility of pet dogs with cancer as a relevant system for advancing liquid biopsy platforms.

BackgroundCerebrospinal fluid (CSF) culture is the diagnostic gold standard for neuroinfectious diseases such as bacterial meningitis, but its sensitivity is limited and results are often delayed. Natural language processing (NLP) offers a powerful approach to extract meaningful clinical signals from unstructured data such as chief complaints and ICD notes. This study applies machine learning, including BioBERT-enhanced NLP models (not traditional TF-IDF approaches), to support early diagnosis and outcome prediction in ICU patients. MethodsTraining and validation datasets were derived from MIMIC-IV (internal) and MIMIC-III/eICU (external) databases. Fully connected neural network (FCNN) and other machine learning models were trained to predict CSF culture results using structured lab features. Labels were refined using clinical criteria to reduce false negatives. For ICU survival prediction, three multimodal deep learning architectures (mCNN, mFCNN, and mLSTM) were developed using two ICU survival cohorts with different inclusion criteria. The Strict ICU Survival Cohort included CSF culture results as an input feature, while the Lenient ICU Survival Cohort excluded this requirement, allowing for a broader patient population. In both cohorts, models integrated structured variables with unstructured text encoded by BioBERT, a deep contextual language model, rather than simpler methods like TF-IDF, effectively capturing clinical meaning from free-text ICD entries and chief complaints. ResultsFor CSF culture prediction (training n = 9261), the FCNN model achieved the highest performance (AUROC = 0.853) in independent validation. For ICU survival prediction in the Strict ICU Survival Cohort (training n = 5,795), the mCNN model achieved an AUROC of 0.889 in external validation. In the expanded Lenient ICU Survival Cohort (n = 58,615), the same model achieved an AUROC of 0.974 and an AUPRC of 0.868 during external validation. During model training and development, the predictive performance declined when text features were excluded (AUROC from 0983 to 0.946) or when ICD entries were converted from free-text (BioBERT-encoded) to coded format (AUROC to 0.947). ConclusionsMultimodal machine learning models, enhanced by advanced NLP through BioBERT embeddings of clinical free text, effectively predicted CSF culture results and ICU survival outcomes.