Diagnosis of sepsis in neonates

Abstract

Methods for diagnosing sepsis in neonates by detecting specific biomarkers in a sample, preferably a sample comprising saliva.

Description

[0001]Attorney Docket No.29618-0502WO1 / BWH 2024-540 Diagnosis of Sepsis in Neonates CLAIM OF PRIORITY This application claims the benefit of U.S. Provisional Application Serial No. 63 / 696,783, filed on September 19, 2024. The entire contents of the foregoing are incorporated herein by reference. FEDERALLY SPONSORED RESEARCH OR DEVELOPMENT This invention was made with Government support under Grant No. HD097081 awarded by the National Institutes of Health. The Government has certain rights in the invention. TECHNICAL FIELD Methods for diagnosing sepsis in neonates by detecting specific biomarkers in a sample, preferably a sample comprising saliva. BACKGROUND Neonatal infection and in its most severe form, sepsis, are leading causes of infant mortality worldwide1. Sepsis is defined by a systemic illness caused by a dysregulated host response to an infection with end-organ failure. SUMMARY Provided herein are methods comprising: obtaining a sample, preferably a sample comprising saliva, or alternatively blood or serum, from a mammalian neonatal subject, and determining a level of expression of biomarker proteins in the sample, preferably using an ultrasensitive protein assay, wherein the biomarker proteins comprise one, two, three, four, five, seven or more of Interleukin 1 beta (IL- 1b), IL-4, IL-5, IL-6, IL-8, IL-10, IL-12p70, IL-22, interferon gamma (IFNg), tumor necrosis factor alpha (TNFa), C-reactive protein (CRP), and total protein levels. In some embodiments, levels of up to 11, 10, 9, 8, 7, 6, or 5 biomarkers are determined. In some embodiments, levels of IL-12p70, IL-22, IL-4, IFNg, IL-6, and / or IL-10 are not determined. In some embodiments, the biomarker proteins comprise one, two, or all three of Interleukin 1 beta (IL-1b), tumor necrosis factor alpha (TNFa), and IL-6, and Attorney Docket No.29618-0502WO1 / BWH 2024-540 optionally further comprise C reactive protein (CRP), IL-5 and / or IL-8, optionally wherein the biomarkers comprise: (i) IL-4, I-L1b, IL-6, IL-10, IL-22, and IL-12p70; (ii) IL1b, TNFa, IL6, and IL8; (iii) CRP, TNFa, IL-10, IL-6, IL-12p70, and IL-22; (iv) IL-1b; or (v) IL-1b, IL-10, TNFa, CRP, IL-4, and IL-6. In some embodiments, levels of CCL20 and / or CXCL6 are determined, in addition to or as an alternative to the biomarkers listed above. In some embodiments, the methods further comprise comparing the level of expression of the biomarker proteins to a disease reference, wherein a level of expression of one, two, or more the biomarker proteins above the reference levels indicates that the subject has or is at risk of developing sepsis or an infection. In some embodiments, the subject is a human neonate of up to about 28 days after birth or up to about 52 weeks postmenstrual age (PMA). In some embodiments, the ultrasensitive protein assay is Single-Molecule Arrays (SIMOA); SIMOA Planar Arrays; Molecular On-bead Signal Amplification for Individual Counting (MOSAIC); Meso Scale Discovery (MSD); Single-Molecule Counting (SMC); nucleic acid linked immune-sandwich assay (NULISA); LUMINEX; SOMAscan Assays; mass spectrometry (optionally MALDI-MS), and / or mass cytometry (optionally CyTOF). In some embodiments, the methods further comprise recommending or sending the subject for additional evaluation, optionally by culture. In some embodiments, the methods further comprise administering a treatment for sepsis or an infection to a subject who has been identified as having or at risk of developing sepsis. In some embodiments, the treatment comprises administration of an antibiotic. In some embodiments, the methods further comprise determining a level of biomarker proteins in the subject after administration of the treatment, and comparing the level of biomarker proteins prior to treatment with the level of biomarker proteins during and / or after treatment, wherein a decrease in the level of biomarker proteins indicates that the treatment is effective in treating the infection or sepsis. Attorney Docket No.29618-0502WO1 / BWH 2024-540 Also provided herein are methods of diagnosing a subject as having or likely to develop sepsis or an infection. The methods comprise obtaining a sample, preferably a sample comprising saliva, or alternatively blood or serum, preferably comprising saliva, from a mammalian neonatal subject, determining a level of biomarker proteins in the sample with an ultrasensitive protein assay, wherein the biomarker proteins comprise one, two, three, four, five, seven or more of Interleukin 1 beta (IL-1b), IL-4, IL-5, IL-6, IL-8, IL-10, IL-12p70, IL-22, interferon gamma (IFNg), tumor necrosis factor alpha (TNFa), C-reactive protein (CRP), and total protein levels, and comparing the level of biomarker proteins to a disease reference, wherein differential expression of the biomarker proteins between the sample and the reference indicates that the subject has or is at risk of developing sepsis or an infection. In some embodiments, levels of up to 11, 10, 9, 8, 7, 6, or 5 biomarkers are determined. In some embodiments, levels of IL-12p70, IL-22, IL-4, IFNg, IL-6, and / or IL-10 are not determined. In some embodiments, the biomarker proteins comprise one, two, or all three of Interleukin 1 beta (IL-1b), tumor necrosis factor alpha (TNFa), and IL-6, and optionally further comprise C reactive protein (CRP), IL-5 and / or IL-8. In some embodiments, the biomarker proteins comprise: (i) IL-4, I-L1b, IL-6, IL-10, IL-22, and IL-12p70; (ii) IL1b, TNFa, IL6, and IL8; (iii) CRP, TNFa, IL-10, IL-6, IL-12p70, and IL-22; or (iv) IL-1b; (v) IL-1b, IL-10, TNFa, CRP, IL-4, and IL-6. In some embodiments, levels of CCL20 and / or CXCL6 are determined, in addition to or as an alternative to the biomarkers listed above. In some embodiments, the subject is a human neonate of up to about 28 days after birth or up to about 290 days gestational age. In some embodiments, the ultrasensitive assay is Single-Molecule Arrays (SIMOA); SIMOA SP-X Planar Arrays, Molecular On-bead Signal Amplification for Individual Counting (MOSAIC); Meso Scale Discovery (MSD); Single-Molecule Counting (SMC); nucleic acid linked immune-sandwich assay (NULISA); Attorney Docket No.29618-0502WO1 / BWH 2024-540 LUMINEX; SOMAscan Assays; mass spectrometry (optionally MALDI-MS), and / or mass cytometry (optionally CyTOF). Additionally provided herein are methods of diagnosing a subject as having or likely to develop sepsis or an infection. The methods comprise providing a level of biomarker proteins in a sample, preferably a sample comprising saliva, or alternatively blood or serum, from a mammalian neonatal subject, wherein the biomarker proteins comprise one, two, three, four, five, seven or more of Interleukin 1 beta (IL-1b), IL-4, IL-5, IL-6, IL-8, IL-10, IL-12p70, IL-22, interferon gamma (IFNg), tumor necrosis factor alpha (TNFa), C-reactive protein (CRP), and total protein levels, and comparing the level of biomarker proteins to a disease reference, to provide a value representing the differential expression of each biomarker protein; providing a value representing one or more medical history feature selected from weight at collection, postmenstrual age (PMA) or post-conception age (PCA), age of the mother, sex, birth weight, clinical chorioamnionitis, complete blood count (CBC), and white blood cell (WBC) counts in the subject; calculating a score using the values for the biomarkers and medical history features, wherein the score indicates that the subject has or is at risk of developing sepsis or an infection. In some embodiments, levels of up to 11, 10, 9, 8, 7, 6, or 5 biomarkers are determined. In some embodiments, levels of IL-12p70, IL-22, IL-4, IFNg, IL-6, and / or IL-10 are not determined. In some embodiments, the biomarker proteins comprise one, two, or all three of Interleukin 1 beta (IL-1b), tumor necrosis factor alpha (TNFa), and IL-6, and optionally further comprise C reactive protein (CRP), IL-5 and / or IL-8, and the medical history features comprise birth weight and PMA or PCA. In some embodiments, the biomarker proteins and medical history features comprise: (i) birth weight, sex, PMA or PCA, IL-4, I-L1b, IL-6, IL-10, IL-22, and IL-12p70; (ii) birth weight, PMA or PCA, IL1b, TNFa, IL6, and IL8; (iii) birth weight, CRP, TNFa, IL-10, IL-6, IL-12p70, and IL-22; (iv) IL-1b; or (v) birth weight, WBC, CBC, PMA or PCA, IL-1b, IL-10, TNFa, CRP, IL-4, and IL-6. In some embodiments, levels of CCL20 and / or CXCL6 are determined, in addition to or as an alternative to the biomarkers listed above. Attorney Docket No.29618-0502WO1 / BWH 2024-540 In some embodiments, the subject is a human neonate of up to about 28 days after birth or up to about 290 days gestational age. In some embodiments, the score is calculated using a machine learning algorithm. In some embodiments, the machine learning algorithm is selected from a Logistic Regression, a Linear Discriminant Analysis, a LightGradient Boosted classifier; a CatBoost classifier; and TabPFN. In some embodiments, the biomarker proteins, medical history features, and algorithm comprise: (i) birth weight, sex, PMA or PCA, IL-4, I-L1b, IL-6, IL-10, IL-22, and IL-12p70, and L2 penalized Logistic Regression; (ii) birth weight, PMA or PCA, IL1b, TNFa, IL6, and IL8, and LightGBM; (iii) birth weight, CRP, TNFa, IL-10, IL-6, IL-12p70, and IL-22 and Logistic Regression; (iv) IL-1b and Linear Discriminant Analysis (LDA); or (iv) birth weight, WBC, CBC, PMA or PCA, IL-1b, IL-10, TNFa, CRP, IL-4, and IL- 6, and LightGBM. In some embodiments, the methods further comprise recommending or sending the subject for additional evaluation, optionally by culture. In some embodiments, the methods further comprise recommending, administering or continuing administration of a treatment for sepsis or an infection to a subject who has been identified as having or at risk of developing sepsis or an infection, or stopping administration of a treatment for sepsis to a subject who has been identified as not having or at risk of developing sepsis or an infection. In some embodiments, the treatment comprises administration of an antibiotic. In some embodiments, the methods further comprise determining a level of biomarker proteins in the subject after administration of the treatment, and comparing the level of biomarker proteins prior to treatment with the level of biomarker proteins during and / or after treatment, wherein a decrease in the level of biomarker proteins indicates that the treatment is effective in treating the sepsis or an infection. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs. Methods and materials are described herein for use in Attorney Docket No.29618-0502WO1 / BWH 2024-540 the present invention; other, suitable methods and materials known in the art can also be used. The materials, methods, and examples are illustrative only and not intended to be limiting. All publications, patent applications, patents, sequences, database entries, and other references mentioned herein are incorporated by reference in their entirety. In case of conflict, the present specification, including definitions, will control. Other features and advantages of the invention will be apparent from the following detailed description and figures, and from the claims. DESCRIPTION OF DRAWINGS FIGs.1a-d. Schematic of a salivary Simoa assay. a Saliva is collected from an infant via a modified syringe and a low wall suction (collection device not pictured). For a detailed protocol description, see “Saliva acquisition” in the Methods. b Simoa consists of the formation of sandwich immunocomplexes on micron-sized beads. Substrate is added to activate a fluorescent reaction. c Each microwell array has the space to confine a single bead within each well, enabling single molecule counting. d Measurements are performed by the Quanterix HD-X Analyzer, with the output being the average number of enzymes per bead (AEB). Figure created, in part, in BioRender.com. Rolando, J. (2025) BioRender.com / u36z800. FIGs.2a-f: Optimized Simoa assay calibration curves. a CXCL6. b CCL20. c SAA1. d CXCL12. e Resistin. f LBP. For all assays, a recombinant protein standard was serially diluted to maximize analytical sensitivity, signal-to-noise ratio, and dynamic range. Error bars represent the standard deviations of triplicate technical replicates. A 4PL curve with 1 / y2weighting was fit to each curve. FIGs.3a-f: Dilution linearity of Simoa assays. a CXCL6. b CCL20. c SAA1. d CXCL12. e Resistin. f LBP. Each assay was tested by diluting three unique healthy-adult saliva samples, with an additional fourth sample spiked with the recombinant protein at a concentration in the linear regime of the assay calibration curve. Error bars represent the standard deviation of duplicate technical replicates. Concentration values were interpolated based on the assay calibration curves in FIGs.2a-f. FIGs.4a-f: Observed biomarker concentrations in saliva from uninfected and infected neonates. a CXCL6. b CCL20. c SAA1. d CXCL12. e LBP. f Resistin. Attorney Docket No.29618-0502WO1 / BWH 2024-540 Each sample measurement is the mean of duplicate technical replicates. Filled points indicate measurements within the assay dynamic range; open points indicate measurements above the upper limit of quantification (ULOQ) or below the limit of detection (LOD), as indicated by the dotted lines. Red lines indicate median concentrations. Significance level is reported as the p-value of the interval-censored parametric proportional hazard model, which accounts for data censoring at the LOD or ULOQ. FIG.5: Pearson’s correlation matrix of salivary biomarkers and demographic features. Correlation coefficients represent potential linear relationships between pairs of features from strongly negative (−1) to strongly positive (+1), with values near zero indicating no linear relationship. FIGs.6a-f: Biomarker relative abundance in saliva from uninfected and infected neonates. a CXCL6. b CCL20. c SAA1. d CXCL12. e LBP. f Resistin. Each sample measurement is the mean of duplicate technical replicates. Filled points indicate measurements within the assay dynamic range; open points indicate measurements above the upper limit of quantification (ULOQ) or below the limit of detection (LOD). Black lines indicate median concentrations. Significance level is reported as the p-value of the interval-censored parametric proportional hazards model, which accounts for data censoring at the LOD or ULOQ. FIGs.7a-f: Extended dilution linearity testing of Simoa assays. a CXCL6. b CCL20. c SAA1. d CXCL12. e Resistin. f LBP. Testing was performed as described in Figure 3. All analytes were initially diluted four-fold before subsequent two-fold serial dilution until 256-fold dilution was achieved. Omitted values represent concentrations measured below the LOD. Shading represents the linear regime chosen for spike and recovery testing. FIGs.8a-f: Observed biomarker concentrations in saliva from uninfected, septic culture-positive, and culture-negative neonates. a CXCL6. b CCL20. c SAA1. d CXCL12. e LBP. f Resistin. Each sample measurement is the mean of duplicate technical replicates. Filled points indicate measurements within the assay dynamic range; open points indicate measurements above the upper limit of quantification (ULOQ) or below the Limit of Detection (LOD), as indicated by the dotted lines. Black lines indicate median concentrations. Attorney Docket No.29618-0502WO1 / BWH 2024-540 FIG.9. Observed salivary biomarker concentrations among infected and uninfected infants. Each data point is the mean of duplicate technical replicates. **** is P < 0.0001. Salivary Biomarker data were clipped to the mean LOD and MODE ULOQ of all sample runs. Black lines indicate median concentrations. FIGs.10A-H. Model Performance and Thresholds. ROC for each of the models shown in Table 8. a, Timepoint 1, L2 penalized Logistic Regression, with 9 Features: IL-4, Birthweight, IL-1b, sex, IL-6, IL-10, IL-22, PCA, IL-12p70. b, Timepoint 1, Linear Discriminant Analysis (LDA) with all Features (n=19). c, Timepoint 1, LightGBM with 7 Features: birthweight, IL-1b, TNFa, PCA, IL-6, IL-8, and CBC; d. Timepoint 1, Linear Discriminant Analysis (LDA) with 1 Feature: IL-1b. e. Timepoint 1-Timepoint 2, Logistic Regression with 7 Features: Birthweight, CRP, TNFa, IL-10, IL-6, IL-12p70, IL-22. f. Either Timepoint, LightGBM with all Features (n=19). g. Either Timepoint, LightGBM with 10 Features: Birthweight, PCA, IL-1b, IL-10, TNFa, CRP, CBC, WBC, IL-4, IL-6. h. Either Timepoint, CatBoost with all Features. DETAILED DESCRIPTION The gold standard for the diagnosis of infection remains blood or other bodily fluid culture results that may take up to 36 hours or longer to return.49,50To shorten the time to diagnosis and limit antibiotic exposure, alternative approaches to forgo reliance on culture results have emerged. These include the development of neonatal early-onset sepsis calculators51, 52and polymerase chain reaction (PCR) platforms for the rapid identification of microbes.53Additionally, monitoring the human response to an infection through cytokine quantification has shown discriminatory promise, specifically in its specificity for accurately identifying noninfected newborns.6, 54-55Cytokine profiling is independent of type of infection (e.g. blood-borne, urinary tract, meningitis) or causative organism. However, cytokine quantification remains dependent upon serial phlebotomy and is limited by the number of cytokines that can be quantified simultaneously in newborn blood. Thus, serial serum cytokine quantification is impractical, invasive and exacerbates anemia in the newborn. While most cases of sepsis are associated with positive culture results, the inability to isolate a causative organism does not definitively exclude a sepsis diagnosis2. Such cases are referred to as “culture negative sepsis,” and often result in clinicians providing prolonged antibiotic administration at levels equal to or greater Attorney Docket No.29618-0502WO1 / BWH 2024-540 than culture-confirmed sepsis cases3. This diagnostic ambiguity has contributed to the overuse of antibiotics in neonatal intensive care units4,5. Furthermore, challenges inherent to culture-based approaches—such as slow time-to-result, high blood volume requirements (≥1 mL), and false positives due to contamination—highlight the critical need for improved sensitivity and speed in diagnostic tests for neonatal infections6. Research efforts to identify sepsis in the newborn have led to the evaluation of over 200 candidate biomarkers, with a focus on serum-derived inflammatory proteins such as acute phase reactants, cytokines, and chemokines7,8. Serum measurements of the acute phase reactants C-reactive protein (CRP) and procalcitonin (PCT) are already routinely incorporated in the clinical assessment of neonatal sepsis9. Similarly, other acute phase reactants are elevated during infection including liposaccharide binding protein (LBP)10,11and serum amyloid A1 (SAA1)12,13. In addition, several studies have supported the diagnostic value of measuring blood- based chemokines including CCL20, CXCL6, CXCL1214,15, as well as adipokines such as resistin16,17. Yet to date, no single biomarker has proven to be sufficient for the accurate and rapid diagnosis of neonatal sepsis. Instead, blood-based panels combining early- and late-phase inflammation markers are being considered as the path forward for neonatal diagnostics18. A limitation of this approach is that repeated blood sampling poses significant risks to neonates, particularly preterm infants, due to an increased risk of anemia and potential adverse effects on neurodevelopment due to painful stimuli associated with phlebotomy19,20,21. Saliva has emerged as a promising non-invasive alternative to blood for the detection of neonatal infections22,23,24. Salivary glands serve as a filtration system, allowing blood analytes to enter into the oral cavity via passive diffusion or active transport25,26. These mechanisms have enabled the detection of serum-based inflammatory proteins such as cytokines15,27,28and CRP23in saliva, with the latter demonstrating a significant correlation with serum levels22. Preliminary research has shown that measuring these biomarkers in saliva can effectively discriminate between infected and uninfected infants27,29,30,31,32,33,34,35. To date, approximately 14 inflammatory markers have been identified in neonatal saliva; however, with nearly 200 candidate biomarkers known, the diagnostic potential of the neonatal salivary proteome is only beginning to be explored. Given that analyte concentrations in saliva Attorney Docket No.29618-0502WO1 / BWH 2024-540 are reported to be 100- to 1000-fold lower than in blood36, ultra-sensitive detection methods are essential for advancing salivary diagnostics. Single-molecule array (Simoa) is a digital ELISA platform that can detect analytes in biofluids at sub-femtomolar concentrations (10−15 M), a 1000-fold improvement in sensitivity compared to traditional ELISAs37. An overview of the Simoa assay and the workflow for detecting salivary proteins can be seen in Fig.1. In Simoa, target analytes are first captured onto antibody-coated microscopic beads. The captured analyte is labeled with biotinylated detector antibody and streptavidin- conjugated β-galactosidase (SβG), forming single immunocomplex sandwiches. Individual beads are isolated into femtoliter sized wells, flooded with resorufin-β-D- galactopyranoside (RGP) substrate, and sealed with oil. The catalytic reaction of SβG and RGP results in the formation of a fluorescent product that is detectable only when a bead carries a full immunocomplex sandwich. Single molecule resolution is achieved by using a large excess of beads relative to the number of target analytes in the sample, which ensures that the percentage of beads carrying an immunocomplex follows a Poisson distribution where each bead captures either zero or one analyte molecule38. Thus, the number of analytes is determined by counting the number of “on” wells, which contain a complete ELISA complex. The proportion of active, enzyme-associated beads relative to the total detected beads is used to determine the average number of enzymes per bead (AEB). The Simoa platform was used in the present examples to quantify sepsis-associated biomarkers from salivary samples of infected and uninfected neonates. As shown in Example 1, we analyzed non-interleukin inflammatory proteins that are known to be elevated in the serum of infected infants15including the acute phase reactants serum amyloid A1 (SAA1), lipopolysaccharide-binding protein (LBP), the chemokines CCL20, CXCL6, CXCL12, and the adipokine resistin, representing a diverse range of protein types involved in the immune response to infection. These biomarkers were chosen to capture both early-phase (SAA1, resistin, CXCL6, LBP)6,7,39,40and late-phase (CXCL12, CCL20)14,15,41immune responses. The study aimed to determine if salivary biomarkers detected using ultra-sensitive assays are differentially expressed between infected and uninfected infants. As shown in Example 2, we also used cytokine quantification in predictive models to identify additional markers correlated with sepsis in neonates. Of the 1,215 Attorney Docket No.29618-0502WO1 / BWH 2024-540 rule out sepsis evaluations that informed our predictive model(s) in this observational study, nearly 90% had culture negative results but the neonates were nevertheless exposed to unwarranted antibiotics, validating the urgent need to develop improved assessment tools to limit antibiotic exposure. Consistent with the clinical definition of sepsis as a host response disease, this study hypothesized that inflammatory biomarkers, rather than identifying the causative organism itself, would be able to correctly discern between infectious status. The assay demonstrated near equal specificity and sensitivity exceeding 80% for determination of infectious status across the spectrum of gestational ages and weights in the neonatal population undergoing an infectious work-up. It is the first assay capable of simultaneous quantification of 11 cytokines in this population in a noninvasive and safe manner that is not dependent upon phlebotomy to assess a newborn’s inflammatory response. Interestingly, results from this study revealed that salivary cytokine profiling at the initiation of therapy is more informative of true infectious status than either the profiles assessed hours into therapy or changes in quantified levels between timepoints. This finding has important clinical significance, as clinicians could incorporate results from the assay at time of initial presentation to inform their decision regarding the initiation and / or duration of antibiotic therapy. The most informative cytokines on the 11-protein biomarker panel were IL-6, TNF-α, and IL-1β. These early-phase inflammatory biomarkers are not routinely used to inform care in NICU patients. Conversely, CRP, the most commonly assessed inflammatory biomarker in the neonatal population, demonstrated diagnostic inferiority on our salivary assay. Prior studies assessing the utility of neonatal serum CRP levels for the determination of infectious status have been conflicting, resulting in wide variation of CRP monitoring in NICUs. Indeed, CRP may not be the ideal cytokine for neonatal infection monitoring, as it is a late phase reactant that is more informative if followed serially. Importantly, our study validates that no single biomarker or clinical phenotype can accurately discern between infected and noninfected neonates. Rather, a panel of biomarkers provides superior diagnostic capabilities than the reliance on any single serum biomarker. The salivary panel described here is the first such panel that could be incorporated into newborn care to provide a comprehensive analysis of a newborn’s inflammatory response and may be Attorney Docket No.29618-0502WO1 / BWH 2024-540 used repeatedly in the same newborn without the risk of inducing noxious stimuli, trauma, or anemia. These studies are the first to demonstrate that cytokines, chemokines, adipokines, and acute phase reactants are present in neonatal saliva and quantifiable. These results were enabled by the development of novel Simoa assays that accurately measure these inflammatory markers in neonatal saliva with sub-picomolar detection limits; an average improvement of 44-fold over conventional well-plate ELISAs (Table 2C). Furthermore, several biomarkers were significantly elevated in infected infants, relative to uninfected infants, indicating that clinical infection status could be determined using profiles of salivary inflammatory proteins. Notably, we found that salivary CCL20 and CXCL6 levels were significantly elevated in infected infants, supporting their use as biomarkers for neonatal infectious screening. We hypothesize that biomarkers with large amounts of data clipping at the upper end of the analytical range (e.g., CCL20, CXCL6) would yield more valuable insights with further expansion of the assay’s analytical dynamic range. We have also shown that some chemokines were only marginally detectable in neonatal saliva. In particular, the low concentrations of CXCL12 were unexpected given its well-established presence in serum and involvement in bacterial sepsis14,41. CXCL12 concentrations were less than 42.0 pg / mL in both infected and uninfected neonates. We anticipate this protein should be detectable in saliva using our assays, because the reported blood concentration is 100-fold greater than the assay LOD. Other proteins in this study only had 10-fold lower concentrations in saliva than in blood, and other inflammatory markers, such as CRP, have shown a strong serum- salivary correlation in concentrations23. Interestingly, previous studies on other inflammatory conditions similarly did not show any significant elevation of salivary CXCL12 levels in cases relative to controls43,44. These lower-than-expected salivary CXCL12 levels could indicate the involvement of the chemokine in an uncharacterized pathway that prevents it from being detected in saliva compared to serum. This pathway could include a post- secretion modification or hydrolysis45. While it is known that CXCL12 is cleaved by matrix metalloproteases 2 and 946, all saliva samples in this study were immediately incubated with protease inhibitors after collection to prevent the degradation of the biomarkers. Attorney Docket No.29618-0502WO1 / BWH 2024-540 In summary, through the development of sensitive assays using the Simoa platform, we have demonstrated the ability to measure neonatal host inflammatory markers in saliva and presented two salivary biomarkers, CCL20 and CXCL6, that could be indicative of infection in neonates. Advancing salivary-based diagnostics for neonatal conditions is a crucial step towards reducing phlebotomy-associated morbidities and improving care for this vulnerable population. Biomarkers of Neonatal Sepsis As described herein, levels of certain biomarkers correlate with the presence of sepsis in neonates, including protein biomarkers as well as medial record features. The protein biomarkers (also referred to herein as biomarker proteins) can include one, two, three, four, five, six, seven, eight, nine, ten, or more of Interleukin 1 beta (IL-1b), IL-4, IL-5, IL-6, IL-8, IL-10, IL-12p70, IL-22, interferon gamma (IFNg), tumor necrosis factor alpha (TNFa), C-reactive protein (CRP), and total protein. Preferably in human subjects, human protein sequences are detected. In addition to the protein biomarkers, certain medical history elements can also be useful in diagnosing sepsis in neonates, including weight at collection, postmenstrual age (PMA) or post-conception age (PCA), age of the mother, sex, birth weight, clinical chorioamnionitis, and white blood cell (WBC) counts. In exemplary methods, the biomarkers include birthweight, age (e.g., PCA or PMA), IL-1b, TNFa, and IL-6. WBCs and / or CRP can also be evaluated, and IL-5 and / or IL-8. Methods of Diagnosis Included herein are methods for diagnosing sepsis in a neonatal subject. As measured at the population level, a biomarker can be statistically different between groups. At the individual level, a combination of markers can be used to assign a probability of infection. The methods can include assigning risk scores, even for markers that are not statistically different at the population level. However, generally speaking, markers that are statistically different at the population level are better for assigning risk scores to the individual. The methods rely on detection of protein biomarkers in saliva as described herein. The methods include providing or obtaining a sample comprising saliva from a subject and evaluating the presence and / or level (e.g., concentration) of protein Attorney Docket No.29618-0502WO1 / BWH 2024-540 biomarkers in the sample. The subjects who can be diagnosed with the present methods include mammalian neonates, e.g., a human or non-human veterinary subject, e.g., cat, dog, cow, horse, goat, or non-human primate. Preferably the subject is a human neonate, e.g., up to 28 days after birth, e.g., at least about 20, 21, 22, or 23 weeks and up to about 44, 45, 46, 47, 48, 49, or 52 weeks postmenstrual age (PMA). In some embodiments, the subject is a human neonate >23 weeks but <44 weeks postmenstrual age (PMA). “About” as used herein means plus or minus 10%. As used herein the term “sample”, when referring to the material to be tested for the presence of the protein biomarkers using a method as described herein, includes saliva. In some methods, whole blood, plasma, or serum can be used instead of saliva. The sample is preferably treated with protease inhibitors immediately after being obtained, e.g., within 5, 10, 15, 20, or 30 minutes of being obtained. One or more protease and / or phosphatase inhibitors can be used, preferably one or more (optionally all) of inhibit serine-proteases, cysteine-proteases, aspartic acid-proteases and aminopeptidases. For example, AEBSF, aprotinin, bestatin, E-64, leupeptin and / or pepstatin A can be used. As one example, a combination of the above can be used, the Thermo Scientific HALT Protease Inhibitor Cocktail which comprises all of AEBSF, aprotinin, bestatin, E-64, leupeptin and pepstatin A stabilized in high-quality dimethylsulfoxide (DMSO), or Pierce Protease tablets that include all of the above as well. The methods can include obtaining saliva from the subject, e.g., 5-500 µl , e.g., 100-250 uL saliva, and incubating a sample comprising all or a portion of the saliva, e.g., a portion of the saliva comprising 5-250 µl, 10-100 µl, 15-50 µl, 10-20 µl, 15-20 µl, or 15-25 µl of the saliva from the subject, with a capture reagent (e.g., an antibody, nanobody, or antigen-binding fragment thereof as described herein). In some embodiments, the sample is diluted, e.g., 1:1, 1:2, 1:3, 1:4, 1:5, 1:6, 1:8, 1:10, or 1:20, and any ranges therebetween having the foregoing as endpoints, e.g., 1:1 to 1:20, or 1:1 or 1:10. In some embodiments, the sample is diluted with a buffer; an exemplary sample diluent buffer is described herein, and can comprise a detergent, e.g., Triton- X 100, Tween 20, NP-40, Brij35, Brij58, or C12E8, for example, present at about 0.05%-2% of the sample. Attorney Docket No.29618-0502WO1 / BWH 2024-540 In some embodiments, the sample is contacted with the capture reagent for a time sufficient for protein biomarkers present in the sample to bind to the capture reagent, e.g., for at least 5, 10, 15, 20, 30, or 45 minutes, or at least 1, 2, 3, 4, 5, or 6 hours, up to 1, 2, 3, 4, 5, 6, 8, 10, 12, 18, or 24 hours. In preferred embodiments, the capture reagent is on a bead, e.g., a paramagnetic bead. The capture reagent bound to the protein biomarkers is then incubated in the presence of detection antibodies (e.g., an antibody, nanobody, or antigen-binding fragment thereof as described herein), and the presence and / or quantity of bound antibodies is determined. The presence and / or level of a protein biomarker can be evaluated using methods known in the art. In preferred embodiments, the methods include the use of highly sensitive or ultrasensitive and preferably multiplex detection methods including Meso Scale Discovery (MSD); Single-Molecule Arrays (SIMOA), SIMOA planar arrays; droplet digital ELISA (ddELISA); Molecular On-bead Signal Amplification for Individual Counting (MOSAIC); Single-Molecule Counting (SMC); proximity ligation-based ELISA, e.g., proximity ligation assay (PLA), proximity extension assay (PEA), and proximity proteolysis assay (PPA); nucleic acid linked immune-sandwich assay (NULISA); NOMIC nELISA Platform; Olink; LUMINEX; SOMAscan Assays; mass spectrometry (e.g., MALDI-MS) and mass cytometry (e.g., CyTOF) (see, e.g., Cohen and Walt, Chem. Rev.2019, 119, 293−321). Standard hospital / core laboratory systems, e.g., automated systems, can also be used, e.g., Roche Cobas; Siemens Healthineers Atellica, Dimensions, and Advia lines; Abbott Diagnostics Alinity or Architect systems; Beckman Coulter AU and DxC series. In some embodiments, the protein biomarker in saliva is measured using SIMOA, SIMOA planara array, or MOSAIC assays as described herein. SIMOA assays have several advantages over the conventional ELISA, the current gold standard for protein detection in blood. First, SIMOA is 1000X more sensitive than ELISA and allows for quantification of analytes present at low concentrations. SIMOA can detect protein concentrations as low as 10-19M compared to conventional ELISA’s ability to detect only 10-12M. SIMOA has a wide dynamic range that spans four orders of magnitude in concentration, and thus a single assay can be used to detect both low and high abundance markers. In some embodiments, the SIMOA technique achieves this high sensitivity by digitally counting the number of molecules Attorney Docket No.29618-0502WO1 / BWH 2024-540 in a sample by labeling and physically isolating each immunocomplex into femtoliter- sized wells. In some embodiments, mass spectrometry, and particularly matrix-assisted laser desorption / ionization mass spectrometry (MALDI-MS) and surface-enhanced laser desorption / ionization mass spectrometry (SELDI-MS), are used for the detection of biomarkers. (See U.S. Patent No.5,118,937; 5,045,694; 5,719,060; 6,225,047). In some embodiments, other methods can be used, e.g., standard electrophoretic and quantitative immunoassay methods for biomarker proteins, including but not limited to, Western blot; enzyme linked immunosorbent assay (ELISA); Enzyme-Linked Immunospot (ELISPOT); biotin / avidin type assays; protein array detection, e.g., protein microarrays; radio-immunoassay; immunohistochemistry (IHC); immune-precipitation assay; flow cytometry / FACS (fluorescent activated cell sorting); Proximity Ligation Assay (PLA); lateral flow assay; surface plasmon resonance (SPR); optical imaging; and mass spectrometry (Kim (2010) Am J Clin Pathol 134:157-162; Yasun (2012) Anal Chem 84(14):6008-6015; Brody (2010) Expert Rev Mol Diagn 10(8):1013-1022; Philips (2014) PLOS One 9(3):e90226; Pfaffe (2011) Clin Chem 57(5): 675-687; Cohen and Walt, Chem. Rev.2019, 119, 293−321). The methods typically include revealing labels such as fluorescent, chemiluminescent, radioactive, and enzymatic or dye molecules that provide a signal either directly or indirectly, or oligo labels that can be used, e.g., in MOSAIC or other digital ELISA platforms, immuno-PCR. As used herein, the term “label” refers to the coupling (i.e. physical linkage) of a detectable substance, such as a radioactive agent or fluorophore (e.g. phycoerythrin (PE) or indocyanine (Cy5)), to an antibody or probe, as well as indirect labeling of the probe or antibody (e.g. horseradish peroxidase, HRP) by reactivity with a detectable substance. The methods can also include comparing the presence and / or level with one or more references, e.g., a control reference that represents a normal level of protein biomarkers, e.g., a level in an unaffected subject, and / or a disease reference that represents a level of the proteins associated with sepsis, e.g., a level in a subject having sepsis. Suitable reference values can include detectable / undetectable levels of protein biomarkers or the median values shown in FIG.9. In some embodiments, the presence and / or level of the biomarker proteins is comparable to the presence and / or level of biomarker proteins in the disease Attorney Docket No.29618-0502WO1 / BWH 2024-540 reference, and the subject has one or more symptoms associated with sepsis, then the subject can be diagnosed with sepsis. In some embodiments, the subject has no overt signs or symptoms of sepsis, but the presence and / or level of one or more of the proteins evaluated is comparable to the presence and / or level of the protein(s) in the disease reference, then the subject has sepsis or an increased risk of developing sepsis. In some embodiments, once it has been determined that a person has sepsis, or has an increased risk of developing sepsis, then the subject can be selected or identified for further evaluation, e.g., using other diagnostics (e.g., culture), and / or a treatment, e.g., an antibiotic as known in the art or as described herein, can be selected and / or administered. Suitable reference values can be determined using methods known in the art, e.g., using standard clinical trial methodology and statistical analysis. The reference values can have any relevant form. In some cases, the reference comprises a predetermined value for a meaningful level of the biomarker proteins, e.g., a control reference level that represents a normal level of biomarker proteins, e.g., a level in an unaffected subject or a subject who is not at risk of developing a disease described herein, and / or a disease reference that represents a level of biomarker proteins associated with sepsis, e.g., a level in a subject having sepsis. The predetermined level can be a single cut-off (threshold) value, such as a median or mean, or a level that defines the boundaries of an upper or lower quartile, tertile, or other segment of a clinical trial population that is determined to be statistically different from the other segments. It can be a range of cut-off (or threshold) values, such as a confidence interval. It can be established based upon comparative groups, such as where association with risk of developing disease or presence of disease in one defined group is a fold higher, or lower, (e.g., approximately 2-fold, 4-fold, 8-fold, 16-fold or more) than the risk or presence of disease in another defined group. It can be a range, for example, where a population of subjects (e.g., control subjects) is divided equally (or unequally) into groups, such as a low-risk group, a medium-risk group and a high-risk group, or into quartiles, the lowest quartile being subjects with the lowest risk and the highest quartile being subjects with the highest risk, or into n-quantiles (i.e., n regularly spaced intervals) the lowest of the n-quantiles being subjects with the lowest risk and the highest of the n- quantiles being subjects with the highest risk. Attorney Docket No.29618-0502WO1 / BWH 2024-540 In some embodiments, the predetermined level is a level or occurrence in the same subject, e.g., at a different time point, e.g., an earlier time point. Subjects associated with predetermined values are typically referred to as reference subjects. For example, in some embodiments, a control reference subject does not have sepsis, does not have a risk of developing sepsis, or does not later develop sepsis. A disease reference subject is one who has (or has an increased risk of developing) sepsis. An increased risk is defined as a risk above the risk of subjects in the general population. In some embodiments, the level of protein biomarkers in a subject being statistically different from a reference level or range of levels of the protein biomarkers is indicative of the presence or risk of developing sepsis, and the level of protein biomarkers in a subject being substantially the same as a reference level or range of levels of the protein biomarkers is indicative of the absence of disease or normal risk of sepsis. In some embodiments, to assess whether a subject has sepsis in the clinic, the method can include first log transforming the biomarker protein values and then assigning a predicted probability, e.g., to produce a predicted probability score of having or not having sepsis. If a subject has a predicted probability score above a selected threshold, e.g., at least 50%, the subject would be predicted to have sepsis (e.g., assigned to a sepsis category). If the predicted probability score is below the selected threshold, e.g., 50%, the subject would be predicted to be healthy (e.g., assigned to a healthy category). In some embodiments, the levels of the biomarker proteins are used to calculate a score, e.g., along with one or more additional variable, e.g., gestational age. The score can be calculated, e.g., using an algorithm such as summation, or weighted summation, of the (normalized) levels of the variables. Specific algorithms can be identified using known statistical methods including PCA, linear regression, logistic regression, SVM (support vector machine), decision tree (Extra Trees Classifier, Random Forest, or gradient boosting, e.g., Light Gradient Boosting Machine, Gradient Boosting Classifier, Extreme or Gradient Boosting), KNN (K- nearest neighbors), K-means, or Quadratic Discriminant Analysis, Ada Boost Attorney Docket No.29618-0502WO1 / BWH 2024-540 Classifier, CatBoost, TabPFN, Neural Nets, Linear Discriminant Analysis, or Naïve Bayes. In some embodiments, first, saliva would be obtained from the screenee. Second, the screenee’s saliva level (e.g., concentration) of protein biomarkers in the panel would be measured, e.g., using Simoa. Third, the screenee’s predicted probability of having sepsis would be calculated based on score determined using an algorithm, optionally a logistic regression formula with a dependent variable of the natural log of [(probability of having sepsis) / (probability of not having sepsis)], and with independent variables of age and biomarker proteins. The predicted probability could then inform discussions between the parents or guardians of the screenee and physician or other healthcare provider as to how best to proceed, such as a decision that no further follow-up or treatment is necessary or to pursue confirmatory culture or administer or stop administering antibiotics. In some embodiments, the amount by which the level or score in the subject is different from (greater or less than) the reference level or score is sufficient to distinguish a subject from a control subject, and optionally is less than the level or score in a control subject. In cases where the level or score of the biomarker(s) in a subject being equal to the reference level (or score) of the biomarker(s), the “being equal” refers to being approximately equal. The predetermined value can depend upon the particular population of subjects (e.g., human subjects) selected. For example, an apparently healthy population will have a different ‘normal’ range of levels of the biomarker(s) than will a population of subjects which have, are likely to have, or are at greater risk to have, a disorder described herein. Accordingly, the predetermined values selected may take into account the category (e.g., sex, age, health, risk, presence of other diseases) in which a subject (e.g., human subject) falls. Appropriate ranges and categories can be selected with no more than routine experimentation by those of ordinary skill in the art. In characterizing likelihood, or risk, numerous predetermined values can be established. Method of Treatment Once a subject has been identified using a method as described herein as having or likely to develop sepsis, a treatment comprising antibiotics can be Attorney Docket No.29618-0502WO1 / BWH 2024-540 administered. Treatments can include administration of a beta-lactam antibiotic (e.g., ampicillin, flucloxacillin, or penicillin), optionally in combination with an aminoglycoside (e.g., gentamicin); or a cephalosporin (e.g., cefotaxime) or a glycopeptide (e.g., vancomycin) ; or a beta-lactam antibiotic (e.g., ampicillin, flucloxacillin, or penicillin) in combination with a third-generation cephalosporin agent (e.g., cefotaxime); or cephalosporins administered as a monotherapy. Combinations or single agents comprising ampicillin, gentamicin, cefotaxime, vancomycin, erythromycin, and piperacillin can be used. See, e.g., Korang et al., Syst Rev.2019 Dec 5;8:306; and Neonatal infection: antibiotics for prevention and treatment. London: National Institute for Health and Care Excellence (NICE); 2024 Mar 19. (NICE Guideline, No.195.) Available from: ncbi.nlm.nih.gov / books / NBK571222 / EXAMPLES The invention is further described in the following examples, which do not limit the scope of the invention described in the claims. Example 1. Ultrasensitive detection of sepsis-associated proteins in neonatal saliva Sepsis is a life-threatening condition that affects millions of newborns every year. Current diagnostic practices require painful, often repetitive, blood collections to isolate and culture the causative microorganism. Definitive return of results can take up to 5 days, exposing newborns to potentially unnecessary antibiotics. Developing a more rapid, non-invasive assessment of infectious status based upon host immune response provides an alternative diagnostic approach to infection screening. However, additional blood collection for multiplex quantification of inflammatory biomarkers in neonates remains prohibitive. Alternatively, saliva may serve as an informative biofluid, though detection of biologically relevant proteins in saliva remains challenging, with analyte concentrations 100 to 1000 times lower compared to blood. To address this challenge, we used ultra-sensitive single-molecule array (Simoa) assays to evaluate inflammatory biomarkers known to be associated with neonatal sepsis: serum amyloid A1 (SAA1), lipopolysaccharide-binding protein (LBP), chemokines CCL20, CXCL6, CXCL12; and the adipokine resistin. When these assays Attorney Docket No.29618-0502WO1 / BWH 2024-540 were tested on 40 neonatal salivary samples, we found that CXCL6 and CCL20 were significantly elevated in infected and septic neonates compared to those uninfected. The small volumes of fluid used (~10 μL saliva) and limited concentrations of these analytes in saliva (pg / mL) underscore the importance of ultra-sensitive measurements in the development of non-invasive diagnostics and reinforces the value of neonatal saliva as a viable sample matrix for monitoring infection. These assays can be used in diagnostic tests to discriminate between infected and uninfected neonates. Materials and Methods The following materials and methods were used in the Examples below. Saliva acquisition With Institutional Review Board approval and parental consent, neonatal saliva samples were collected at the initiation of an infectious work up requiring antibiotic administration and 18–36 h into therapy. All samples were obtained from neonates enrolled in Salivary Profiling in Infants Treated for Suspected Sepsis: The SPITSS Study (NCT05490212) and quantified in accordance with a Mass General Brigham Institutional Review Board protocol. Saliva samples were collected with previously described techniques23. Briefly, the oropharynx was gently suctioned with a 1 mL syringe attached to low wall suction. Saliva was immediately stabilized at the bedside with a HALT protease inhibitor cocktail and placed at −80 °C pending analysis. All samples were collected in duplicate. If a sample appeared either bloody or contaminated with breast milk or formula, the sample was discarded and an additional sample was obtained. Demographic data including gestational age, sex, and weight were recorded. Clinical outcome data including blood culture results and duration of antibiotic exposure were also recorded and used to determine infectious outcome. Preparation of antibody-coated beads ELISA DuoSet kits were obtained from R&D Systems. Capture antibodies (CRP (R&D Systems DY1707), CCL20 (R&D Systems DY360), CXCL6 (R&D Systems DY333)) were coupled to 2.7 µm carboxylated paramagnetic beads (Quanterix) using 1-ethyl-3-(3-dimethylaminopropyl) carbodiimide (EDC) hydrochloride chemistry (ThermoFisher Scientific). First, lyophilized capture antibody (R&D Systems) was reconstituted in bead conjugation buffer (Quanterix). Attorney Docket No.29618-0502WO1 / BWH 2024-540 4.2 × 108beads were washed three times using bead wash buffer (Quanterix) and three times in bead conjugation buffer. The beads were incubated with EDC (0.15 mg / mL) for 30 min with rotation at 4 °C. Afterwards, beads were washed once in bead conjugation buffer and capture antibody (0.15 mg / mL) was added. The beads and antibody were incubated for 2 h with rotation at 4 °C. The beads were then washed two times in bead wash buffer and incubated in bead blocking buffer (Quanterix) for 45 min. Afterwards, the beads were washed once in bead wash buffer and twice in bead diluent (Quanterix) and placed in tinted vials for long-term storage at 4 °C. The next day, the coupling efficiency for all beads was calculated. Coupling efficiency was defined as the amount of residual antibody in the bead wash solutions, compared to the amount of starting capture antibody. The amount of antibody in the bead wash solutions was quantified using a NanoDrop OneC Spectrophotometer (ThermoFisher Scientific). The fraction of beads existing as monomers was measured using a Z2 Coulter Particle Count and Size Analyzer (Beckman Coulter). Successful bead conjugations had a high fraction of monomeric beads (≥80%). Calibration curves To generate eight-point calibration curves, recombinant proteins for each biomarker (R&D Systems) were serially diluted in a sample-buffer solution. This solution consisted of StartingBlock Blocking Buffer, 0.5 M EDTA solution (1X), and HALT Protease Inhibitor Cocktail (1X) (ThermoFisher Scientific). Initial Simoa reagent concentrations used to generate the calibration curves were 150 pM streptavidin-β-galactosidase (Quanterix) and 0.2 µg / mL biotinylated detector antibody (R&D Systems). Subsequently, each assay was optimized by titrating both the detector antibody concentrations (0.1 and 0.2 µg / mL) and streptavidin-beta- galactosidase (SβG) enzymatic label (25, 100, and 150 pM). All assays were run as two-step assays with default incubation times. In the Simoa two-step assay configuration, a sample is first incubated with capture beads and detector antibody, and then SβG is added in the second step. Homebrew Helper beads, passivated beads that lack any conjugated capture antibody on the surface, were also used in all assays to improve resuspension of beads and ensure a consistent bead fill. A 1:1 ratio of helper to active beads was used for a total of 500,000 beads per assay, as recommended by the manufacturer. All measurements were taken in triplicate using the Quanterix HD-X Analyzer instrument and the mean concentration was calculated Attorney Docket No.29618-0502WO1 / BWH 2024-540 by the HD-X software. Average enzyme per bead (AEB) values were also calculated by the HD-X software. Prism 10 (GraphPad Software) was used to fit a 4-parameter logistic regression (4PL) curve with 1 / y2weighting and interpolate concentration values from the AEB measurements. LOD values were calculated as the mean signal of the blank plus three times the standard deviation of the noise. ULOQ was defined as the most concentrated calibrator in each assay. Dilution linearity Healthy adult saliva samples (BioIVT) were serially diluted six times (initially 4-fold, and then 2-fold for every subsequent dilution) in the sample-buffer solution described above, and recovered signal was measured (parallelism). Percent linearity was calculated by dividing the measured signal between concentrations while correcting for the dilution factor. An assay demonstrated good linearity if measurements agreed within 80% to 120% of the subsequent expected concentration, e.g. half of the previous concentration for 2-fold serial dilutions. Three samples were measured per assay, with an additional fourth sample spiked with the standard protein to represent a highly concentrated sample. Each sample was measured in duplicate technical replicates. Sample dilution factors were chosen based on the dilution factor where the signal linearized. The percent linearity was calculated using the following equation: Spike and recovery Recombinant protein standard was spiked into diluted healthy-adult saliva samples (BioIVT) at low, medium, and high concentrations relative to each assay calibration curve. The dilution factor of the saliva was established through dilution linearity testing. The observed recovered protein signal was measured using three distinct saliva samples. A control was also included in each spike and recovery test to account for user error in the execution of serial dilutions. The control experiment consisted of spiking the same three concentrations into the sample-buffer solution, rather than saliva, and measuring recovered signal to account for the impact of sample matrix on quantification. Each sample was measured in duplicate technical replicates. The assay demonstrated good recovery if the observed concentration was within 80% Attorney Docket No.29618-0502WO1 / BWH 2024-540 to 120% of the expected spiked concentration. The percent recovery was calculated using the following equation: The patient cohort (n = 20 subjects providing 40 samples) was run in two batches (n = 20), and the same lot of each reagent was used across all samples except for the prepared sample buffer. Samples were thawed at 4 °C for 15 min, before subsequent centrifugation for 30 min at 4 °C at 21,000 RCF. Each batch of samples included three quality controls per assay to determine run-to-run variability. Calibration curves were generated the day prior to the first day of sample processing and all sample concentrations were determined from these curves. The cohort was matched to the clinical study and designed to ensure an equal number of infected and uninfected samples. Operators of the HD-X were blinded to the cases status of each sample (infected or uninfected) until all measurements were disclosed. All sample measurements were taken in technical duplicates using the Quanterix HD-X Analyzer instrument and the mean concentration was calculated by the HD-X software. Average enzyme per bead (AEB) values were calculated by the HD-X software. Total Protein in neonatal samples was quantified using a NanoDrop OneC Spectrophotometer (ThermoFisher Scientific). Statistical analyses Interpolated concentration values were calculated from each assay calibration curve using a 4PL curve with 1 / y2weighting. We used the icenReg package in RStudio to perform an interval-censored parametric proportional hazard analysis, starting from a Weibull baseline distribution, to assess the difference between biomarker concentrations in infected and uninfected neonates. Interval censoring reflects the limits of detection and upper limits of quantification of our assays, treating them as left- and right-censored, respectively. The analysis provided estimates for the log hazard ratio, standard error, z-value, and two-tailed level of significance for each biomarker. To evaluate the statistical power of the regression results, we performed a Attorney Docket No.29618-0502WO1 / BWH 2024-540 z-test. The observed z-value was calculated as the log hazard ratio divided by its standard error. Statistical power was calculated as 1−β, where β represents the Type II error rate, using a two-tailed test at a significance level (α) of 0.05. All measurements outside the assay range were assigned values of the LOD or ULOQ, or zero and positive infinity for censoring, as appropriate. All samples were treated as independent, regardless of time of collection and whether the sample came from the same infant at different timepoints. This approach was chosen to maximize statistical power of the limited sample size. Statistical analyses were conducted in Prism 10 (GraphPad Software); Python 3.10 using nupy 1.26.4, scikit-learn 1.5.2, and statsmodels 0.14.4; or RStudio using icenReg v2.0.16 for interval censored data48. Example 1.1. Neonatal participants Twenty neonates participated in this study, providing 40 saliva samples for analysis. Pertinent demographic and clinical data can be found in Table 1A. Forty-five percent of participants were deemed to be infected based upon positive blood culture results and / or prolonged antibiotic administration (Table 1A). Among the infected category, half of the infants had culture positive results for bacteria, while half of the infants were treated for five or more days with antibiotics, indicating symptoms of a clinical infection despite negative culture results (Table 1B). Table 1A Demographic and clinical characteristics of sample cohort Uninfected Infected s an ar ev a on. Table 1B. Demographic and clinical characteristics of infected and uninfected samples. Uninfected Culture-positive Culture-negative Attorney Docket No.29618-0502WO1 / BWH 2024-540 Male (%) 20 20 80 Birth weight 1821± 752 921 ± 785 2979 ± 1494 Example 1.2. Assay development We developed a panel of six Simoa assays to detect inflammatory biomarkers in saliva, related to the neonatal immune response to infection. Each assay was evaluated by performing a serial dilution of the recombinant protein to generate eight- point calibration curves. Each assay was also optimized for the most sensitive limit of detection (LOD), highest signal-to-noise ratio, and largest dynamic range by titrating both the detector antibody concentrations and streptavidin-beta-galactosidase (SβG) enzymatic label. We defined the LOD as the mean signal of the blank plus three times the standard deviation of the noise. Signal-to-noise ratio was calculated as both the ratio of the mean signal of the highest calibrant to the blank, as well as the lowest detectable calibrant to the blank. We defined the dynamic range as the difference in concentrations between the lowest detectable calibrant to the smaller of 15 AEB, or the upper asymptote of the 4PL regression. Through this systematic process of optimization, we enhanced the performance of the Simoa assays, to yield the optimized calibration curves (Fig.2), and final assay conditions (Table 2B). Titrating both SβG and detector was necessary for assays with initial background signals less than 0.005 AEB, such as CXCL6 and CCL20, where low background AEB would result in assays with unstable LODs. Antithetically, assays for SAA1 and resistin required decreasing reagent concentrations to expand the dynamic range, as the fluorescence signal intensity at the upper end of the calibration curves exceeded the quantification limit of the HD-X Analyzer. Compared to the commercial ELISA kits, these assays show, on average, 44-fold improvements in analytical sensitivity (Table 2C). To validate the assays’ performance in saliva, we performed dilution linearity of both spiked and endogenous proteins (parallelism) in commercial healthy-adult saliva (Fig.3). We acknowledge that there are differences between neonatal and adult saliva compositions (protein profiles, enzymatic content, electrolyte balance), however, adult saliva was used as a surrogate due to the limited availability of Attorney Docket No.29618-0502WO1 / BWH 2024-540 neonatal saliva and to facilitate assay optimization. For dilution linearity testing, all analytes were initially diluted four-fold before subsequent two-fold serial dilution until 256-fold dilution was achieved (Figure 7). A “linear regime” was selected if the signal from three adjacent dilutions yielded 80–120% of the expected recovery range for most samples, as recommended by the FDA Bioanalytical Method Validation Guidance for Industry guidelines42. Recovered signals for the majority of biomarkers were measured within the 80–120% linearity, over a wide range of concentrations, indicating minimal interfering matrix effects from saliva with the antigen-antibody binding mechanism of Simoa (Fig.3). Spike-and-recovery was performed at low, medium, and high concentrations to cover the full range of expected endogenous concentrations by adding recombinant protein to three separate healthy-adult saliva samples, each at a selected dilution factor. Spike recovery was calculated by subtracting the endogenous signal of the saliva from the observed concentration after spiking, then dividing by the expected concentration of the spiked protein. An assay demonstrated good recovery if the Percent Recovery fell between 80 and 120% of the expected spiked concentration42, which we observed for most analytes (Table 2). The assays for CXCL12 and resistin each had one spike concentration fall below the desired 80–120% range; however, since the other measurements fell between this range, we chose to move forward with these two biomarkers, taking note of this limitation while quantifying clinical samples. Table 2A Spike and recovery of Simoa assays Biomarker Spike 1 Mean % Spike 2 Mean % Spike 3 Mean % ry a 3 Attorney Docket No.29618-0502WO1 / BWH 2024-540 Each spike concentration was chosen from the upper, lower, and center ranges of the assay calibration curve. The calculated percent value is the mean of three samples. Each sample was measured in duplicate. Table 2B: Simoa assay conditions. Serial Dilution Volume of Detector SβG Factor in Sample Biomarker Ab Dilution stabilized saliva Table 2C: Simoa assay performance characteristics and comparison with commercial ELISA performance. Simoa Simoa Dilution Simoa Dilution Simoa LOD Fold ia . . LOD data were gathered from R&D Systems. Example 1.3. Quantification of protein analytes from neonatal saliva Simoa assays detected and quantified all biomarker candidates in neonatal saliva in the majority of clinical samples. Concentrations generally fell within the quantifiable range of the assay. However, CXCL12 was detected in only 18 of 40 samples, indicating even the ultra-sensitive Simoa had insufficient analytical sensitivity for this biomarker. In cases where biomarkers could not be measured, the concentrations were typically below the assay’s limit of detection (LOD) (Table 3). Attorney Docket No.29618-0502WO1 / BWH 2024-540 Conversely, biomarkers CCL20, CXCL6, SAA1, and resistin were occasionally measured at concentrations exceeding the assay upper limit of quantification (ULOQ), defined as the most concentrated calibrator. In total, out of 240 possible measurements (six biomarkers per sample for 40 samples), 61.7% of measurements were successfully quantified (Table 3). Table 3 Biomarker assay performance Samples detectable Samples above ULOQ Samples below LOD (%) (#) (#) The median concentrations of most of the tested biomarkers were elevated in infected infants relative to uninfected infants, with the log hazard ratio of biomarkers CXCL6, CCL20, and SAA1 being significantly elevated, p < 0.05, and sufficiently powered, 1−β > 0.8 (Fig. 4). LBP showed a trend towards significance (p = 0.06), but its statistical power was insufficient to draw definitive conclusions (1−β = 0.468). Future studies with larger sample cohorts may help confirm differential expression of this protein. For CXCL12, resistin, and total protein, no significant differences were observed between infected and uninfected groups (p > 0.05), and their respective power values indicate limited capacity to detect meaningful associations in the current dataset. The specific median concentrations, interquartile range (IQR), log hazard ratio, two-tailed level of significance, and power are reported in Table 4. Additional information on replicate measurements and their corresponding coefficient of variation (CV) percentages are available as a supplementary file. Within the infected infants, we found that the median biomarker concentration was elevated in the culture-positive group compared to the culture-negative group, except for CXCL12 (FIGs.8a-f). Attorney Docket No.29618-0502WO1 / BWH 2024-540 Table 4 Median biomarker concentrations in uninfected and infected neonatal saliva Bio- Uninfected Infected median Log Std. z-value p (two- Power marker median and and [IQR] hazards error tailed) (1−β) 3 he interquartile ranges (IQR) are reported for uninfected (n = 22) and infected (n = 18) neonates. Significance level is reported as the p-value of the interval-censored parametric proportional hazard model, which accounts for data censoring at the LOD or ULOQ. Example 1.4. Statistical examination of possible confounders to protein analytes To assess if demographic differences between the infected and uninfected groups, or relationships among multiple features (biomarkers or demographics data), influenced the observed protein concentrations, we performed a series of analyses including a Pearson Correlation Matrix, Multiple Linear Regression by Ordinary Least Squares, and Variance Inflation Factor (VIF) calculations. The bivariate Pearson’s Correlation Matrix (Fig.5) was calculated to highlight linear relationships between features. This analysis showed a strong positive correlation between birth weight and gestational age, a biologically plausible finding. Similarly, a strong positive correlation between CCL20 and CXCL6, suggesting potential collinearity. In addition, CCL20 and CXCL6 both exhibited moderate Attorney Docket No.29618-0502WO1 / BWH 2024-540 negative correlations with gestational age and birth weight, and moderate positive correlation with LBP. We next assessed if the demographic features gestational age, birth weight, or sex were possible confounders to the observed biomarker concentrations. Multiple Linear Regression (MLR) analyses were performed using Ordinary Least Squares (OLS), with infection status as the primary predictor and gestational age, birth weight, and sex included as potential confounders. Model-level metrics, including R-squared, adjusted R-squared, F-statistics, and their associated p-values, were used to evaluate the explanatory power of the predictors for each biomarker (Table 5A). The R- squared values ranged from 0.060 (CXCL12) to 0.474 (CXCL6), indicating varying degrees of variance explained by the predictors. Adjusted R-squared values were lower for most biomarkers, reflecting the modest contribution of the confounders after accounting for the number of predictors. Significant F-statistics for CCL20 (p = 2.60 × 10−4), CXCL6 (p = 1.21 × 10−4), LBP (p = 0.023), and SAA1 (p = 0.011) indicate that the included predictors collectively explained a significant proportion of the variance for these biomarkers. Table 5A: Model-level metrics of Ordinary Least Squares (OLS) Multiple Linear Regression (MLR) with Possible Confounders: Gestational Age, Birth Weight, and Sex. Biomarker R- Adj. R- F- Prob. (F- squared squared statistic statistic) AIC BIC 2 4 7 4 5 9 7 nd poss e co ou e s suc as ges a o a age, e g , a se , a e a jus ed R-squared accounting for the number of predictors. The F-statistic tests whether at least one predictor significantly explains the observed variance, with its associated probability (p-value) indicating the statistical significance of the overall model. Predictor-level metrics provided further insight into individual contributions to the biomarker concentrations (Table 5B). Infection status, the primary hypothesis of this manuscript, was significantly associated with CCL20 (β = 619, p = 0.049) and CXCL6 (β = 3240, p = 0.027), indicating infection increases the observed Attorney Docket No.29618-0502WO1 / BWH 2024-540 concentrations of these biomarkers; a finding consistent with the interval-censored parametric proportional hazard model. Despite both CCL20 and CXCL6 exhibiting moderate negative correlations with gestational age and birth weight in the Pearson Correlation Matrix, neither was significantly associated with these features in the MLR analysis (p > 0.05). Conversely, gestational age and birth weight were significant predictors for SAA1 (β = −542, p = 0.023 and β = 23, p = 0.004, respectively), suggesting that the observed significance in the interval-censored model was due to demographic confounders. The remaining biomarkers showed no statistically significant associations with infection status or confounders. Table 5B: Predictor-level metrics of Ordinary Least Squares (OLS) Multiple Linear Regression (MLR) with Possible Confounders: Gestational Age, Birth Weight, and Sex. Bio- Co- t- k Predictor efficient Std. E Valu p- 95% CI 95% CI V l L U Attorney Docket No.29618-0502WO1 / BWH 2024-540 S 9040 9720 0.93 0.359 -10700 28800 Geexstationa g . , , the statistical significance of the relationship. Significant results (p < 0.05) indicate a statistical association with the possible confounder. A statistically significant intercept indicates that the observed concentration of a biomarker is non-zero in the reference group (in the absence of possible confounders). Variance inflation factor (VIF) calculations were performed to assess collinearity among multiple predictors (Table 5C). The VIF values indicated potential collinearity between birth weight and gestational age when all features are included, consistent with the strong positive correlation observed between these two features in the Pearson’s correlation matrix. Excluding either of these features from the analysis reduces their respective VIFs, confirming their collinearity. In addition, large VIF values for CCL20 (18.0) and CXCL6 (20.7) suggest significant collinearity, likely due to their strong positive correlation as shown in the Pearson’s correlation matrix (see Attorney Docket No.29618-0502WO1 / BWH 2024-540 FIG.5). Excluding either CCL20 or CXCL6 substantially reduced their respective VIFs (to 4.01 and 4.59, respectively), demonstrating the impact of collinearity on the regression models. Notably, LBP, which appeared moderately correlated in the Pearson’s Correlation Matrix, was not clearly correlated with other predictors in the regression analysis. For the remaining features, such as CXCL12, SAA1, LBP, resistin, and total protein, VIF values were below 3, indicating that these predictors do not contribute significantly to multi-collinearity. Table 5C. Variance inflation factor (VIF) of salivary biomarkers and demographic features Feature All Exclude Exclude Exclude Exclude birth CCL20 CXCL6 gestational age weight e n ercep row represen s e or e mo e n ercep an s n uence y e o her features. High VIF values indicate potential multi-collinearity among predictors, with VIFs above 10 generally considered problematic. N / A indicates that the variable was excluded from the respective model. Attorney Docket No.29618-0502WO1 / BWH 2024-540 Example 1.5. Relative abundance of target biomarkers to total protein content To determine if salivary production rates or hydration status had an impact on the observed concentration of biomarkers, the observed salivary protein concentration was divided by the total protein concentration in the saliva to obtain the relative abundance of the target analyte protein to total protein concentration. The process of normalizing protein concentration did not affect the significance, as the log hazard ratio of the relative abundance of CCL20, CXCL6, and SAA1 remained significantly elevated in infected neonatal saliva relative to the uninfected group (Fig.6, Table 6). LBP became statistically significant (p = 0.021), but insufficient power limited the robustness of this finding. For the remaining analytes, relative abundance did not show statistical difference between the infected and uninfected groups, consistent with the findings before the normalization process. In total, these results emphasize that the observed differences in biomarker concentrations were independent of variations in total protein content, and thus, could be markers of infection. Table 6. Biomarker median relative abundance in uninfected and septic neonatal saliva Biomarker Uninfected Infected Log Std. z- p (two- Power i i h l t il 1 β) 7 0 7 7 7 Attorney Docket No.29618-0502WO1 / BWH 2024-540 Biomarker Uninfected Infected Log Std. z- p (two- Power median and median and hazard error value tailed) (1−β) 3 and interquartile ranges (IQR) are reported for uninfected (n = 22) and infected (n = 18) neonates. Significance level is reported as the p-value of the interval- censored parametric proportional hazards model, which accounts for data censoring at the LOD or ULOQ. Example 2. Salivary Profiling in Infants with Suspected Sepsis (SPITSS) Materials and Methods The following materials and methods were used in the Examples below. Study: The Salivary Profiling in Infants Treated for Suspected Sepsis (SPITSS) trial was a prospective, observational study with central Institutional Review Board (IRB) approval (Tufts Medical Center, Boston, MA IRB: 13372) and local IRB approval from participating sites (Women & Infants Hospital of Rhode Island, Providence, RI and University of Florida, Gainesville, FL). Given the noninvasive nature of saliva collection and the continuous assessment of infection in a neonatal intensive care unit (NICU), each site was granted approval to collect saliva prior to parental consent. Samples could only be analyzed once parental consent was obtained; saliva samples collected from infants whose parents declined the study were destroyed. Eligibility: Hospitalized infants were eligible for participation if they were >23 weeks but <44 weeks postmenstrual age (PMA) at time of sample collection. All infants undergoing an infectious work-up with administration of antibiotic, antifungal and / or antiviral therapy were eligible for enrollment. An infant could be enrolled regardless of infectious organism (bacterial, fungal, viral) or type of suspected infection (blood- borne, urinary tract, cerebral spinal fluid, lung, peritoneum or skin). Infants with a noninfectious, life-threatening condition were excluded. Attorney Docket No.29618-0502WO1 / BWH 2024-540 Saliva Collection and Stabilization: Saliva samples were collected at the initiation of antibiotic therapy (within 6 hours) and 18-36 hours into treatment with previously described techniques.21, 22Two samples were collected at each time point. For ease of sample collection, premade collection kits were prepared and distributed to each site and stored in 4°C refrigerators housed at each NICU. Each kit contained two 1 mL syringes for saliva collection and two 1.5 mL Eppendorf LoBind tubes containing a saliva protein stabilization matrix (e.g., StartingBlock Blocking Buffer, 0.5 M EDTA solution (1X), and Halt Protease Inhibitor Cocktail (1X). See Example 1). Each collection tube was barcode labeled with the eLabInventory tracking system (eLabNext, Groningen, NL); each site was equipped with a barcode scanner to ensure proper chain of custody for each sample. Once saliva was collected and stabilized, it was immediately frozen in a -80°C freezer housed within each NICU. Batched samples were shipped to Brigham and Women’s Hospital (Boston, MA) on dry ice for multiplexed cytokine analysis on the single molecule array (Simoa) platform, a digital immunoassay platform that uses single-molecule arrays to detect biomarkers in small fluid samples or Simoa SP-X Planar Arrays (Quanterix, Billerica, MA USA). Infectious Classification: Neonates were classified into three infectious categories based on culture results and antibiotic therapy duration: 1.) infected: positive culture results and antibiotics administered > 5 days; 2.) noninfected: negative culture results and antibiotics halted at ≤48 hours; 3.) clinically infected: negative culture results but antibiotics administered >48 hours. Given the ambiguity of true infectious status in those classified as clinically infected, these infants were excluded from the model. Infants with pneumonia or necrotizing enterocolitis (NEC) were also excluded from model development. Clinical laboratory values were linked with the dataset if the clinical-laboratory-based C-reactive protein (CRP) measurement was run within 6 hours of saliva sampling, and white blood cell count was quantified within 12 hours of saliva collection. Salivary Cytokine Quantification: Research-derived C-reactive protein (CRP) was quantified using Simoa on the HD-X platform (Quanterix), and ten saliva cytokines were quantified on the SP-X platform (Quanterix): tumor necrosis factor-alpha (TNF-α), interferon gamma (IFN-γ) Attorney Docket No.29618-0502WO1 / BWH 2024-540 and interleukin 1 beta (IL-1β), IL-4, IL-5, IL-6, IL-8, IL-10, IL-12p70 and IL-22. Total protein was used as a feature to normalize samples for varying starting volumes. Sample preparation Samples were stored at –80 °C until analysis. For each subject, two to four aliquots (~250 μL each) were available (representing two aliquots collected simultaneously from the same infant, and two aliquots 18-36 hrs later for a subset of the infants). One aliquot per subject was thawed for approximately 1 hr at 4 °C on a Hula Mixer. Thawed samples were centrifuged at 21,000 × g for 30 minutes at 4°C to pellet debris. The supernatant was aspirated from the meniscus and diluted directly into the dilution buffer. If the sample was too viscous to accurately aspirate, the centrifugation step was repeated. Samples were maintained on ice throughout processing. Operators were blinded to the case status of each sample (infected or uninfected) until all measurements were completed and submitted. All sample measurements were performed in technical duplicates and the mean concentration was calculated automatically by either the SP-X or HD-X software. SP-X quantification Cytokine concentrations were measured using the Quanterix Human CorPlex Cytokine Panel 1 (a 10-Plex Array of IFN-γ, IL-1β, IL-4, IL-5, IL-6, IL-8, IL-10, IL- 12P70, IL-22, and TNFα). Briefly, recombinant standards for the ten cytokines were reconstituted in Sample Diluent according to the manufacturer’s instructions, with the exceptions of IL-1β and IL-8. To adjust the range of the calibration curve to better fit endogenous concentrations in neonatal saliva, the concentration of the IL-1β standard was increased two-fold, and the concentration of the IL-8 standard was increased six- fold. Protein standards were combined to create a single calibrator stock, and a calibration curve was prepared by four-fold serial dilution. Processed samples were diluted 1:4 by mixing 40 μL of sample with 120 μL of Sample Diluent, followed by brief vortexing and pulse-spinning. The plate, plate lid, and SP-X Wash Buffer were prepared according to the manufacturer’s instructions. 50 μL of each calibrator or diluted sample were added to the pre-coated 96- well plate in duplicate and incubated for 2 hours at room temperature on a Quanterix Simoa Microplate Shaker (525 rpm). Following incubation, plates were washed using Attorney Docket No.29618-0502WO1 / BWH 2024-540 the Quanterix Simoa Microplate Washer with four cycles comprising 300 µL of SP-X Wash Buffer dispensed and aspirated per cycle. Residual Wash Buffer in the plate was removed by blotting. Next, 50 µL of Biotinylated Antibody Reagent was added to each well, followed by a 30-minute incubation at room temperature with shaking (525 rpm). Plates were washed again using the previous protocol. Subsequently, 50 μL of Streptavidin-HRP conjugate was added to each well and incubated for 30 minutes under the same conditions. During this final incubation, the chemiluminescent substrate was prepared by gently mixing 3 mL of Stable Peroxide with 3 mL of SuperSignal Luminol Enhancer. After incubation, plates underwent 5 wash cycles (each consisting of four dispenses and aspirations of 300 µL of SP-X Wash Buffer). Immediately prior to imaging, 50 μL of the prepared SuperSignal Substrate was added to each well. Luminescence was measured within 5 minutes using the Quanterix SP-X Imaging System. Between plate assay performance was evaluated by verifying the consistency of the limit of detection (LOD), lower limit of quantification (LLOQ), upper limit of quantification (ULOQ), and R2of the calibration curves. Plates that demonstrated variation in any of these categories were remeasured. HD-X quantification The lyophilized capture antibody, biotinylated detector antibody, and recombinant protein standard for CRP were purchased from R&D Systems (DY1707). For each batch, 2.8×108488-dyed singleplex beads (Quanterix) were washed three times with 200 µL of Bead Wash Buffer (Quanterix), followed by three washes with 200 µL of Bead Conjugation Buffer (Quanterix). In the final wash step, the beads were resuspended in 190 µL of Bead Conjugation Buffer. A 10 mg / mL 1-ethyl-3-(3-dimethylaminopropyl) carbodiimide (EDC, ThermoFisher Scientific) solution was freshly prepared by dissolving 1 mg EDC in 100 µL of Bead Conjugation Buffer.10 µL of this EDC solution was quickly added to the bead suspension, vortexed to create a final concentration of 0.5 mg / mL EDC, and incubated for 30 minutes at room temperature while shaking. Activated beads were washed once with 200 µL of Bead Conjugation Buffer and then resuspended in 200 µL of 0.5 mg / mL CRP capture antibody (100 µg total) and incubated for two hours at room temperature with shaking. Following conjugation, beads were washed twice Attorney Docket No.29618-0502WO1 / BWH 2024-540 with 200 µL Bead Wash Buffer, and then incubated with 200 µL Bead Blocking Buffer for 30 minutes at room temperature with shaking. After blocking, beads were washed once with 200 µL Bead Wash Buffer, followed by a final wash with 200 µL Bead Diluent. CRP-conjugated beads were stored at 4 °C until use. Processed samples were diluted 1:50 by mixing 6 μL of sample with 294 μL of Starting Block PBST (ThermoFisher Scientific), followed by brief vortexing and pulse-spinning. Calibration curves were prepared in triplicate using a five-fold serial dilution from 1000 pg / mL. Internal control samples of 10 pM and 200 pM (corresponding to ~0.1 and ~2 AEB, respectively) were included in each run. Between day assay performance was verified by evaluating the signal-to-noise ratio (SNR) of the calibration curve, consistency of the LOD, and variation in the control sample measurements. CRP concentrations were measured on the Quanterix HD-X platform Analyzer using 0.02 µg / mL biotinylated detector antibody and 150 pM Streptavidin- β-galactosidase. Total Protein Total protein concentration was quantified using a NanoDrop OneC spectrophotometer (ThermoFisher Scientific) by absorbance at 280 nm with baseline correction at 340 nm, assuming 1 absorbance unit equals 1 mg / mL. A Millipore water blank was run every 10 samples for quality control. Data Cleaning and feature engineering All data processing was performed in Python 3.10 using the following libraries: numpy (1.26), scipy (1.11), pandas (2.1), matplotlib (3.7), sklearn (1.4), pycaret (3.3), seaborn(0.13), shap (0.44), optuna (4.0), xgboost (3.0), lightgbm (4.6), tabpfn (2.0), and catboost (1.2). Additional features were calculated, including binned APGAR scores, the weight change between timepoints, birthweight categories (<1 kg, 1-2 kg, and >2 kg), the change in Post-conception age (PCA) between timepoints, and Early-onset Sepsis vs Late-onset Sepsis labels (EOS < 72 hrs; LOS > 72 hrs after birth). Research-derived sample measurements were adjusted to account for batch-to- batch analytical variability. Values were first replaced at the mean LOD or mode Upper Limit of Quantification (ULOQ). Of 34,440 biomarker measurements, 4,757 (15.1%) were replaced and 6,106 (17.7%) were subsequently clipped to the assay- specific dynamic ranges. Values replaced in the first step were excluded from the clipping count to avoid double-counting. Secondary features were created by Attorney Docket No.29618-0502WO1 / BWH 2024-540 calculating the difference in biomarker concentrations between timepoints, as well as normalizing concentrations to the total protein concentration. Example 2.1. Model Development (training and validation) There were 1,446 infants consented for this study, 57% male, 42% female and < 1% indeterminate sex from Tufts Medical Center (Boston, MA), Women & Infants Hospital of Rhode Island (Providence, RI), n=178, 15%, and University of Florida, Gainesville (Gainesville, FL). Of these, 486 infants underwent more than one rule out sepsis evaluation. The majority of infants underwent an early onset sepsis evaluation (61%) defined as occurring at < 72 hours after birth. Neonates were deemed non- infected (69.8%), clinically infected (13.9%) or infected + / - septic with end organ failure and hypotension (16.3%). The majority of infections were bacterial (93%), followed by viral and fungal infections. Fifty-three percent of infants were born via cesarean section. Additional pertinent demographic and clinical information for all consented infants can be found in Table 7a. 1,215 rule out sepsis evaluations were included in the analysis providing of 2,319 salivary samples. Of these, 1,215 were obtained at the initiation of antibiotic therapy and 1,113 were obtained 18-36 hours into treatment. There were 88.1% non- infected infants and 11.9% infected + / - septic infants included in the analysis. Pertinent clinical data is provided in Table 7b. Table 7a - Demographic and clinical information for all consented infants With Number by Site Used in Number by Site Biomarker Modeling H Attorney Docket No.29618-0502WO1 / BWH 2024-540 >37 weeks 357 24. 46 261 50 306 25. 37 223 46 7 2 Post-Menstrual Age at Sample Collection: Attorney Docket No.29618-0502WO1 / BWH 2024-540 Infected 144 10. 28 99 17 144 11. 28 99 17 2 9 Clinically 197 13 22 164 11 0 Table 7b - clinical data Enrolled Number by Site Used in Number by Site Modeling H Attorney Docket No.29618-0502WO1 / BWH 2024-540 28 to 33 658 35. 205 314 139 434 35. 151 124 28 6 / 7 weeks 2 7 32 to 36 283 15 52 170 61 190 15 38 124 28 Attorney Docket No.29618-0502WO1 / BWH 2024-540 <48 hours 809 43. 234 436 139 644 53. 204 355 85 2 0 >48 hours 331 17 43 147 141 208 17 23 128 57 or necrosis factor-alpha (TNF-α), interferon gamma (IFN-γ) and interleukin 1 beta (IL- 1β), IL-4, IL-5, IL-6, IL-8, IL-10, IL-12p70 and IL-22. Total protein was used as a feature to normalize samples for varying starting volumes. The results are shown in FIG.9. As can be seen in FIG.9, median levels of each the biomarkers (except total protein, which was used to normalize) were statistically significantly higher in infected subjects than in uninfected. A number of candidate models were set us as follows. First, the data was pre- processed. Multiple strategies for splitting the data were tested, including random draws in cross-validation and splitting by site. Sample measurements were either (1) split by site, by creating roughly equal sized groupings of Tufts and WIHRI samples against UF-Gainesville, with z-score standardization applied using the mean and standard deviation of the training data (to prevent data leakage), or (2) pooled and z- score standardized before a random 70%:30% split into training and hold-out test data. In the pooled approach, the training data were evaluated using 10-fold cross- validation. Demographic features included post-menstrual age (PMA), weight at sample collection, clinical chorioamnionitis, sex, complete blood count (CBC) white blood cell counts (WBC), corrected gestational age CGA), and maternal age (included as an uninformative control feature). Models were trained on four different feature sets, Attorney Docket No.29618-0502WO1 / BWH 2024-540 including: (1) data obtained from the first timepoint only (Tp1), (2) data from the second timepoint (Tp2), (3) the change between timepoints (Tp2-Tp1), or (4) the combined data of either Tp1 or Tp2, to maximize the size of the training data. Initial model selection was performed using the AutoML pipeline PyCaret to compare multiple machine learning algorithms, including: tree-based methods (Random Forest, Gradient Boosting Classifier, LightGBM, XGBoost Decision tree, Extra Trees), Linear discriminant Analysis (LDA), Logistic Regression (LR), Support Vector Machine (SVM), Ridge classifier, Naïve Bayes, K-Nearest Neighbors, and Quadratic Discriminant analysis (QDA). Data were processed using an iterative imputer repeated five times, and a 70% / 30% train-test split for initial model screening. Models were ranked by F1 score, and the top-performing models evaluated in 10-fold cross validation on the training set, with final performance assessed on the independent hold-out test data for evaluation. The best performing models of each feature set were then manually implemented starting from default parameters. These models included Linear Discriminant Analysis (LDA) and Logistic Regression (LR), LightGBM, CatBoost, and TabPFN. LDA and LR models were optimized using a comprehensive grid search across all solver types and shrinkage parameters for LDA, and all regularization types, regularization strengths, solvers, and maximum number of iterations for LR. Class imbalance was addressed using class weights (priors) and SMOTE. LightGBM (LGBM) models were optimized sequentially by tuning boosting type, number of leaves and tree depth, Regularization parameters (L1 and L2 penalties), and learning rate. LGBM natively handles missing data and class imbalance, so no additional imputation or oversampling was performed. CatBoost models were trained using default parameters, except for a reduced learning rate (0.01) and early stopping after 20 rounds without improvement. Models were initialized with 500 iterations using LogLoss as the objective function, and AUC or Logloss as the evaluation metric. Class probabilities were predicted, and an optimal binarization threshold was selected based on the ROC curve to generate final predictions. TabPFN models were trained using the default parameters of the AutoTabPFN Classifier without additional tuning. Model performance was evaluated using ROC AUC, F1 score, precision, recall, and accuracy. Calibration plots with bin-wise sample counts, learning curves, and confusion matrices were generated for all models, with attention to minimizing Attorney Docket No.29618-0502WO1 / BWH 2024-540 misclassification of uninfected infants by adjusting class weights and penalties, where applicable. Class imbalance was addressed using Synthetic Minority Oversampling Technique (SMOTE) augmentation, adjusted class priors, and native model support. Hyperparameter tuning in tree-based models was performed both manually using a grid-search and automatically using Optuna In general, hyperparameter tuning resulted in moderate improvements over default settings. Feature-level importance was evaluated using SHAP plots (where supported), model coefficients, Permutation Feature (PFI), and Recursive Feature Elimination (RFE). Final models were selected based on a combination of performance metrics and feature minimization. As noted above, the underlying data used in the models included the following: Levels (concentrations) of biomarker proteins: some or all of IL-1b, IL-4, IL-5, IL-6, IL-8, IL-10, IL-12p70, IL-22, IFNg, TNFa, CRP, and total protein. Medical record features included: some or all of weight at collection, PCA, age of the mother, sex, birth weight, clinical chorioamnionitis, and WBC counts. Among all datasets, “Timepoint 1” and “Either timepoint” were able to generate informative classifiers (defined as models that could reliably distinguish between infected and uninfected infants). Most models achieved an AUC-ROC of 0.8+ / -0.02 on the test sets, and correct classification of both infected and infected infants ~75% of the time. (These parameters can be adjusted by altering the threshold for classification of the infant as infected / uninfected.) The models could be restricted to 7 or 8 features with minimal degradation in performance (AUC-ROC 0.78 vs 0.80). Feature selection by SHAP, Recursive Feature Elimination, Permutation Feature Importance, and feature importance indicated that: ^ Birthweight is regularly the most important feature. ^ Post Conception Age is also a strong feature. ^ IL-1b, TNFa, and IL-6 were generally important. ^ There was disagreement between feature identification methods for the importance of White Blood Cell counts and CRP (C-Reactive Protein). ^ IL-5 was useful in one model, IL-8 in another. ^ IL-12p70, sex, the presence of clinical chorioamnionitis, IL-22, IL-4, IFNg, and IL-10 were uninformative in most models. Attorney Docket No.29618-0502WO1 / BWH 2024-540 Table 8 provides a summary of the performance of the top models; FIGs.10A-H provide the ROC curves for each of the models. Table 8 - Performance of example classification models Either Either Data set Tp 1 Tp 1 Tp 1 Tp 1 Tp2-Tp1 Tp Either Tp Tp 02 86 55 82 77 - LDA, Linear Discriminant Analysis References 1. Attia Hussein Mahmoud, H. et al. Insight into neonatal sepsis: an overview. Cureus 15, e45530 (2023). 2. Cantey, J. B. & Prusakov, P. A proposed framework for the clinical management of neonatal ‘culture-negative’ sepsis. J. Pediatr.244, 203–211 (2022). 3. Klingenberg, C., Kornelisse, R. F., Buonocore, G., Maier, R. F. & Stocker, M. Culture-negative early-onset neonatal sepsis—at the crossroad between efficient sepsis care and antimicrobial stewardship. Front. Pediatr.6, 285 (2018). 4. Cantey, J. B. & Baird, S. D. Ending the culture of culture-negative sepsis in the neonatal ICU. Pediatrics 140, e20170044 (2017). Attorney Docket No.29618-0502WO1 / BWH 2024-540 5. Bouadma, L. et al. Use of procalcitonin to reduce patients’ exposure to antibiotics in intensive care units (PRORATA trial): a multicentre randomised controlled trial. Lancet 375, 463–474 (2010). 6. Cantey, J. B. & Lee, J. H. Biomarkers for the diagnosis of neonatal sepsis. Clin. Perinatol.48, 215–227 (2021). 7. Memar, M. Y., Alizadeh, N., Varshochi, M. & Kafil, H. S. Immunologic biomarkers for diagnostic of early-onset neonatal sepsis. J. Matern. Fetal Neonatal Med.32, 143–153 (2019). 8. Ludwig, K. R. & Hummon, A. B. Mass spectrometry for the discovery of biomarkers of sepsis. Mol. Biosyst.13, 648–664 (2017). 9. Eichberger, J., Resch, E. & Resch, B. Diagnosis of neonatal sepsis: the role of inflammatory markers. Front. Pediatr.10, 840288 (2022). 10. Mussap, M., Noto, A., Fravega, M. & Fanos, V. Soluble CD14 subtype presepsin (sCD14-ST) and lipopolysaccharide binding protein (LBP) in neonatal sepsis: new clinical and analytical perspectives for two old biomarkers. J. Matern. Fetal Neonatal Med.24, 12–14 (2011). 11. Pavcnik-Arnol,M.,Hojker, S. &Derganc,M. Lipopolysaccharide-binding protein, lipopolysaccharide, and soluble CD14 in sepsis of critically ill neonates and children. Intensive Care Med.33, 1025–1032 (2007). 12. Arnon, S. et al. Serum amyloid A: an early and accurate marker of neonatal early-onset sepsis. J. Perinatol.27, 297–302 (2007). 13. Cetinkaya, M., Ozkan, H., Köksal, N., Celebi, S.&Hacimustafaoğlu, M. Comparison of serum amyloid A concentrations with those of C-reactive protein and procalcitonin in diagnosis and follow-up of neonatal sepsis in premature infants. J. Perinatol.29, 225–231 (2009). 14. Badr, H. S., El-Gendy, F. M. & Helwa, M. A. Serum stromal-derived factor-1 (CXCL12) and its alpha chemokine receptor (CXCR4) as biomarkers in neonatal sepsis. J. Matern. Fetal Neonatal Med.31, 2209–2215 (2018). 15. Chen, S. et al. Expression of serum cytokines profile in neonatal sepsis. Infect. Drug Resist.15, 3437 (2022). 16. Aliefendioglu, D., Gürsoy, T., Çağlayan, O., Aktaş, A. & Ovalı, F. Can resistin be a new indicator of neonatal sepsis? Pediatr. Neonatol.55, 53–57 (2014). Attorney Docket No.29618-0502WO1 / BWH 2024-540 17. Khattab, A. A., El-Mekkawy, M. S., Helwa, M. A. & Omar, E. S. Utility of serum resistin in the diagnosis of neonatal sepsis and prediction of disease severity in term and late preterm infants. J. Perinat. Med.46, 919–925 (2018). 18. Ng, P. Diagnostic markers of infection in neonates. Arch. Dis. Child. Fetal Neonatal Ed.89, F229–F235 (2004). 19. Grunau, R. E. et al. Neonatal pain, parenting stress and interaction, in relation to cognitive and motor development at 8 and 18 months in preterm infants. Pain 143, 138–146 (2009). 20. Grunau, R. Early pain in preterm infants. A model of long-term effects. Clin. Perinatol.29, 373–394 (2002). 21. Grunau, R. E., Holsti, L. & Peters, J. W. B. Long-term consequences of pain in human neonates. Semin. FetalNeonatalMed.11, 268–275 (2006). 22. Iyengar, A. & Maron, J. L. Detecting infection in neonates: promises and challenges of a salivary approach. Clin. Ther.37, 523–528 (2015). 23. Iyengar, A., Paulus, J. K., Gerlanc, D. J. & Maron, J. L. Detection and potential utility of C-reactive protein in saliva of neonates. Front. Pediatr.2, 131 (2014). 24. Romano-Keeler, J., Wynn, J. L. & Maron, J. L. Great expectorations: the potential of salivary “omic” approaches in neonatal intensive care. J. Perinatol.34, 169–173 (2014). 25. Chiappin, S., Antonelli, G., Gatti, R. &De Palo, E. F. Saliva specimen: a new laboratory tool for diagnostic and basic investigation. Clin. Chim. Acta 383, 30– 40 (2007). 26. Lin, G. C. et al. Directed transport of CRP across in vitro models of the blood-saliva barrier strengthens the feasibility of salivary CRP as biomarker for neonatal sepsis. Pharmaceutics 13, 256 (2021). 27. Chen, I.-L. et al. A novel method to detect bacterial infection in premature infants: using a combination of inflammatory markers in blood and saliva. J. Microbiol. Immunol. Infect.53, 892–899 (2020). 28. da Rocha, V. A. et al. Identification of inflammatory mediators in saliva samples from hospitalized newborns: potential biomarkers? Clin. Nurs. Res.33, 207– 219 (2024). Attorney Docket No.29618-0502WO1 / BWH 2024-540 29. Barekatain, B., HasanGhalyaei, N., Mohammadizadeh, M. & Tavakolifard, N. Investigation of salivary C-reactive protein and interleukin-18 for the diagnosis of neonatal sepsis. J. Res. Med. Sci.26, 131 (2021). 30. Omran, A. et al. Salivary and serum interleukin-10, C-reactive protein, mean platelet volume, and CRP / MPV ratio in the diagnosis of late onset neonatal sepsis in full-term neonates. J. Immunol. Res.2021, 4884537 (2021). 31. Wang, X., Fei, Q. & Yuan, T. The diagnostic value of salivary C-reactive protein in neonatal infections: a meta-analysis. Infection 52, 2071–2082 (2024). 32. Datla, S., Kitchanan, S. & Sethuraman, G. Diagnostic reliability of salivary C-reactive protein as an alternative noninvasive biomarker of neonatal sepsis. Indian Pediatr.58, 745–748 (2021). 33. Omran, A., Maaroof, A., Mohammad, M. H. S. & Abdelwahab, A. Salivary C-reactive protein, mean platelet volume and neutrophil lymphocyte ratio as diagnostic markers for neonatal sepsis. J. Pediatr.94, 82–87 (2018). 34. Ramavath, C., Katam, S. K., Vardhelli, V., Deshabhotla, S. & Oleti, T. P. Examining the utility of rapid salivary C-reactive protein as a predictor for neonatal sepsis: an analytical cross-sectional pilot study. Diagn. Basel Switz.13, 867 (2023). 35. Tosson, A. M. S., Koptan, D., Abdel Aal, R. & Abd Elhady, M. Evaluation of serum and salivary C-reactive protein for diagnosis of late-onset neonatal sepsis: a single center cross-sectional study. J. Pediatr.97, 623–628 (2021). 36. Cui, Y. et al. Developments in diagnostic applications of saliva in human organ diseases. Med. Nov. Technol. Devices 13, 100115 (2022). 37. Rissin, D. M. et al. Single-molecule enzyme-linked immunosorbent assay detects serum proteins at subfemtomolar concentrations. Nat. Biotechnol.28, 595– 599 (2010). 38. Zhang, J. et al. Improving the accuracy, robustness, and dynamic range of digital bead assays. Anal. Chem.95, 8613–8620 (2023). 39. Delanghe, J. R. & Speeckaert, M. M. Translational research and biomarkers in neonatal sepsis. Clin. Chim. Acta 451, 46–64 (2015). 40. Linge, H. M. et al. The human CXC chemokine granulocyte chemotactic protein 2 (GCP-2) / CXCL6 possesses membrane-disrupting properties and is antibacterial. Antimicrob. Agents Chemother.52, 2599–2607 (2008). Attorney Docket No.29618-0502WO1 / BWH 2024-540 41. Tunc, T. et al. Diagnostic value of elevated CXCR4 and CXCL12 in neonatal sepsis. J. Matern. Fetal Neonatal Med.28, 356–361 (2015). 42. U.S. Food and Drug Administration. Bioanalytical Method Validation Guidance for Industry. fda.gov / regulatory-information / search-fda-guidance- documents / bioanalytical-method-validationguidance- industry (2018). 43. Urvasizoglu, G., Kilic, A., Capik, O., Gundogdu, M. & Karatas, O. F. CXCL14 and miR-4484 serves as potential salivary biomarkers for early detection of peri-implantitis. Odontology 112, 864–871 (2023). 44. Hernández-Molina, G., Michel-Peregrina, M., Hernández-Ramírez, D. F., Sánchez-Guerrero, J. & Llorente, L. Chemokine saliva levels in patients with primary Sjögren’s syndrome, associated Sjögren’s syndrome, pre-clinical Sjögren’s syndrome and systemic autoimmune diseases. Rheumatology 50, 1288–1292 (2011). 45. Prasad, G. & McCullough, M. Chemokines and cytokines as salivary biomarkers for the early diagnosis of oral cancer. Int. J. Dent.2013, 813756 (2013). 46. McQuibban, G. A. et al. Matrix metalloproteinase activity inactivates the CXC chemokine stromal cell-derived factor-1. J. Biol. Chem.276, 43503–43508 (2001). 47. Nesin, M. & Cunningham-Rundles, S. Cytokines and neonates. Am. J. Perinatol.17, 393–404 (2000). 48. Anderson-Bergman, C. icenReg: regression models for interval censored data in R. J. Stat. Softw.81, 1–23 (2017). 49. Bromiker R, Elron E, Klinger G. Do Neonatal Infections Require a Positive Blood Culture? Am J Perinatol.2020 Sep;37(S 02):S18-S21. 50. Mukhopadhyay S, Briker SM, Flannery DD, et al. Time to positivity of blood cultures in neonatal late-onset bacteraemia. Arch Dis Child Fetal Neonatal Ed. 2022 Nov;107(6):583-588. 51. Kuzniewicz MW, Puopolo KM, Fischer A, et al. A Quantitative, Risk- Based Approach to the Management of Neonatal Early-Onset Sepsis. JAMA Pediatr. 2017 Apr 1;171(4):365-371. 52. Kuzniewicz MW, Escobar GJ, Forquer H, et al. Update to the Neonatal Early-Onset Sepsis Calculator Utilizing a Contemporary Cohort. Pediatrics.2024 Oct 1;154(4):e2023065267. Attorney Docket No.29618-0502WO1 / BWH 2024-540 53. Straub J, Paula H, Mayr M, et al. Diagnostic accuracy of the ROCHE Septifast PCR system for the rapid detection of blood pathogens in neonatal sepsis— A prospective clinical trial. PLOS ONE; 2017;12(11): e0187688. 54. An X, Zhang X, ShangGuan Y. Application of PCT, IL-6, CRP, and WBC for Diagnosing Neonatal Sepsis. Clin Lab.2023 Aug 1;69(8). 55. Kurt AN, Aygun AD, Godekmerdan A, et al. Serum IL-1beta, IL-6, IL-8, and TNF-alpha levels in early diagnosis and management of neonatal sepsis. Mediators Inflamm.2007;2007:31397. OTHER EMBODIMENTS It is to be understood that while the invention has been described in conjunction with the detailed description thereof, the foregoing description is intended to illustrate and not limit the scope of the invention, which is defined by the scope of the appended claims. Other aspects, advantages, and modifications are within the scope of the following claims.

Claims

Attorney Docket No.29618-0502WO1 / BWH 2024-540 WHAT IS CLAIMED IS:

1. A method comprising: obtaining a sample comprising saliva from a mammalian neonatal subject, and determining a level of expression of biomarker proteins in the sample, preferably using an ultrasensitive protein assay, wherein the biomarker proteins comprise one, two, three, four, five, seven or more of Interleukin 1 beta (IL-1b), IL-4, IL-5, IL-6, IL-8, IL-10, IL-12p70, IL-22, interferon gamma (IFNg), tumor necrosis factor alpha (TNFa), C-reactive protein (CRP), and total protein levels.

2. The method of claim 1, wherein the biomarker proteins comprise one, two, or all three of Interleukin 1 beta (IL-1b), tumor necrosis factor alpha (TNFa), and IL-6, and optionally further comprise C reactive protein (CRP), IL-5 and / or IL-8, optionally wherein the biomarkers comprise: (i) IL-4, I-L1b, IL-6, IL-10, IL-22, and IL-12p70; (ii) IL1b, TNFa, IL6, and IL8; (iii) CRP, TNFa, IL-10, IL-6, IL-12p70, and IL-22; (iv) IL-1b; or (v) IL-1b, IL-10, TNFa, CRP, IL-4, and IL-6.

3. The method of claim 1 or 2, further comprising comparing the level of expression of the biomarker proteins to a disease reference, wherein a level of expression of the biomarker proteins above the reference levels indicates that the subject has or is at risk of developing sepsis or an infection.

4. The method of any of claims 1 to 3, wherein the subject is a human neonate of up to about 28 days after birth or up to about 52 weeks postmenstrual age (PMA).

5. The method of claims 1 to 4, wherein the ultrasensitive protein assay is Single- Molecule Arrays (SIMOA); SIMOA Planar Arrays; Molecular On-bead Signal Amplification for Individual Counting (MOSAIC); Meso Scale Discovery (MSD); Single-Molecule Counting (SMC); nucleic acid linked immune-sandwich assay (NULISA); LUMINEX; SOMAscan Assays; mass spectrometry (optionally MALDI-MS), and / or mass cytometry (optionally CyTOF).Attorney Docket No.29618-0502WO1 / BWH 2024-540 6. The method of claims 1 to 4, further comprising recommending or sending the subject for additional evaluation, optionally by culture.

7. The method of claims 1 to 4, further comprising administering a treatment for sepsis or an infection to a subject who has been identified as having or at risk of developing sepsis.

8. The method of claim 7, wherein the treatment comprises administration of an antibiotic.

9. The method of claim 1, further comprising determining a level of biomarker proteins in the subject after administration of the treatment, and comparing the level of biomarker proteins prior to treatment with the level of biomarker proteins during and / or after treatment, wherein a decrease in the level of biomarker proteins indicates that the treatment is effective in treating the sepsis.

10. A method of diagnosing a subject as having or likely to develop sepsis or an infection, the method comprising: obtaining a sample comprising saliva from a mammalian neonatal subject, determining a level of biomarker proteins in the sample with an ultrasensitive protein assay, wherein the biomarker proteins comprise one, two, three, four, five, seven or more of Interleukin 1 beta (IL-1b), IL-4, IL-5, IL-6, IL-8, IL-10, IL- 12p70, IL-22, interferon gamma (IFNg), tumor necrosis factor alpha (TNFa), C- reactive protein (CRP), and total protein levels, and comparing the level of biomarker proteins to a disease reference, wherein differential expression of the biomarker proteins between the sample and the reference indicates that the subject has or is at risk of developing sepsis or an infection.

11. The method of claim 10, wherein the biomarker proteins comprise one, two, or all three of Interleukin 1 beta (IL-1b), tumor necrosis factor alpha (TNFa), and IL-6, and optionally further comprise C reactive protein (CRP), IL-5 and / or IL-8.

12. The method of claim 10, wherein the biomarker proteins comprise: (i) IL-4, I-L1b, IL-6, IL-10, IL-22, and IL-12p70; (ii) IL1b, TNFa, IL6, and IL8;Attorney Docket No.29618-0502WO1 / BWH 2024-540 (iii) CRP, TNFa, IL-10, IL-6, IL-12p70, and IL-22; or (iv) IL-1b; (v) IL-1b, IL-10, TNFa, CRP, IL-4, and IL-6.

13. The method of any of claims 10 to 12, wherein the subject is a human neonate of up to about 28 days after birth or up to about 290 days gestational age.

14. The method of any of claims 10 to 12, wherein the ultrasensitive assay is Single- Molecule Arrays (SIMOA); SIMOA SP-X Planar Arrays, Molecular On-bead Signal Amplification for Individual Counting (MOSAIC); Meso Scale Discovery (MSD); Single-Molecule Counting (SMC); nucleic acid linked immune-sandwich assay (NULISA); LUMINEX; SOMAscan Assays; mass spectrometry (optionally MALDI-MS), and / or mass cytometry (optionally CyTOF).

15. A method of diagnosing a subject as having or likely to develop sepsis or an infection, the method comprising: providing a level of biomarker proteins in a sample comprising saliva from a mammalian neonatal subject, wherein the biomarker proteins comprise one, two, three, four, five, seven or more of Interleukin 1 beta (IL-1b), IL-4, IL-5, IL-6, IL- 8, IL-10, IL-12p70, IL-22, interferon gamma (IFNg), tumor necrosis factor alpha (TNFa), C-reactive protein (CRP), and total protein levels, and comparing the level of biomarker proteins to a disease reference, to provide a value representing the differential expression of each biomarker protein; providing a value representing one or more medical history feature selected from weight at collection, postmenstrual age (PMA) or post-conception age (PCA), age of the mother, sex, birth weight, clinical chorioamnionitis, complete blood count (CBC), and white blood cell (WBC) counts in the subject; calculating a score using the values for the biomarkers and medical history features, wherein the score indicates that the subject has or is at risk of developing sepsis or an infection.

16. The method of claim 15, wherein the biomarker proteins comprise one, two, or all three of Interleukin 1 beta (IL-1b), tumor necrosis factor alpha (TNFa), and IL-6, and optionally further comprise C reactive protein (CRP), IL-5 and / or IL-8, and the medical history features comprise birth weight and PMA or PCA.Attorney Docket No.29618-0502WO1 / BWH 2024-540 17. The method of claim 15, wherein the biomarker proteins and medical history features comprise: (i) birth weight, sex, PMA or PCA, IL-4, I-L1b, IL-6, IL-10, IL-22, and IL-12p70; (ii) birth weight, PMA or PCA, IL1b, TNFa, IL6, and IL8; (iii) birth weight, CRP, TNFa, IL-10, IL-6, IL-12p70, and IL-22; (iv) IL-1b; or (v) birth weight, WBC, CBC, PMA or PCA, IL-1b, IL-10, TNFa, CRP, IL-4, and IL-6.

18. The method of any of claims 15 to 17, wherein the subject is a human neonate of up to about 28 days after birth or up to about 290 days gestational age.

19. The method of any of claims 15 to 17, wherein the score is calculated using a machine learning algorithm.

20. The method of claim 19, wherein the machine learning algorithm is selected from a Logistic Regression, a Linear Discriminant Analysis, a LightGradient Boosted classifier; a CatBoost classifier; and TabPFN.

21. The method of claim 19, wherein the biomarker proteins, medical history features, and algorithm comprise: (i) birth weight, sex, PMA or PCA, IL-4, I-L1b, IL-6, IL-10, IL-22, and IL-12p70, and L2 penalized Logistic Regression; (ii) birth weight, PMA or PCA, IL1b, TNFa, IL6, and IL8, and LightGBM; (iii) birth weight, CRP, TNFa, IL-10, IL-6, IL-12p70, and IL-22 and Logistic Regression; (iv) IL-1b and Linear Discriminant Analysis (LDA); or (iv) birth weight, WBC, CBC, PMA or PCA, IL-1b, IL-10, TNFa, CRP, IL-4, and IL-6, and LightGBM.

22. The method of any of claims 9 to 21, further comprising recommending or sending the subject for additional evaluation, optionally by culture.

23. The method of claims 9 to 21, further comprising recommending, administering or continuing administration of a treatment for sepsis or an infection to a subject who has been identified as having or at risk of developing sepsis or an infection, orAttorney Docket No.29618-0502WO1 / BWH 2024-540 stopping administration of a treatment for sepsis to a subject who has been identified as not having or at risk of developing sepsis or an infection.

24. The method of claim 23, wherein the treatment comprises administration of an antibiotic.

25. The method of claim 24, further comprising determining a level of biomarker proteins in the subject after administration of the treatment, and comparing the level of biomarker proteins prior to treatment with the level of biomarker proteins during and / or after treatment, wherein a decrease in the level of biomarker proteins indicates that the treatment is effective in treating the sepsis or an infection.