Volatile biomarkers for discriminating valley fever from other lung infections

Abstract

Disclosed herein are methods and devices for minimally invasive or non-invasive detection of pneumonia in a subject, and discrimination of a fungal pneumonia from a bacterial pneumonia, using discriminatory volatile organic compounds (VOCs). Also provided are methods of treatment in a subject having or suspected of having pneumonia.

Description

Atty. Docket No.055743-824202 Client’s Ref.: M23-284L-WO1-a VOLATILE BIOMARKERS FOR DISCRIMINATING VALLEY FEVER FROM OTHER LUNG INFECTIONS STATEMENT REGARDING FEDERALLY SPONSORED RESEARCH OR DEVELOPMENT

[0001] This invention was made with government support under ADHS18-198861 awarded bythe Arizona Biomedical Research Centre. The government has certain rights in the invention. CROSS REFERENCE TO RELATED APPLICATIONS This is a PCT patent application that claims benefit to U.S. Provisional Application Serial No. 63 / 604,469 filed on November 30, 2023, which is herein incorporated by reference in its entirety. FIELD

[0002] The present disclosure relates to methods, systems and devices for determining thepresence and type of pneumonia in a subject, and related treatment methods. BACKGROUND

[0003] Pneumonia is an acute pulmonary infection and classified by the “pneumonia triad” aseither community-acquired pneumonia (CAP), which is the most common, hospital-acquired, which includes ventilator-associated pneumonia, or pneumonia in the immunocompromised host. Pneumonia is a major health concern and is a leading cause of infectious death worldwide with more than 2.3 million deaths and more than 91 million years of life lost in 2016. Even excluding the effects of the COVID-19 pandemic, in the United States, pneumonia is the most common cause of hospital admission outside of birth; approximately 1 million adults seek care in a hospitaldue to pneumonia every year, and 50,000 die from this disease. Pneumonia is a huge burden onthe US healthcare system as it is one of the top ten most expensive conditions during inpatient hospitalization. Current strategies used for diagnosing pneumonia and determining etiology are based on clinical, radiological, and microbial criteria; however, these criteria are time-consuming, lack accuracy, suffer from large interobserver variability, and can be invasive as accessing a lower lung specimen is difficult.

[0004] Coccidioidomycosis or Valley fever is a fungal pneumonia endemic to the arid and semi-arid regions of North and South America. It is estimated there are 350,000 new cases each year and in endemic and highly populated regions up to 30% of community acquired pneumonia may 1 100243977.1Atty. Docket No.055743-824202 Client’s Ref.: M23-284L-WO1-a be caused by Valley fever. It is currently difficult to diagnose, in part because symptomatic primary pulmonary infection often resembles bacterial pneumonia leading to a misdiagnosis and inappropriate treatment with antibiotics.

[0005] Therefore, there is an unmet need for developing methods for distinguishing anddiagnosing Valley fever from pneumonia caused by other lung infections, such as e.g., bacterial pneumonia. SUMMARY

[0006] In some aspects, the disclosure encompasses a minimally invasive or non-invasivecomputer-implemented method for identifying a pulmonary infection in a subject, the method comprising of: (a) a test sample from the lungs of the subject, determining or having determined the relative abundance of each of a plurality of volatile organic compounds (VOCs) to prepare a sample VOC dataset; and (b) generating, by a processor, a machine learning prediction of a diagnosis selected from a pneumonia, a bacterial pneumonia, and a fungal pneumonia infection afflicting the subject using the sample VOC dataset as input to a machine learning model in view of a plurality of VOC abundances predetermined to be predictive of the pneumonia, bacterial pneumonia or pneumonia.

[0007] In some aspects, the machine learning model includes Random Forest (RF) classificationand is trained, by the processor, by steps including: (a) accessing at least one sample VOC dataset and at least one VOC training dataset associated with the pneumonia, bacterial pneumonia or fungal pneumonia, wherein the VOC training dataset comprises the plurality of VOCs predetermined to be predictive of the pneumonia, bacterial pneumonia or fungal pneumonia; and (b) analyzing the relative abundance of VOCs of each sample VOC dataset; and (c) ranking an ability of each VOC to classify a sample for a diagnosis of pneumonia, bacterial pneumonia or fungal pneumonia.

[0008] In some aspects, the training further comprises: (d) building a first plurality of classifiertrees; (e) selecting a first set of top VOCs from the first plurality of classifier trees; (f) building a second plurality of classifier trees using the set of top VOCs from (e); (g) selecting and saving a second set of top VOCs from the second plurality of classifier trees, the second set of top classifiers defining selected VOCs; and (h) performing multiple times to select the best set of top VOCs.

[0009] In some aspects, the Random Forest classification includes building an ensemble of aplurality of classifier trees, where the final prediction for a test sample is obtained by majority vote on the combination of predictions of all of the plurality of classifier trees. 2 100243977.1Atty. Docket No.055743-824202 Client’s Ref.: M23-284L-WO1-a

[0010] In some aspects, the fungal pneumonia is coccidioidomycosis, caused by the fungusCoccidioides. In some aspects, the method further comprises generating, by the processor, a plurality of VOC predictors for pneumonia, bacterial pneumonia and fungal pneumonia. In some aspects, the plurality of VOC predictors comprises [161, 215, 368, 377, and 515]. In some aspects, the plurality of VOC predictors comprises [161, 199, 215, 254, 368, 377, and 515]. In some aspects, the plurality of VOC predictors comprises [46, 150, 255, and 369]. In some aspects, the plurality of VOC predictors comprises [150, 182, 255, 352, and 374].

[0011] In some aspects, the VOC predictors correctly predict pneumonia infection in the subjectwith an error rate of no more than 12%. In some aspects, the VOC predictors correctly predict fungal pneumonia infection in the subject with an error rate of no more than 7%.

[0012] Further provided herein is a method for distinguishing a fungal pneumonia from a bacterialpneumonia, the method comprising: (a) from a test sample from the lungs of the subject, determining or having determined the relative abundance of each of a plurality of volatile organic compounds (VOCs) to prepare a sample VOC dataset; and (b) generating, by a processor, a machine learning prediction of fungal pneumonia infection afflicting the subject using the sample VOC dataset as input to a machine learning model in view of a plurality of VOC abundances predetermined to be predictive of the fungal pneumonia.

[0013] In some aspects, the machine learning model includes Random Forest (RF) classificationand is trained, by the processor, by steps including: (a) accessing at least one sample VOC dataset and at least one VOC training dataset associated with the fungal pneumonia, wherein the VOC training dataset comprises the plurality of VOC abundances predetermined to be predictive of the fungal pneumonia; and (b) analyzing the relative abundance of VOCs of each sample VOC dataset; and (c) ranking an ability of each VOC to classify a sample for a diagnosis of fungal pneumonia.

[0014] In some aspects, the training further comprises: (d) building a first plurality of classifiertrees; (e) selecting a first set of top VOCs from the first plurality of classifier trees; (f) building a second plurality of classifier trees using the set of top VOCs from (e); (g) selecting and saving a second set of top VOCs from the second plurality of classifier trees, the second set of top classifiers defining selected VOCs, and (h) performing multiple times to select the best set of top VOCs. In some aspects, the Random Forest classification includes building an ensemble of a plurality of classifier trees, where the final prediction for a test sample is obtained by majority vote on the combination of predictions of all of the plurality of classifier trees.

[0015] In some aspects, the fungal pneumonia is coccidioidomycosis, caused by the fungusCoccidioides. In some aspects, the method further comprises generating, by the processor, a 3 100243977.1Atty. Docket No.055743-824202 Client’s Ref.: M23-284L-WO1-a plurality of VOC predictors for fungal pneumonia. In some aspects, the plurality of VOC predictors comprises [150, 182, 255, 352, and 374], wherein the VOC predictors correctly predict fungal pneumonia infection in the subject with an error rate of no more than 7%. In some aspects, the plurality of VOC predictors further comprises [46, 150, 255, and 369], wherein the VOC predictors correctly predict fungal pneumonia infection in the subject with an error rate of no more than 5%.

[0016] In some aspects, the sample is a bronchoalveolar lavage sample, a sputum sample, or abreath sample. In some aspects, the relative abundance of each of plurality of VOCs is determined using a mass spectrometer. In some aspects, the mass spectrometry comprises two-dimensional gas chromatography-time-of-flight mass spectrometry (GC×GC-TOFMS).

[0017] In further aspects, provided herein is a method for treating a subject having or suspectedof having a pneumonia, the method comprising: selecting the subject’s treatment from a fungal infection treatment and an antibiotic, by distinguishing or having distinguished a suspected pneumonia as a bacterial pneumonia or a fungal pneumonia by the method disclosed herein.

[0018] The disclosure further encompasses a sensor platform for identifying a pulmonaryinfection, distinguishing a fungal pneumonia from a bacterial pneumonia, or any combination thereof, in a subject comprising: (a) a receiver configured to receive a sample from a subject in need thereof; (b) a sensor configured to detect and / or assess relative abundance of one or more VOCs of any preceding claim; (c) a processor configured to analyze, process and / or identify the one or more of the detected VOCs; (d) an output module configured to transmit a result generated by the processor.

[0019] In some aspects, the processor comprises a machine learning algorithm configured toanalyze, process, and / or identify the VOCs contained in the sample. In some aspects, the sample is a bronchoalveolar lavage sample, a sputum sample, or a breath sample. BRIEF DESCRIPTION OF THE FIGURES

[0020] The patent or application file contains at least one drawing executed in color. Copies ofthis patent or patent application publication with color drawing(s) will be provided by the Office upon request and payment of the necessary fee.

[0021] FIG. 1 is a schematic of processing steps of mass spectrometer data.

[0022] FIG. 2 is a Bee Swarm plot of class probabilities using Random Forest classification ofInfected vs. Negative samples based on VOC analysis of seven VOC metabolites labeled with an X followed by the VOC number [161, 199, 215, 254, 368, 377, 515] from human bronchoalveolar lavage fluid (BALF) samples, showing a model accuracy of 0.88. Each point represents a technical 4 100243977.1Atty. Docket No.055743-824202 Client’s Ref.: M23-284L-WO1-a replicate of a BALF sample, grouped into their true classifications of Infected (purple circles) and Negative (gray triangles), and plotted based on the probability that the sample is classified as Negative based on 100 iterations of Random Forest. Points towards the top of the plot indicate a higher probability of classifying as negative, while those towards the bottom of the plot indicate a higher probability of classifying as infected.

[0023] FIG. 3 is a principal component analysis biplot using seven discriminatory VOCs [161,199, 215, 254, 368, 377, 515] as variables, and infected or negative human BALF samples as observations. Each observation represents the geometric mean of VOC signal intensities for technical replicates for the sample. Ellipses represent 95% confidence intervals, with infected in purple and negative in gray. Volatile compounds are labeled in black with an X preceding the VOC number. Samples are labeled and color-coded according to infected category: Coccidioides (C) – blue, bacteria (B) – green, fungi (F) – pink, viral (V) – orange, multiple pathogens (M) – red; the negative (N) samples are in gray.

[0024] FIG. 4 is a Bee swarm plot of class probabilities using Random Forest classification ofCoccidioides positive vs. Non-Cocci Infected samples based on VOC analysis of five VOC [150, 182, 255, 352, 374] metabolites from human BALF samples, showing a model accuracy of 0.93; each point represents a technical replicate of a BALF sample, grouped into their true classifications of Coccidioides (blue circles) and Non-Cocci Infected (purple triangles), and plotted based on the probability that the sample is classified as Non-Cocci Infected based on 100 iterations of Random Forest. Points towards the top of the plot indicate a higher probability of classifying as Non-Cocci Infected, while those towards the bottom of the plot indicate a higher probability of classifying as Coccidioides.

[0025] FIG.5 is a Principal component analysis biplot using five discriminatory VOCs [150, 182,255, 352, 374] as variables and Coccidioides or Non-Cocci Infected human BALF samples as observations. Each observation represents the geometric mean of VOC signal intensities of technical replicates for the sample. Ellipses represent 95% confidence intervals, with Coccidioides in blue and Non-Cocci Infected in purple. Volatile compounds are labeled in black with an X preceding the VOC number. Samples are labeled and color-coded according to infected category: Coccidioides (C)– blue; Non-Cocci Infected: bacteria (B) – green, fungi (F) – pink, viral (V) – orange, multiple pathogens (M)–red.

[0026] FIG. 6 is a Bee swarm plot of class probabilities using Random Forest classification ofCoccidioides positive vs. bacteria-positive samples using four human BALF VOCs [46, 150, 255, 369] with a model accuracy of 0.95. Bee swarm plot depicting class probabilities (Coccidioides versus Bacteria); each point represents a technical replicate of a BALF sample, grouped into their 5 100243977.1Atty. Docket No.055743-824202 Client’s Ref.: M23-284L-WO1-a true classifications of Coccidioides (blue circles) and Bacteria (green triangles), and plotted based on the probability that the sample is classified as Bacteria based on 100 iterations of Random Forest. Points towards the top of the plot indicate a higher probability of classifying as Bacteria, while those towards the bottom of the plot indicate a higher probability of classifying as Coccidioides.

[0027] FIG.7 is a Principal component analysis biplot using discriminatory VOCs [46, 150, 255,369] as variables and Coccidioides or bacterial human BALF samples as observations. Each observation represents the geometric mean of VOC signal intensities of technical replicates for the sample. Volatile compounds are labeled in black with an X preceding the VOC number. Samples and ellipses representing 95% confidence intervals are labeled and color-coded according to category: Coccidioides (C) – blue, bacteria (B) – green. DETAILED DESCRIPTION

[0028] The present disclosure describes methods, system, and device for diagnosing Valley fever.The methods, system and device disclosed herein are partly based on the surprising discovery that a set of certain volatile organic compounds (VOCs) can identify a pulmonary infection and / or distinguish coccidioidal pneumonia from bacterial pneumonia. These VOCs can be obtained from a subject in a minimally invasive manner from bronchoalveolar lavage fluid (BALF) sample, or in a non-invasive fashion from a sputum sample or a breath sample.

[0029] Untargeted volatile metabolomics analysis on BALF samples from patients with suspectedpneumonia was performed, and machine learning was used to identify specific VOCs that discriminate between patients infected with Coccidioides, infected with another microbial lung pathogen, or not having any detectable infection. The methods, system, and device described herein can also be applied to detect the presence of other fungal infections of pneumonia.

[0030] The identified discriminatory VOCs can be detected from a BALF, or a breath samplewhich can distinguish between negative and positive samples of pneumonia, and also from Valley fever-positive pneumonia against non-Valley fever pneumonia. The identified discriminatory VOCs can further be used to identify or diagnose fungal pneumonia (Valley fever) from bacterial pneumonia. In addition to the minimally or non-invasive aspects, the use of VOCs as described herein provides more accurate and faster diagnosis of fungal pneumonia or Valley fever. The methods, system, and device disclosed herein help avoid clinical misdiagnosis and determine an appropriate course of treatment.

[0031] All publications mentioned herein are incorporated herein by reference to disclose anddescribe the methods and / or materials in connection with which the publications are cited. The 6 100243977.1Atty. Docket No.055743-824202 Client’s Ref.: M23-284L-WO1-a publications discussed herein are provided solely for their disclosure prior to the filing date of the present application. Nothing herein is to be construed as an admission that the present invention is not entitled to antedate such publication by virtue of prior invention.A. Terminology

[0032] Before the present compounds, compositions, articles, systems, devices, and / or methodsare disclosed and described, it is to be understood that they are not limited to specific synthetic methods unless otherwise specified, or to particular reagents unless otherwise specified, as such may, of course, vary. It is also to be understood that the terminology used herein is for the purpose of describing particular aspects only and is not intended to be limiting. Although any methods and materials similar or equivalent to those described herein can be used in the practice or testing of the present invention, example methods and materials are now described.

[0033] This disclosure describes inventive concepts with reference to specific examples.However, the intent is to cover all modifications, equivalents, and alternatives of the inventive concepts that are consistent with this disclosure.

[0034] As used in the specification and the appended claims, the singular forms “a”, “an”, and“the” include plural referents unless the context clearly dictates otherwise.

[0035] The phrase ‘consisting essentially of’ limits the scope of a claim to the recited componentsin a composition or the recited steps in a method as well as those that do not materially affect the basic and novel characteristic or characteristics of the claimed composition or claimed method. The phrase ‘consisting of’ excludes any component, step, or element that is not recited in the claim. The phrase ‘comprising’ is synonymous with ‘including’, ‘containing’, or ‘characterized by’, and is inclusive or open-ended. ‘Comprising’ does not exclude additional, unrecited components or steps.

[0036] As used herein, when referring to any numerical value, the term ‘about’ means a valuefalling within a range that is ± 10% of the stated value.

[0037] Ranges can be expressed herein as from ‘about’ one particular value, and / or to ‘about’another particular value. When such a range is expressed, a further aspect includes from the one particular value and / or to the other particular value. Similarly, when values are expressed as approximations, by use of the antecedent ‘about,’ it will be understood that the particular value forms a further aspect. It will be further understood that the endpoints of each of the ranges are significant both in relation to the other endpoint and independently of the other endpoint. It is also understood that there are a number of values disclosed herein, and that each value is also herein disclosed as ‘about’ that particular value in addition to the value itself. For example, if the value 7 100243977.1Atty. Docket No.055743-824202 Client’s Ref.: M23-284L-WO1-a ‘10’ is disclosed, then ‘about 10’ is also disclosed. It is also understood that each unit between two particular units are also disclosed. For example, if 10 and 15 are disclosed, then 11, 12, 13, and 14 are also disclosed.

[0038] References in the specification and concluding claims to parts by weight of a particularelement or component in a composition denotes the weight relationship between the element or component and any other elements or components in the composition or article for which a part by weight is expressed. Thus, in a compound containing 2 parts by weight component X and 5 parts by weight component Y, X and Y are present at a weight ratio of 2:5, and are present in such ratio regardless of whether additional components are contained in the compound.

[0039] As used herein, the terms ‘optional’ or ‘optionally’ means that the subsequently describedevent or circumstance can or cannot occur, and that the description includes instances where said event or circumstance occurs and instances where it does not. In an aspect, a disclosed method can optionally comprise one or more additional steps, such as, for example, repeating an administering step or altering an administering step.

[0040] The present disclosure also contemplates that in some aspects, any feature or combinationof features set forth herein can be excluded or omitted. To illustrate, if the specification states that a complex comprises components A, B and C, it is specifically intended that any of A, B or C, or a combination thereof, can be omitted and disclaimed singularly or in any combination.

[0041] As used herein, “and / or” refers to and encompasses any and all possible combinations ofone or more of the associated listed items, as well as the lack of combinations where interpreted in the alternative (“or”).

[0042] As used herein “pulmonary infection” refers to lung infection.

[0043] As used herein “Pneumonia” refers to an inflammatory condition of the lung, especiallyof the alveoli (microscopic air sacs in the lungs), associated with fever, chest symptoms, and consolidation on a chest radiograph. Patients suffering from infectious pneumonia often have a productive cough, fever accompanied, shortness of breath, sharp or stabbing chest pain during deep breaths, confusion, and an increased respiratory rate. Pneumonia can be classified in several ways. Pneumonia is most commonly classified by where or how it was acquired (community- acquired (CAP), aspiration, healthcare-associated, hospital-acquired, and ventilator-associated pneumonia), but may also be classified by the area of lung affected (lobar pneumonia, bronchialpneumonia and acute interstitial pneumonia), or by the causative organism. Although more thanone hundred strains of microorganisms can cause pneumonia, only a few are responsible for most cases. The most common types of infectious agents are viruses and bacteria and to a lesser extend fungi or parasites. Bacteria causing pneumonia can include Haemophilus influenzae, 8 100243977.1Atty. Docket No.055743-824202 Client’s Ref.: M23-284L-WO1-a Chlamydophila pneumoniae, Mycoplasma pneumoniae, Staphylococcus aureus, Moraxella catarrhalis, Legionella pneumophila and gram-negative bacilli. Pneumonia caused by fungus include Coccidioidomycosis or Valley fever. The two species of Coccidioides that can cause coccidioidomycosis include Coccidioides posadasii and Coccidioides immitis. Other fungi cause pneumonia, non-limiting examples of which include pneumonia caused by Histoplasmacapsulatum, Blastomyces, Cryptococcus neoformans, Aspergillus, Candida, and Pneumocystisjiroveci.

[0044] As used interchangeably herein, the terms “Coccidioidomycosis” and “Valley fever” referto a fungal pneumonia endemic to the arid and semi-arid regions of North and South America. It is estimated there are 350,000 new cases each year and in endemic and highly populated regions up to 30% of community acquired pneumonia may be caused by Valley fever. It is currently difficult to diagnose, in part because symptomatic primary pulmonary infection often resembles bacterial pneumonia leading to a misdiagnosis and inappropriate treatment with antibiotics.

[0045] As used herein, “Volatile organic compounds” or “VOCs” refer to compounds in the gasphase that are produced by the subject and / or by any microorganism that may be infecting the subject. VOC further refers to organic compounds that have a nonzero vapor pressure at room- temperature (higher than 1 part-per-trillion, 1 ppt). Their vapor pressure results from a comparatively low boiling point, which causes large numbers of molecules to evaporate from the liquid or solid form of the compound and enter the surrounding air. An organic compound is any member of a large class of gaseous, liquid, or solid chemical compounds whose molecules contain carbon. In some aspects, a VOC is any organic compound having an initial boiling point less than or equal to 350° C, particularly 300° C, measured at a standard atmospheric pressure of 101.3 kPa. In some aspects, VOC has a boiling point in the range of 50 to 250° C.

[0046] As used herein “discriminatory VOCs” refers to VOCs, the relative abundance of whichcan discriminate between subjects infected with either Coccidioides, another microbial lung pathogen, or no detectable infection.

[0047] As used herein, the term "Random Forest" refers generically to a learning algorithm thatincorporates an ensemble classifier that uses the methods developed by Breiman and Cutler, and specifically to the algorithm as implemented in the commercially available Radom Forest software package. (See, e.g., Breiman, L. and Cutler, A. (2005) Random Forests. www.stat.berkeley.edu / ~breiman / RandomForest; and Breiman, L. and Cutler, A. (2012) Breiman and Cutler’s Random Forests for Classification and Regression. Documentation for Package ‘randomForest’ Version 4.6-2). 9 100243977.1Atty. Docket No.055743-824202 Client’s Ref.: M23-284L-WO1-a

[0048] As used herein, “abundance” refers to an amount of a particular analyte (e.g., VOCs)present in the sample. The amount may be a concentration, number, ratio, proportion, or a percentage of the analyte compared to the control sample or determined using a standard curve. The amount may be an absolute amount, or a relative amount in comparison to a reference amount.

[0049] As used herein, “relative abundance” refers to relative abundance. Relative abundance canbe the abundance of one VOC relative to another VOC or relative to all detected VOCs.

[0050] As used herein, “profile” refers to a set of at least two VOC abundances obtained from atest sample or a reference sample, which may be presented in graphic form such as a spectrum, and which portray the significant VOC features of the test sample or reference sample. In some aspects, the profile comprises composite information from the measurement of two or more VOCs that is diagnostic.

[0051] As used herein, “individual”, “subject”, “host”, and “patient” can be used interchangeablyherein and refer to any mammalian subject for whom diagnosis, treatment, prophylaxis or therapy is desired, for example, humans, pets, livestock, horses or other animals. As used herein, the term “subject” and “patient” are used interchangeably herein and refer to both human and nonhuman animals. The term “nonhuman animals” of the disclosure includes all vertebrates, e.g., mammals and non-mammals, such as nonhuman primates, sheep, dog, cat, horse, cow, chickens, amphibians, reptiles, and the like. In some aspects, the subject can be a human. In other aspects, the subject can be a human in need of treating a pneumonia.

[0052] As used herein, “treatment,” “therapy” and / or “therapy regimen” refer to the clinicalintervention made in response to a disease, disorder or physiological condition manifested by a patient or to which a patient may be susceptible. The aim of treatment includes the alleviation or prevention of symptoms, slowing or stopping the progression or worsening of a disease, disorder, or condition and / or the remission of the disease, disorder or condition.

[0053] As used herein, “effective amount” and “amount effective” can refer to an amount that issufficient to achieve the desired result such as, for example, the treatment and / or prevention of a disease and / or disorder. As used herein, the terms “effective amount” and “amount effective” can refer to an amount that is sufficient to achieve the desired effect on an undesired condition (a disease and / or disorder). For example, a “therapeutically effective amount” refers to an amount that is sufficient to achieve the desired therapeutic result or to have an effect on undesired symptoms but is generally insufficient to cause adverse side effects.

[0054] As used herein, a sample may be of any biological tissue, fluid, or cell from the subject.The sample can be solid or fluid. The sample can be a heterogeneous cell population. Non-limiting examples of suitable biological samples include sputum, bronchoalveolar lavage fluid (BALF), 10 100243977.1Atty. Docket No.055743-824202 Client’s Ref.: M23-284L-WO1-a exhaled breath, serum, blood, blood cells (e.g., white cells), a biopsy, urine, peritoneal fluid, pleural fluid, or cells derived therefrom. The biopsy can be a fine needle aspirate biopsy, a core needle biopsy, a vacuum assisted biopsy, an open surgical biopsy, a shave biopsy, a punch biopsy, an incisional biopsy, a curettage biopsy, or a deep shave biopsy. Biological samples may also include fresh, frozen, or preserved samples. Methods of collecting a biological sample from a subject are well known in the art. In some aspects, the biological sample is a peripheral blood sample. In some aspects, the biological sample is bronchoalveolar lavage fluid (BALF). In some aspects, the biological sample is a sputum sample. In some aspects, the biological sample is a breath sample.

[0055] Sample from the subject can be procured one or more times, before, during and / or afterdiagnosis. In some aspects, samples can be procured from the subject before, during, and / or after treatment. In some aspects, sample can be procured from the subject prior to the start of treatment. In some aspects, sample can be procured from the subject undergoing a treatment. Additionally, samples can be procured repeatedly at multiple stages after initial sample procurement, to determine and / or monitor the disease progression in a subject.

[0056] In some aspects, a control sample can be procured from a healthy subject. In some aspects,the control is a person or persons with similar characteristics to the subject. In some aspects, the control is a person or persons with a pneumonia caused by another pathogen other than Coccidioides. Put differently, since the model discriminates Valley fever from other causes of pneumonia, the control is not necessarily a healthy subject, but rather an individual with a pneumonia that caused by a pathogen other than Coccidioides.

[0057] In some aspects, the control can be an average of the combination of disclosed VOCbiomarkers relative abundance from different healthy sources (e.g., more than one healthy control subject). In some aspects, the control sample can be a pooled sample.

[0058] Unless otherwise defined, all technical terms used herein have the same meaning ascommonly understood by one of ordinary skill in the art to which this disclosure belongs.B. Sample Collection and Processing

[0059] Disclosed herein is a non-invasive method for identifying or diagnosing fungal pneumoniain subjects suspected of having pneumonia. The present disclosure further encompasses methods of diagnosis, prognosis, monitoring for presence of pneumonia by determining the relative abundance of signature set of volatile organic compounds (VOCs) in a sample from the subject. In some aspects, the disclosure encompasses a method for diagnosing and / or distinguishing subjects with fungal pneumonia from subjects with bacterial pneumonia. In further aspects, the 11 100243977.1Atty. Docket No.055743-824202 Client’s Ref.: M23-284L-WO1-a methods comprise diagnosing and / or distinguishing subjects with Coccidioides infection from subjects having other infections (for e.g., bacterial infection).

[0060] In some aspects, the subject is a mammal, preferably a human. In some aspects, the subjectis a subject who is at risk of developing pneumonia, a subject who is suspected of having pneumonia, or a subject who is afflicted with pneumonia.

[0061] In such aspects, the method comprises obtaining a sample from the subject. In someaspects, the sample is a sample obtained from the lung of the subject. In some aspects, the sample obtained from the subject is a breath sample. In some aspects, the breath sample comprises exhaled breath from a subject suspected, at risk or diagnosed with pneumonia. In some aspects, obtaining the breath sample can be performed using any method known in the art. In some aspects, the breath sample may be collected using a breath collector apparatus. In some aspects, the breath collector apparatus may be designed to collect alveolar breath samples. Exemplary breath collector apparatuses within the scope of the present invention include apparatuses approved by the American Thoracic Society / European Respiratory Society (ATS / ERS); Silkoff et al., Am. J. Respir. Crit. Care Med., 2005, 171, 912). In some aspects, breath sample may be obtained by sampling, e.g., at the intensive care unit, manually using an air-bag (for e.g., Tedlar bag) or a glass tube. In such aspects, an air-bag such as a gas tight pouch with typical 1-2 liter volume can be used to obtain breath. In some aspects, a glass tube may further comprise a suitable adsorbent, contained in an appropriate device such as a thermal desorption tube. In some aspects, the method may involve direct monitoring of VOCs in the breath by aspiration or by injection into analytical instrument (for e.g., proton-transfer-reaction time-of-flight mass spectrometer, ion mobility spectrometer, gas chromatography-mass spectrometry). In some aspects, VOCs can be directly extracted from the air bag (for e.g., Tedlar bag). In some aspects, the VOCs can be transferred from the air-bag to a sorbent (for e.g., by a pump).

[0062] In further aspects, a sample comprising bronchoalveolar lavage fluid (BALF) may beobtained from the subject. In such aspects, BALF may be obtained using a bronchoscope. In some aspects, BALF may be obtained using a catheter, a syringe, an aspiration device, or a vacutainer.

[0063] In further aspects, a sample obtained from the subject can be analyzed directly for one ormore VOCs, for example, from the breath of a patient suspected of having pneumonia. Alternatively, the sample can be cultured in a suitable growth medium to allow growth and metabolism of microorganism in the sample. In certain aspects, the disclosure involves taking a BALF from a subject and placing it in media, for example, with microfluidics, or in culture, for example, with conventional culturing methods. The microorganism, if present, is stimulated to metabolize. The headspace (gaseous phase) generated as a result of this metabolism may be 12 100243977.1Atty. Docket No.055743-824202 Client’s Ref.: M23-284L-WO1-a collected and may be tested for the presence of at least one metabolite indicative of the microorganism in that growth media.

[0064] In certain aspects, the sample is pre-concentrated prior to the measurement of VOCs. Non-limiting examples of breath concentrators include, but are not limited to, Solid Phase Microextraction (SPME), Sorbent Tubes, and Cryogenic Condensation.

[0065] SPME - The SPME technique is based on a fiber coated with a liquid (polymer), a solid(sorbent), or combination thereof. The fiber coating extracts the compounds from the sample either by absorption (where the coating is liquid) or by adsorption (where the coating is solid). Non-limiting examples of coating polymers include polydimethylsiloxane, polydimethylsiloxane- divinylbenzene and polydimethylsiloxane-carboxen.

[0066] Sorbent Tubes - Sorbent tubes are typically made of glass and contain various types ofsolid adsorbent material (sorbents). Commonly used sorbents include activated charcoal, silica gel, and organic porous polymers such as Tenax® and Amberlite™ XAD resins. Sorbent tubes are attached to air sampling pumps for sample collection. A pump with a calibrated flow rate in ml / min draws a predetermined volume of air through the sorbent tube. Compounds are trapped onto the sorbent material throughout the sampling period.

[0067] Cryogenic Condensation - Cryogenic condensation is a process that allows recovery ofvolatile organic compounds (VOCs) for reuse. The condensation process requires very low temperatures so that VOCs can be condensed. In some aspects, chlorofluorocarbon (CFC) refrigerants are used to condense the VOCs. In some aspects, liquid nitrogen is used in the cryogenic (less than -160 °C) condensation process.C. Detection of VOC

[0068] VOCs as disclosed herein may be detected or relative abundance assessed using varioustechnologies or combinations of technologies including, but not limited to: gas chromatography (GC); spectrometry, for example mass spectrometry (including quadrupole, time of flight, tandem mass spectrometry, ion cyclotron resonance, and / or sector (magnetic and / or electrostatic)), ion mobility spectrometry, field asymmetric ion mobility spectrometry, Differential Mobility Spectrometry (DMS); fuel cell electrodes; light absorption spectroscopy; nanoparticle technology; flexural plate wave (FPW) sensors; biosensors that mimic naturally occurring cellular mechanisms; electrochemical sensors; photoacoustic equipment; laser-based equipment; electronic noses (bio-derived, surface coated); and / or various ionization techniques.

[0069] In some aspects, the VOCs are detected or assessed using GC. In some aspects, the GC iscoupled to a mass spectrometer (GC-MS). In this method, the GC utilizes a capillary column 13 100243977.1Atty. Docket No.055743-824202 Client’s Ref.: M23-284L-WO1-a having characteristic dimensions (length, diameter, film thickness) as well as characteristic phase properties. The difference in the chemical properties of different molecules in a mixture allows the separation of the molecules as the sample travels through the column, wherein each molecule has a characteristic time (termed retention time) in which it passes through the column under set conditions. This allows the mass spectrometer to capture, ionize, accelerate, deflect, and detect the ionized molecules separately. The MS signal is obtained by ionization of the molecules or molecular fragments and measurement of their mass to charge ratio by comparing it to a reference collection. In some aspects, the GC-MS method includes but not limited to two-dimensional gas chromatography mass spectrometry (GC^GC-MS) and gas chromatography-tandem mass spectrometry (GC-MS / MS).

[0070] In some aspects, VOCs are prepared from samples using head space solid phasemicroextraction (SPME) coupled to gas chromatography and mass spectrometry (SPME-GC- MS). In some aspects, SPME-GC×GC-TOFMS is used. In some aspects, thermal desorption coupled to GC×GC-TOFMS (TD-GC×GC-TOFMS) is used. In such aspects, VOCs in the headspace of the samples transferred and sealed into sterilized headspace vials (for e.g., 2 mL gas chromatography headspace vials) are absorbed to concentrating solid phase microextraction (SPME) fiber. The SPME fiber is heated and VOCs are desorbed into one or more detection systems, for example a GC-MS detection system, and data is acquired. In further aspects, the data output is further analyzed for one or more VOCs that distinguish Coccidioides infection in the sample. The detection and analysis of VOCs can include determining the presence or absence, or alternatively the relative abundance, of headspace VOCs in the gas phase.

[0071] In one aspect, the present disclosure provides a VOC signature that provides accuratedetection of pneumonia, and discrimination of pneumonia types, with high degrees of accuracy as described herein. Methods for identifying VOC signatures associated with pneumonia, bacterial pneumonia and coccidioidomycosis or Valley fever generally comprise obtaining test samples from the lung, and control samples; analyzing the test and control samples by mass spectrometry to obtain abundance values for a plurality of VOCs; and applying a statistical modeling technique to select for a plurality of VOCs that distinguish test samples from control samples with a predetermined accuracy. Test samples are from subjects with coccidioidomycosis or another cause of pneumonia, either of which is confirmed using known diagnostic methods as described above. Control samples are from subjects confirmed to be free of bacterial pneumonia or coccidioidomycosis or free of both, also using known diagnostic methods for each. For avoidance of confusion, it should be understood that in some aspects, the control is a person or persons with a pneumonia caused by another pathogen other than Coccidioides. Put differently, since the model 14 100243977.1Atty. Docket No.055743-824202 Client’s Ref.: M23-284L-WO1-a discriminates Valley fever from other causes of pneumonia, the control is not necessarily a healthy subject, but rather an individual with a pneumonia that is caused by a pathogen other than Coccidioides.

[0072] The VOCs that distinguish test samples from control samples comprise the VOC signaturefor that disease.

[0073] Samples from a subject’s lungs can be obtained as detailed above. Samples are analyzedusing mass spectrometry, preferably coupled with gas chromatography, to yield abundance measures for a plurality of molecular features, i.e., VOCs. The abundance value for each VOC may be obtained from a measurement of the area under the peak for the nominal mass or the monoisotopic mass of the VOCs. Identification and extraction of VOCs involves finding and quantifying all the known and unknown compounds / metabolites down to the lowest abundance and extracting all relevant spectral and chromatographic information. Algorithms are available to identify and extract molecular features such as VOCs. Such algorithms include for example the Molecular Feature Extractor (MFE) by Agilent. MFE locates ions that are covariant (rise and fall together in abundance) but the analysis is not exclusively based on chromatographic peak information. The algorithm uses the accuracy of the mass measurements to group related ions- related by charge-state envelope, isotopic distribution, and / or the presence of adducts and dimers. It assigns multiple species (ions) that are related to the same neutral molecule (for example, ions representing multiple charge states or adducts of the same neutral molecule) to a single compound that is referred to as a feature. Using this approach, the MFE algorithm can locate multiple compounds within a single chromatographic peak. Specific parameters for MFE may include a minimum ion count of 600, an absolute height of 2,000 ion counts, and compound ion count threshold of 2 or more ions. Once the molecular feature such as a VOC has been identified and extracted, the area under the peak for the nominal or monoisotopic mass is used to determine the abundance value for the molecular feature. The nominal mass and the monoisotopic mass are the sum of the masses of the atoms in a molecule using the unbound, ground-state, rest mass of the principal (most abundant) isotope for each element instead of the isotopic average mass. Nominal mass and monoisotopic mass are typically expressed in unified atomic mass units (u), also called daltons (Da).

[0074] A VOC is a molecular feature identified as a potential molecular feature for utilization ina VOC signature of the present disclosure if it is present in at least 15% of either the Coccidioides positive samples, all test samples, or the control samples. For example, the molecular feature may be present in at least 15%, at least 20%, at least 25%, at least 30%, at least 35%, at least 40%, at least 45%, at least 50%, at least 55%, at least 60%, at least 65%, at least 70%, at least 75%, at 15 100243977.1Atty. Docket No.055743-824202 Client’s Ref.: M23-284L-WO1-a least 80%, at least 85%, at least 90%, at least 95% or 100% of either the test samples or the control samples.

[0075] To increase the stringency of the VOC signature, replicates of the test samples and controlsamples may be analyzed. For example, the test samples and control samples may be analyzed in duplicate. Alternatively, the test biological samples and control biological samples may be analyzed in triplicate. Additionally, the test samples and control samples may be analyzed four, five, or six times. The replicate analysis is used to down-select the plurality of molecular features. The down-selection results in a VOC signature with increased stringency.

[0076] Once a plurality of potential VOC molecular features has been generated, a statisticalmodeling technique may be applied to select for the VOC molecular features that provide an accuracy of disease detection that is clinically meaningful. Several statistical models are available to select the molecular features that comprise a VOC signature of the present disclosure. Non- limiting examples of statistical modeling techniques include linear discriminant analysis (LDA), classification tree (CT) analysis, random forests, and LASSO (least absolute shrinkage and selection operator) logistic regression analysis. Various methods are known in the art for determining an optimal cut-off that maximizes sensitivity and / or specificity to serve as a threshold for discriminating samples obtained from subjects with bacterial pneumonia or coccidioidomycosis. The cut-off can be set as required by situational circumstances. For example, in certain clinical situations it may be desirable to minimize false-positive rates. These clinical situations may include, but are not limited to, the use of an experimental treatment (e.g., in a clinical trial) or the use of a treatment associated with serious adverse events and / or a higher-than- average number of side effects. Alternatively, it may be desirable to minimize false-negative rates in other clinical situations. Non-limiting examples may include treatment with a non- pharmacological intervention, the use of a treatment with a good risk-benefit profile, or treatment with an additional diagnostic agent.

[0077] The pattern of VOCs can be analyzed with a pattern recognition analyzer which utilizesvarious algorithms including, but not limited to, artificial neural networks, multi-layer perception (MLP), generalized regression neural network (GRNN), fuzzy inference systems (FIS), self- organizing map (SOM), radial bias function (RBF), genetic algorithms (GAS), neuro-fuzzy systems (NFS), adaptive resonance theory (ART) and statistical methods including, but not limited to, principal component analysis (PCA), partial least squares (PLS), multiple linear regression (MLR), principal component regression (PCR), discriminant function analysis (DFA) including linear discriminant analysis (LDA), and cluster analysis including nearest neighbor. 16 100243977.1Atty. Docket No.055743-824202 Client’s Ref.: M23-284L-WO1-a

[0078] In some aspects, further analysis is conducted to classify the sample based on the VOCspresent. The relative abundances of the VOCs are input into a classification model. In some aspects, the classification is performed by using an algorithm such as supervised learning, machine learning, or pattern recognition algorithms. Many such algorithms are known in the art. Non- limiting examples of classification algorithms that can be used with the method disclosed herein include Bayes classifier, support vector machine (SVM), linear discriminant functions, Fisher's linear discriminant, C4.5 algorithm tree, K-nearest neighbor (KNN), weighted K-nearest neighbor, Partial least squares (PLS), Hierarchical clustering algorithm, Random Forest, hidden Markov model, Gaussian mixture model (GMM), K-mean clustering algorithm, Ward's clustering algorithm, minimum least squares, and neural network algorithms.

[0079] In an exemplary aspect, the Random Forest algorithm is used for the methods disclosedherein. In this algorithm, decision trees are developed based on different sets of samples and random forest is used to calculate a loss of accuracy of classification when the values of features are randomly permutated between sets of samples. One or more features associated with a loss ofaccuracy of classification are then selected. Features associated with the loss of accuracy are thenselected as indicative features.

[0080] The learning algorithms generally begin with a training protocol to produce a database ofVOCs associated with different pneumonia types and / or control samples. This database is then stored, and the classification of an unknown sample is then performed based on the results of the learning algorithm's comparison of the VOCs extracted from the sample with those of the training procedure. In some aspects, the parameters of the fit are chosen such that the classification is done based on VOCs that have a predetermined minimum significance threshold. In some aspects, this minimum significance threshold is a 95% confidence limit. Non- limiting examples of statistical significance tests for determining whether a particular feature obtained by PCA passes the predetermined significance threshold include χ, Wilcoxon test, and Student's t-test.

[0081] In further aspects, the operating characteristics of the classification model can be furthercharacterized using sensitivity, specificity, and the area under the receiver operating characteristic curve with a specified confidence interval (CI), for example a CI of 95%.

[0082] In some aspects, discriminatory VOCs are identified by comparing samples from subjectswith pneumonia to samples from healthy subjects, samples from subjects with Coccidioides- positive infection to samples from subjects with other infection (i.e., non- Coccidioides infection), and samples from subjects with Coccidioides-positive infection to samples from subjects with bacteria-positive infection. 17 100243977.1Atty. Docket No.055743-824202 Client’s Ref.: M23-284L-WO1-a

[0083] In some aspects, further provided herein is a classification model comprisingdiscriminatory VOCs for diagnosis and / or classification of a subject. In some aspects, 14 discriminatory VOCs are used in the model disclosed herein. In some aspects, the relative abundances of VOCs [161, 199, 215, 254, 368, 377, and 515] is used to distinguish pneumonia infected samples from non-infected samples, with sensitivity and specificity of 95% and 71%, respectively. In some aspects, the relative abundances of VOCs [161, 215, 368, 377, and 515] are used to distinguish presence and absence of pneumonia infection. In some aspects, the relative abundance of VOCs [150, 182, 255, 352, and 374] are used to distinguish Coccidioides-positive pneumonia from non-Coccidioides infected pneumonia, with sensitivity and specificity of 90% and 95%, respectively. In some aspects, the relative abundances of VOCs [46, 150, 255, and 369] are used to distinguish Coccidioides pneumonia (Valley fever) from bacterial pneumonia infection, with sensitivity and specificity of 93% and 96%, respectively.D. Methods

[0084] Further disclosed herein are non-invasive or minimally invasive methods using one ormore discriminatory VOCs for detection and diagnosis of pneumonia.

[0085] In some aspects, the disclosure encompasses a minimally invasive or non-invasivecomputer-implemented method for identifying a pulmonary infection in a subject. In some aspects, the method comprises determining or having determined from a mass spectrum of a test sample from a lung of the subject, the relative abundance of each of a plurality of volatile organic compounds (VOCs) to prepare a sample VOC dataset; and generating, by a processor, a machine learning prediction of a diagnosis selected from a pneumonia, a bacterial pneumonia, and a fungal pneumonia infection afflicting the subject using the sample VOC dataset as input to a machine learning model in view of a plurality of VOC abundances predetermined to be predictive of the pneumonia, bacterial pneumonia or fungal pneumonia.

[0086] In some aspects, the disclosed method includes an algorithmic machine learning techniquecapable of adapting to complex data sets (e.g., panel of VOCs) and of making decisions based on these data sets. In some aspects, a single statistical classifier system of learning such as a decision / classification tree (for e.g., random forest (RF) or classification and regression tree (C & RT)) is used. In other aspects, a combination of 2, 3, 4, 5, 6, 7, 8, 9, 10, or more statistical classifier systems of learning is used, preferably in tandem. Examples of statistical classification systems of learning include, but are not limited to, those that use inductive learning (for example, decision / classification trees such as classification and regression trees (C & RT), reinforced trees, etc.), learning probably approximately correct (PAC), connectionist learning (for example, neural 18 100243977.1Atty. Docket No.055743-824202 Client’s Ref.: M23-284L-WO1-a networks (NN), artificial neural networks (ANN), neuro-diffuse networks (NFN), network structures, perceptrons such as multilayer perceptrons, forward feeding networks of multiple layers, neural network applications, Bayesian learning in belief networks, etc.), reinforcement learning (e.g., passive learning in a known environment such as naive learning, adaptive dynamic learning and time difference learning, passive learning in an unfamiliar environment, active learning in an environment unknown, action-value learning functions, reinforcement learning applications), and genetic algorithms and evolutionary programming. Other statistical classifier systems of learning include support vector machines (for example, Kernel methods), multivariate adaptive regression striations (MARS), Levenberg-Marquardt algorithms, Gauss-Newton algorithms, gaussian mixtures, gradient descending algorithms and quantification of learning vectors (LVQ).

[0087] In some aspects, the disclosed method uses random forests. Random forests use manyindividual decision trees and decide the class by selecting the mode (that is, occurring more frequently) of the classes as determined by the individual trees. Random forest analysis can be carried out, for example, using the Random Forests® software available from Salford Systems (San Diego, CA). See, for example, Breiman, Machine Learning, 45: 5-32 (2001); and http: / / stat- wvvw.berkeley.edu / usersA)reiman / RandomForests / cc_home.htm, for a description of random forests.

[0088] In some aspects, Analysis of classification and regression trees can be carried out, forexample, using the CART® software available from Salford Systems or the TIBO® Statistica data analysis software available from StatSoft, Inc. (Tulsa, OK), A description of classification and regression trees is found, for example, in Breiman et al. "Classification and Regression Trees," Chapman and Hall, New York (1984); and Steinberg et al., "CART: Tree-Structured Non- Parametric Data Analysis," Salford Systems, San Diego, (1995).

[0089] In some aspects, the statistical methods and models described herein use the random forestmethod for model building. A random forest is usually a classifier made up of many decision trees with a random component to build each tree. Each decision tree addresses the same classification problem, but with a different collection of examples and a different subset of features randomly selected from the set of sample data provided.

[0090] In some aspects, a single tree could be constructed using random 2 / 3 of the availablesamples (for example, training set), with a random 2 / 3 of the selected characteristics to make a split decision in each node of the tree. Once the forest is built during training, new examples are classified by taking a vote through all the decision trees. In the simplest case for a two-class random forest classifier, the class with the most votes wins. In other cases, the limit for a winning 19 100243977.1Atty. Docket No.055743-824202 Client’s Ref.: M23-284L-WO1-a number of votes, as established during training is pre-established to optimize performance measurements. In some cases, if false positives are more expensive than false negatives, the limit could be set higher. Typically, many trees are cultivated and trained to random subsets of training data and predictions are made by averaging the results of individual trees.

[0091] In some aspects, the random forest modeling of the present disclosure comprises anassembly method, wherein the forest comprises thousands of decision trees. A decision tree can be used to predict when generating a sequence of questions (for example, divisions). In a given node in the sequence, the question (for example, division) that is made depends on the answers to the previous questions. A tree selects the best division in each node based on how well it separates the classes. Trees are produced by using different random sub-samples of training data. In other cases, trees are produced using different random subsets of markers at each node. During prediction mode, the probability of each class is a fraction of the trees in the algorithm that predicts it.

[0092] In some aspects, the general stages of random forest include: (1) building a conventionalrecursive division tree from almost 2 / 3 of the data available for training, where no subsequent debugging or pruning is employed; (2) consider only a small and different subset of variables available in each node, where the subset is not greater than the square root of the number of variables; (3) repeating steps 1 and 2 a plurality of times to create several trees (one forest); and (4) during the prediction mode, report a grade that is the fraction of trees in the forest that predicts a given class. A limit can then be used to produce a prediction of classes.

[0093] In certain aspects, the diagnostic algorithm is established using a retrospective cohort ofpredicted diagnoses with presence, absence and / or known levels of VOCs. In some aspects, the examples used (for example, training and test sets) to construct the random forest come from studies using the methods described herein.

[0094] In some aspects, correlations, and associations for all possible pairwise combinations ofthe variables (e.g., VOCs and known disease diagnoses) used in some part of the diagnostic algorithm can be calculated. For pairs of continuous variables and for pairs of continuous-binary variables, Pearson correlation coefficients and their p-values can be calculated. For pairs of binary-binary variables, the association can be evaluated with the analysis of squares Chi. All calculations can also be carried out using the MATLAB software program.

[0095] In further aspects, the machine learning model includes random forest (RF) classificationand is trained by the processor. In such aspects, the training can comprise accessing at least one sample VOC dataset and at least one VOC training dataset associated with the pneumonia, bacterial pneumonia or fungal pneumonia, wherein the VOC training dataset comprises the 20 100243977.1Atty. Docket No.055743-824202 Client’s Ref.: M23-284L-WO1-a plurality of VOCs predetermined to be predictive of the pneumonia, bacterial pneumonia or fungal pneumonia; analyzing the VOC mass spectrum of each sample VOC dataset; and ranking an ability of each VOC to classify a sample for a diagnosis of pneumonia, bacterial pneumonia or fungal pneumonia.

[0096] In further aspects, the training further comprises building a first plurality of classifier trees;selecting a first set of top VOCs from the first plurality of classifier trees; building a second plurality of classifier trees using the set of top VOCs; selecting and saving a second set of top VOCs from the second plurality of classifier trees, the second set of top classifiers defining selected VOCs; and performing multiple times to select the best set of top VOCs.

[0097] In some aspects, the random forest classification includes building an ensemble of aplurality of classifier trees, where the final prediction for a test sample is obtained by majority vote on the combination of predictions of all of the plurality of classifier trees.

[0098] In some aspects, the fungal pneumonia is coccidioidomycosis, caused by the fungusCoccidioides. In some aspects, the method further comprises generating, by the processor, a plurality of VOC predictors for pneumonia, bacterial pneumonia and fungal pneumonia. In some aspects, the plurality of VOC predictors comprises [161, 215, 368, 377, and 515]. In some aspects, the plurality of VOC predictors comprises [161, 199, 215, 254, 368, 377 and 515]. In some aspects, the plurality of VOC predictors comprises [46, 150, 255, and 369]. In some aspects, the plurality of VOC predictors comprises [150, 182, 255, 352, and 374].

[0099] In some aspects, the VOC predictors correctly predict pneumonia infection in the subjectwith an error rate of no more than 20%. In some aspects, the disclosed error rate is no more than 20%, 19%, 18%, 17%, 16%, 15%, 14%, 13%, 12%, 11%, 10%, 9%, 8%, 7%, 6%, 5%, 4%, 3%, 2%, or 1%. In some aspects, the error rate is no more than 12%.

[0100] In some aspects, the VOC predictors correctly predict fungal pneumonia infection in thesubject with an error rate of no more than 10%. In some aspects, the disclosed error rate is no more than 10%, 9%, 8%, 7%, 6%, 5%, 4%, 3%, 2%, or 1%. In some aspects, the error rate is no more than 5%.

[0101] In further aspects, provided herein is a method for distinguishing a fungal pneumonia froma bacterial pneumonia. In such aspects, the method comprises, from a mass spectrum of a test sample from a lung of the subject, determining or having determined the relative abundance of each of a plurality of volatile organic compounds (VOCs) to prepare a sample VOC dataset; and generating, by a processor, a machine learning prediction of fungal pneumonia infection afflicting the subject using the sample VOC dataset as input to a machine learning model in view of a plurality of VOC abundances predetermined to be predictive of the fungal pneumonia. 21 100243977.1Atty. Docket No.055743-824202 Client’s Ref.: M23-284L-WO1-a

[0102] In such aspects, the machine learning model includes Random Forest (RF) classification.In some aspects, the machine learning model and is trained by the processor, by steps including: accessing at least one sample VOC dataset and at least one VOC training dataset associated with the fungal pneumonia, wherein the VOC training dataset comprises the plurality of VOC abundances predetermined to be predictive of the fungal pneumonia; analyzing the VOC mass spectrum of each sample VOC dataset; and ranking an ability of each VOC to classify a sample for a diagnosis of fungal pneumonia.

[0103] In further aspects, the training further comprises: building a first plurality of classifiertrees; selecting a first set of top VOCs from the first plurality of classifier trees; building a second plurality of classifier trees using the set of top VOCs; selecting and saving a second set of top VOCs from the second plurality of classifier trees, the second set of top classifiers defining selected VOCs, and performing multiple times to select the best set of top VOCs.

[0104] In some aspects, the Random Forest classification includes building an ensemble of aplurality of classifier trees, where the final prediction for a test sample is obtained by majority vote on the combination of predictions of all of the plurality of classifier trees.

[0105] In some aspects, the fungal pneumonia is coccidioidomycosis, caused by the fungusCoccidioides. In some aspects, the method further comprises generating, by the processor, a plurality of VOC predictors for fungal pneumonia. In some aspects, the plurality of VOCpredictors comprises [150, 182, 255, 352, and 374]. In some aspects, wherein the VOC predictorscorrectly predict fungal pneumonia infection in the subject with an error rate of no more than 10%. In some aspects, the disclosed error rate is no more than 10%, 9%, 8%, 7%, 6%, 5%, 4%, 3%, 2%, or 1%. In some aspects, the error rate is no more than 7%.

[0106] In some aspects, the plurality of VOC predictors further comprises [46, 150, 255, and369], wherein the VOC predictors correctly predict fungal pneumonia infection in the subject with an error rate of no more than 10%. In some aspects, the disclosed error rate is no more than 10%, 9%, 8%, 7%, 6%, 5%, 4%, 3%, 2%, or 1%. In some aspects, the error rate is no more than 5%.

[0107] In some aspects, the sample is a bronchoalveolar lavage sample, a sputum sample, or abreath sample. In some aspects, the mass spectrometry comprises two-dimensional gas chromatography-time-of-flight mass spectrometry (GC×GC-TOFMS).

[0108] In some aspects, the methods disclosed herein aid in the decision on whether to prescribeantifungal therapy to a newly diagnosed patient. There are serious side effects to antifungal medications, which are taken for at least six months, so a test that can determine disease severity and the strength of the patient's immune system would help doctors prescribe antifungals to only patients who will need them to get better. 22 100243977.1Atty. Docket No.055743-824202 Client’s Ref.: M23-284L-WO1-a

[0109] Therefore, the present disclosure further encompasses a method for treating a subject atrisk, having or suspected of having a pneumonia. In such aspects, the method comprises distinguishing or having distinguished the suspected pneumonia as a bacterial pneumonia or a fungal pneumonia. The method further comprises classifying the subject as having a fungal pneumonia or a bacterial pneumonia and treating the subject with an antibiotic when the subject is distinguished as having a bacterial infection. In some aspects, the method further comprises not treating the subject with an antibiotic when the subject is distinguished as having a fungal infection. In such aspects, the disclosed method further comprises administering an effective amount of antifungal medication to subjects determined to have a fungal pneumonia or Coccidioidomycosis (Valley fever).

[0110] In further aspects, the methods disclosed herein aid in monitoring the efficacy of antifungaltherapy. In such aspects, subjects being treated with antifungals medications, can be tested using methods described herein (e.g., weekly, biweekly, or monthly) to monitor the disease severity or progression. In further aspects, the method comprises altering the dosage, or modifying the antifungal medication, or stopping the treatment.

[0111] In some aspects, medication for bacterial pneumonia comprises Glucocorticoids, IVampicillin or nafcillin plus gentamicin or cefotaxime, Ceftriaxone or cefotaxime, Azithromycin erythromycin, cephalosporin, (as single agents) ampicillin and sulbactam (Unasyn®), piperacillin and tazobactam (Zosyn®), or ticarcillin and clavulanate (Timentin®), Cephalosporins, trimethoprim, sulfamethoxazole, Fluoroquinolones, levofloxacin, moxifloxacin, gatifloxacin, or any combination thereof. In some aspects, antifungal medication can comprise Amphotericin B, fluconazole, itraconazole, posaconazole, voriconazole, or any combination thereof. Dosage can vary, and can be administered in one or more dose administrations daily, for one or several days. The dosage of the medication can be determined and adjusted by a physician. Guidance can be found in the literature for appropriate dosages for given classes of medication.E. System and Device

[0112] According to one aspect, the disclosure further encompasses a system for detection anddiagnosis of pneumonia. In such aspects, the system is computer implemented and executes methods disclosed herein. It will be appreciated that the storage devices and storage media are aspects of machine-readable storage that are suitable for storing a program or programs that, when executed, implement various aspects of the present disclosure, such as an output that identifies the presence or absence of the identified VOCs, identifying a pulmonary infection, distinguishing a fungal pneumonia from a bacterial pneumonia, or any combination thereof. Accordingly, in some aspects provided is a program comprising code for implementing a method disclosed herein and 23 100243977.1Atty. Docket No.055743-824202 Client’s Ref.: M23-284L-WO1-a a machine readable storage storing such a program. In some aspects, the system comprises a device (for e.g., mass spectrometer) which can be coupled to a processor configured with computer readable instructions to provide an analysis output that identifies the presence or absence of the identified VOCs. In further aspects, the processor can execute a machine learning algorithm and / or classifier programs (for e.g., Random Forest, neural network). In some aspects, the processor is able to execute a pattern recognition program. In some aspects, a processor may be implemented as appropriate in a hardware, a software, a firmware, or any combination thereof. Software or firmware implementations of the one or more processors may include computer- readable or machine-readable instructions written in any suitable programming language to perform the various functions described herein.

[0113] In various aspects, the disclosure encompasses a sensor platform. In some aspects, thesensor platform can be configured to identify presence or absence of one or more VOCs disclosed herein, identify a pulmonary infection, distinguish a fungal pneumonia from a bacterial pneumonia, or any combination thereof, in a subject. In some aspects, the sensor platform can comprise a receiver, a sensor; a processor; an output module, or any combination thereof.

[0114] In some aspects, the sensor platform disclosed herein comprises a receiver. In someaspects, the receiver of the sensor platform can be configured to receive a sample from a subject. In some aspects, the sample can be a breath sample, a sputum sample, or a BALF sample. In someaspects, the disclosed receiver can include a mouth piece operable to receive a breath sample froma subject. For example, a subject can insert the mouthpiece in his or her mouth, closing his or her lips around the mouth piece, and the subject can breathe into the mouth piece. In some aspects, the mouth piece can further include an inlet opening to receive the breath or a BALF sample from the subject. In some aspects, the receiver can further comprise a flow path, wherein the sample from the subject can, when received from the mouth piece or inlet, can flow towards the sensor of the sensor platform. In some aspects, the sensor platform can further comprise a sensor configured to detect one or more VOCs disclosed herein. In some aspects, the sensor is configured to detect one or more disclosed VOCs, in a breath sample, a sputum sample, or a BALF sample provided by the subject. In some aspects, the sensor is further configured to collect data associated with a detection of the one or more VOCs disclosed herein. In some aspects, the sensor can include at least one component operable to detect a VOC. Depending on the VOCs to be detected, any number of sensors and / or sensor components can be selected and / or designed to provide suitable sensitivity for detecting the specific VOCs as well as providing for suitable assessment of a relative abundance of one or more VOCs. Suitable sensors and sensor components can include, but are not limited to, electronic sensors, electromechanical sensors, electrochemical sensors, or 24 100243977.1Atty. Docket No.055743-824202 Client’s Ref.: M23-284L-WO1-a any other sensing device or technique that can convert detection of VOCs in a breath or a BALF sample to an electrical signal.

[0115] In some aspects, the sensor can comprise an electrode array. In some aspects, the electrodearray can comprise electrodes made up of a metal. In some aspects, the electrode array can comprise electrode made of a metal selected from Au, Ti, Cu, Ag, Pd, Pt, Ni, Al, or any combinations thereof. In further aspects, the sensor can comprise a conducting polymer or a porous film, in electric contact with the disclosed electrode array. In some aspects, non-limiting examples of a conducting polymer can include polyaniline (PANI), polythiophene, poly(3,4- ethylenedioxythiophene)-poly(styrene-sulfonate) (PEDOT:PSS), polypyrrole, polydiketopyrrolopyrrole, and derivatives, or any combination thereof. In some aspects, the sensor can comprise one or more graphene, gold nanowires, nanomaterials, carbon paint, or any combination thereof. Any type of nanomaterials for detecting VOCs can be used, including nanocomposites, nanotubules, or nanofibers. In some aspects, nanofibers, can include polymeric nanofibers, conducting polymers or, for example, carbon nanofibers (CNF). In some aspects, the sensor can be a electrochemical sensor; a chemiresistor, a chemicapacitor, a nanoparticle sensor; a 2D (two-dimensional) metal carbide / nitride such as a MXene; customized forms 2D Mn+iXn (Ts) Mxene compositions; a 2D (two-dimensional) Ti3C2 nanosheet; Ti3C2Tx; Ti3C2(OH)2; Ti3C2Mxene; a 3D Mxene (3D-M) framework; one or more metal oxide nanoparticles blended with a hybrid structure of graphene; a conducting polymer such as polyaniline (Pani) and polypyrrole (PPY); a supramolecule; a cavitand (macrocyclic compounds based on resorcinarenes); a relatively thin film of a supramolecule or a calixresorcinarene.

[0116] In some aspects, when a breath sample, a sputum sample, or a BALF sample is receivedfrom the receiver of the sensor platform, the samples flows through, over, or otherwise adjacent to at least one sensor or sensor component, which can detect one or more of the disclosed VOCs in the sample. In some aspects, the sensor can, upon detection of one or more VOCs, generate an electronic signal correlating to presence of and / or a relative abundance of a specific VOC. In some aspects, any number of electronic signals may be generated by the sensor depending on the number of detected and assessed VOCs. In such aspects, each of the electronic signals would correlate to presence of and / or a relative abundance of a specific VOCs detected in the sample from the subject. In some aspects, the sensor can be customized to detect any number of patterns identified in the one or more VOCs of the sample from the subject. Such predefined pattern can be associated with identifying a pulmonary infection, distinguishing a fungal pneumonia from a bacterial pneumonia, or any combination thereof. For e.g., the sensor, can be customized or otherwise designed to include one or more sensors operable to detect the specific VOCs as well 25 100243977.1Atty. Docket No.055743-824202 Client’s Ref.: M23-284L-WO1-a as assess the relative abundance of the respective VOCs. The predefined patterns can then be specifically correlated with a particular diagnosis or condition. In some aspects, the sensor platform can further comprise a processor configured to analyze, process and / or identify the one or more of the detected VOCs. In some aspects, the processor is in communication with the sensor of the sensor platform. In such aspects, the sensor of the sensor platform can transfer or transmit data to the processor. In some aspects, the processor comprises a machine learning algorithm for analyzing, processing or identifying the detected VOCs. In some aspects, the data from the sensor is processed by a neural network or pattern recognition algorithm. In some aspects, an initial result may include identification of one or more VOCs in the sample, identification of a unique sensor derived signal and / or signal pattern of the one or more the VOCs in the sample, an identification of and / or correlation with a condition such as for e.g., a pulmonary infection, distinguish a fungal pneumonia from a bacterial pneumonia, or any combination thereof. In further aspects, the machine learning algorithm can be initially created and / or subsequently trained by receiving and storing any number of previously detected and identified signals and / or signal patterns of one or more VOCs from samples of any number of subjects. In some aspects, the signals and / or signal patterns of one or more VOCs can be received and stored, wherein each signal and / or signal pattern may be correlated with one or more condition such as for e.g., a pulmonary infection, or distinguish a fungal pneumonia from a bacterial pneumonia.

[0117] In some aspects, the machine learning algorithm can comprise a neural network andpattern recognition algorithm, any number of mathematical and computational tools and techniques which can be implemented to process collected data from a sensor, including, but not limited to, using feature extraction and feature selection processes in conjunction with an artificial neural network (ANN), Random Forest, artificial intelligence techniques, a multifactorial approach, leave-one-out cross-validation (LOOCV), nonlinear support vector machine (SVM), multi-layer perception (MLP), generalized regression neural network (GRNN), fuzzy inference systems (FIS), self-organizing map (SOM), radial bias function (RBF), genetic algorithms (GAS), neurofuzzy systems (NFS), adaptive resonance theory (ART) and statistical methods such as canonical discriminant analysis, canonical correlation, principal component analysis (PCA), partial least squares (PLS), multiple linear regression (MLR), principal component regression (PCR), discriminant function analysis (DFA) including linear discriminant analysis (LDA), and cluster analysis including nearest neighbor.

[0118] In some aspects, the processor of the sensor platform can be in communication with anoutput module configured to transmit a result generated by the processor. In some aspects, the output module can prepare and present a suitable output or result for transmission to a display. 26 100243977.1Atty. Docket No.055743-824202 Client’s Ref.: M23-284L-WO1-a Example outputs for a display can include, but are not limited to, a heat map, a graph, a chart, a relative abundance of a specific VOC, or an indication of a condition such as for e.g., a pulmonary infection, distinguishing a fungal pneumonia from a bacterial pneumonia, or any combination thereof.

[0119] In some aspects, the sensor platform can comprise a receiver configured to receive asample from a subject, a sensor configured to detect one or more VOCs disclosed herein; a processor configured to analyze, process and / or identify the one or more of the detected VOCs; an output module configured to transmit a result generated by the processor, or any combination thereof.

[0120] In some aspects, the sensor platform can further include one or more selectively permeablemembranes to permit certain gases in a breath sample, such as nitrogen, oxygen, and / or carbon dioxide to exit while concentrating or otherwise retaining one or more VOCs, for detection and / or identification by the sensor. For example, silicone rubber-type membranes can be used to separate certain gases from mixtures of nitrogen, oxygen, and / or carbon dioxide and other gases. In some aspects, the sensor platform can comprise a breath reservoir. In some aspects, the sensor platform can comprise a heated breath reservoir which prevents condensation of water. In some aspects the sensor platform can comprise a sorbent tube which captures the VOCs (for e.g, activated carbon). In some aspects, VOCs captured in the device can be desorbed from the sorbent tubes and concentrated, separated, identified and / or quantitated in a mass spectrometer.

[0121] In further aspects, the sensor platform can comprise a housing. In some aspects, thehousing can enclose one or more components of the sensor platform. In some aspects, the sensor platform can be configured as a point of care or portable device for detecting one or more VOCs, or diagnosing a condition in a subject such as for e.g., a pulmonary infection, distinguishing a fungal pneumonia from a bacterial pneumonia, or any combination thereof. I. EXAMPLES

[0122] The Examples that follow are illustrative of specific aspects of the invention, and varioususes thereof. They set forth for explanatory purposes only and are not to be taken as limiting the invention. Methods and Materials

[0123] Human Specimen: Fifty-five bronchoalveolar lavage fluid (BALF) samples from theBiospecimens Accessioning and Processing (BAP) lab biorepository were provided by Tom Grys at Mayo Clinic, Phoenix, Arizona. The study was approved by the Institutional Review Boards at Arizona State University and Mayo Clinic (IRB 18-005235). Using chart review, patient samples 27 100243977.1Atty. Docket No.055743-824202 Client’s Ref.: M23-284L-WO1-a were categorized by the infecting microbial organism(s) using the following categories: Coccidioides, fungal, bacterial, viral, multiple, and negative (Table 1). A sample was considered Coccidioides positive if it had a positive Coccidioides serology (enzyme immunoassay, complement fixation and / or immunodiffusion) and / or was confirmed by culture. Samples categorized as fungal were not Coccidioides positive but had a positive fungal culture or fungal smear for another fungus. Bacterial samples were identified by a positive bacterial culture and / or by polymerase chain reaction (PCR). Samples were categorized as viral by PCR. If a sample was considered positive for more than one type of infectious organism (e.g., bacterial positive and viral positive) it was labeled multiple. Samples that did not fit into an infectious category, as described, were labeled negative. 28 100243977.1Atty. Docket No.055743-824202 Client’s Ref.: M23-284L-WO1-a Table 1: Categorization and clinical notes on pneumonia etiology in the Samples used in the study s ely pr eod* ioliadil e raglavisita et29 100243977.1Atty. Docket No.055743-824202 Client’s Ref.: M23-284L-WO1-a Aspergillus F17 Fungal X fumigatus B t i *Samples er than non- tuberculoy , y TM bacterial infection; if a sample has both a bacterial infection and NTM, the notes column is labeled with a + and then the specific NTM is listed.

[0124] Volatile Metabolomics Analysis by HS-SPME-GC×GC–TOFMS: The BALF samples,which had been stored at -80°C, were allowed to thaw at 4°C overnight, and then split into technical triplicates of 200 μL that were transferred and sealed into sterilized 2 mL gas chromatography (GC) headspace vials with Supelco® PTFE / silicone septum magnetic screw caps (Sigma-Aldrich®, St. Louis, MO). All samples were stored in GC headspace vials for up to 14 d at -20°C until analyzed. Samples were randomized for analysis. Volatile metabolites sampling 30 100243977.1Atty. Docket No.055743-824202 Client’s Ref.: M23-284L-WO1-a was performed by headspace solid phase microextraction (HS-SPME) using a Gerstel® MPS Robotic Pro MultiPurpose autosampler directed by Maestro® software (Gerstel®, Inc., Linthicum, MD). Sample extraction and injection parameters are provided in Table 2 (Autosampler Method). Volatile metabolite analysis was performed by two-dimensional gas chromatography−time-of-flight mass spectrometry (GC×GC–TOFMS) using a LECO® Pegasus® 4D and Agilent® 7890B GC (LECO® Corp., St. Joseph, MI). Chromatographic, mass spectrometric, and peak detection parameters are provided in Table S0.1 (see GC×GC Method and Mass Spectrometry Method). An external alkane standards mixture (C8 – C20; Sigma- Aldrich®) was sampled multiple times for calculating retention indices (RI). The injection, chromatographic, and mass spectrometric methods for analyzing the alkane standards were the same as for the samples. Table 2: Parameters for HS-SPME and GC×GC-TOFMS analysis, and data processing and alignment AUTOSAMPLER METHOD Instrument description Gerstel® MPS Pro®)m 531 100243977.1Atty. Docket No.055743-824202 Client’s Ref.: M23-284L-WO1-a Carrier gas Helium, 2 mL∙min-1 (constant)Front inlet type Gerstel®F t i l t d S litl32 100243977.1Atty. Docket No.055743-824202 Client’s Ref.: M23-284L-WO1-a DATA PROCESSING METHOD Software descriptionLECO®ChromaTOF®and Statistical .1[01g y g p , p g, andalignment were performed using ChromaTOF software version 4.71 with the Statistical Compare package (LECO® Corp.), using the parameters listed in Table 2 (see Data Processing Method). Peaks were assigned a putative identification based on mass spectral similarity and retention index (RI) data, and the confidence of those identifications were indicated by assigning levels 1 to 4 (1 highest). Peaks at level 1 were identified based on mass spectral and RI matches with external standards or homologous series data. Peaks at level 2 were identified based on ≥ 800 mass spectral match by a forward and reverse search of the NIST 14 library and RIs thatconsistent (< 7% error) with a calculated 624 RI using the following equation: mean NIST nonpolar RI ^ 1.0517 + 14.468; this equation was obtained based on the linear relationship between RIs measured on 624Sil and nonpolar stationary phases. Level 1, 2, and 3 compounds were assigned to chemical functional groups based upon characteristic mass spectral fragmentation patterns and second dimension retention times. Level 4 compounds had mass spectral matches <800 or RI that do not match previously published values and were reported as unknowns.

[0126] Data Post Processing and Statistical Analyses: The data post processing steps aredepicted in FIG.1. Before statistical analyses, compounds eluting prior to 358s (acetone retention time) and siloxanes (i.e., chromatographic artifacts) were removed from the peak table. Missing values were handled using RepHM imputation in the R MetabImpute package version 0.1.0, as follows: peaks that were present in only one of the three technical replicates were imputed to zero 33 100243977.1Atty. Docket No.055743-824202 Client’s Ref.: M23-284L-WO1-a for that replicate, while missing values for peaks that were present in two out of three technical replicates were imputed to half of the minimum value within the replicates. The relative abundances of compounds across chromatograms were normalized using probabilistic quotient normalization (PQN) in R version 4.1.2 and the data were log10 transformed. Discriminatory VOCs were identified using 100 iterations of Random Forest (RF) with balanced class sizes in R randomForest package version 4.7.1, and visualized using the beeswarm package version 0.4.0. Sample sizes were balanced with replacement by down-sampling the majority class to have an equal number of samples as the minority class. RF model statistics were calculated using the Agresti-Coull interval in R epiR package version 2.0.63. Principal component analyses of the discriminatory VOCs were performed using prcomp in R stats package version 4.1.2 with the geometric means of the technical replicates as observations and the absolute peak abundance (mean-centered and scaled to unit variance) as variables.

[0127] Data availability: Metabolomic data (chemical feature peak areas and retention timeinformation) included in this study are available at the NIH Common Fund’s National Metabolomics Data Repository (NMDR) website, the Metabolomics Workbench, at www.metabolomicsworkbench.org, where it has been assigned project ID PR0001064 and study ID ST002449 (http: / / dx.doi.org / 10.21228 / M85H6W).Example 1: Human BALF Samples

[0128] Bronchoalveolar lavage fluid (BALF) specimens that had been collected from patientstreated at Mayo Clinic, Phoenix, Arizona presenting with symptoms consistent with community- acquired pneumonia were analyzed by headspace solid phase microextraction (HS-SPME) and two-dimensional gas chromatography coupled to time of-flight mass spectrometry (GC×GC- TOFMS) to characterize their volatilomes. Patient samples were categorized by infecting microbial organisms: Coccidioides, fungal, bacterial, viral, multiple pathogens, and negative (Table 1). After contaminant removal and data clean-up we detected 244 VOCs (Table 3A-3B), of which 69 were identified at a level 1 or 2 and were assigned putative compound names based on mass spectral and chromatographic data (Higgins et al. 2021. mSphere 6:e00040-21, Higgins et al. 2023. Journal of Fungi. 9(1)). Of these 69 named compounds, 19 have been previously associated with Coccidioides spp. either in vitro cultures or in a mouse lung infection model. For the unnamed compounds, 19 were identified at level 3 and assigned chemical classification based on a combination of mass spectral and chromatographic data (Table 3A-3B). Geometric mean of the log-transformed peak areas of three technical replicates with the Coccidioides, fungal, 34 100243977.1Atty. Docket No.055743-824202 Client’s Ref.: M23-284L-WO1-a bacterial, viral, multiple pathogens, and negative samples for the VOCs were calculated (Data not shown). Table 3A: Detected VOCs Molecular Chemical Functional d100243977.1Atty. Docket No.055743-824202 Client’s Ref.: M23-284L-WO1-a 4778 0.98 738 unknown88 C5H12O 2 788 1.20 743 alcohold100243977.1Atty. Docket No.055743-824202 Client’s Ref.: M23-284L-WO1-a 41129 0.94 890 unknown106 C8H10 2 1148 0.97 898 aromatic hydrocarbon37 100243977.1Atty. Docket No.055743-824202 Client’s Ref.: M23-284L-WO1-a 41420 0.91 1023 unknown4 1422 1.20 1024 unknown38 100243977.1Atty. Docket No.055743-824202 Client’s Ref.: M23-284L-WO1-a 184 C13H28 2 1855 0.73 1246 hydrocarbon100243977.1Atty. Docket No.055743-824202 Client’s Ref.: M23-284L-WO1-a 42410 0.64 1549 unknown206 C14H22O 2 2420 1.62 1553 aromatic alcoholMW: Nominal molecular weight in atomic mass units, based on compound identification ID Level: Confidence of the identification (level 1 high) based on the Chemical Analysis Working Group Metabolomics Standards Initiative criteria 1tR: Mean first dimension retention time 2tR: Mean second dimension retention time LRI: Linear retention index on a 624 column Table 3B: Number of human BALF samples per category with VOC present, after Rep Half- Min 40 100243977.1Atty. Docket No.055743-824202 Client’s Ref.: M23-284L-WO1-a Total Total Total Total Total Total VOC Compound ID Cocci Viral (out of Fungal Bacterial (out o Multi Negative (out of 1) (out of 19) f (out of 7) (out of 12)41 100243977.1Atty. Docket No.055743-824202 Client’s Ref.: M23-284L-WO1-a 86 2,3,4-trimethyl-pentane 2 0 5 1 4 4 88 2,3,3-trimethyl-pentane 3 0 9 1 4 6100243977.1Atty. Docket No.055743-824202 Client’s Ref.: M23-284L-WO1-a 147 3-heptanone 0 1 0 0 0 0 149 2-heptanone 11 1 17 2 6 123 100243977.1Atty. Docket No.055743-824202 Client’s Ref.: M23-284L-WO1-a 198 6-methyl-5-hepten-2-one 0 0 1 0 0 1 199 4,6-dimethyl-2-heptanone 12 1 18 3 6 1244 100243977.1Atty. Docket No.055743-824202 Client’s Ref.: M23-284L-WO1-a 270 unknown-270 12 1 19 3 7 12 271 unknown-271 11 1 18 4 7 1245 100243977.1Atty. Docket No.055743-824202 Client’s Ref.: M23-284L-WO1-a 421 unknown-421 1 0 2 1 0 1 422 unknown-422 4 0 5 0 0 2Example 2: Human BALF volatiles that discriminate Coccodioides from other infection etiologies

[0129] Several models were built to determine if VOCs in BALF can be used to classifyspecimens based on the presence vs. absence of infection and the infection etiology. To identify discriminatory VOCs we performed Random Forest (RF) comparing infected samples to negative samples, Coccidioides-positive samples to all other infected samples (i.e., non-Cocci infected), and Coccidioides-positive samples to bacteria-positive samples (Table 1). For the model discriminating Infected vs. Negative samples, a total of 43 subjects were identified as infected with at least 1 microbial organism, yielding 124 total infected samples, while 12 subjects were not infected with a microbial organism, yielding a total of 36 negative samples. RF was performed 46 100243977.1Atty. Docket No.055743-824202 Client’s Ref.: M23-284L-WO1-a with a sample size of 36 for both classes. Seven discriminatory VOCs were identified that separated infected samples from negative samples with a mean-out-of-bag error of 0.12. These seven VOCs were able to correctly classify 112 of the 124 infected samples as well as 30 of the 36 negative samples (Table 4; FIG. 2), yielding a sensitivity and specificity of 95% and 71%, respectively. All three technical replicates for two samples (B136 and B165) were misclassified as negative, and both samples were infected with non-tuberculous mycobacteria (NTM) (Table 1). Additionally, two out of three technical replicates of a third sample (M16) were misclassified as negative, and the patient was noted as having a history of tuberculosis with a possible active viral infection at the time of BALF collection. Two samples that misclassified as positive represent two out of three replicates for a specimen (N41) that contained clinical notes indicating the patient was Coccidioides positive; however, the corroborating lab results could not be located, so it was classified as negative for this study. The remaining misclassified samples, three as negative and three as infected represent a single technical replicate for each specimen, with the remaining replicates correctly classified. A principal component analysis (PCA) biplot of the seven discriminatory volatiles as variables and the human BALF samples as observations shows that five of the volatiles, 161, 215, 368, 377, and 515 are most strongly associated with the presence vs. absence of infection (FIG.3). Table 4: Model error rates, sensitivities, specificities, and areas under the receiver operator curves (AUROCs; (95% CI)) for volatile molecules from BALF as diagnostic for Infected versus Negative, Coccidioides versus Non-Cocci Infected, and Coccidioides versus Bacterial. Model Subjects Samples errorSensitivity Specificity AUROC))

[0130] In a model to identify VOCs that discriminate between Coccidioides-infected samples vs.Non-Cocci Infected BALF, a total of 12 subjects were identified as Coccidioides-positive yielding 33 total samples, while 28 subjects were infected with at least 1 microbial organism while also being Coccidioides-negative, yielding a total of 82 Non-Cocci Infected samples. Three samples 47 100243977.1Atty. Docket No.055743-824202 Client’s Ref.: M23-284L-WO1-a were excluded from this model, two of which, M10 and M49, were multiply infected with Coccidioides and bacteria and were therefore excluded from both classes. The third sample, M815b, was multiply infected with bacteria and virus at the time of BALF sampling; however, this subject had a recent history of Coccidioides and was therefore excluded from the model. RF was performed with a sample size of 33 for both classes. Five discriminatory VOCs were identified that separated Coccidioides samples from Non-Cocci Infected samples with a mean- out-of-bag error of 0.07. These five VOCs were able to correctly classify 29 of the 33 Coccidioides samples as well as 79 of the 82 infected samples (Table 4; FIG. 4) yielding a sensitivity and specificity of 90% and 95%, respectively. Each of the seven samples that were misclassified by RF represent a single technical replicate of a sample, while the remaining two replicates classified correctly. A PCA biplot of the five discriminatory volatiles as variables and the human BALF samples as observations showed that two volatiles, 352 and 374, were positively associated with Coccidioides-positive samples, while 255 was associated with non-Cocci Infected samples (FIG.ato identify VOCs that discriminate between Coccidioides vs. bacterialinfections, a total of 12 subjects were identified as Coccidioides-positive yielding 33 total samples, while 19 subjects were identified as bacterial, yielding a total of 56 bacterial samples. RF was performed with a sample size of 33 for both classes. Four discriminatory VOCs were identified that separated Coccidioides samples from bacterial samples with a mean-out-of-bag error of 0.05. These four VOCs were able to correctly classify 31 of the 33 Coccidioides samples as well as 54 of the 56 bacteria samples (Table 4; FIG. 6), yielding a sensitivity and specificity of 93% and 96%, respectively. Each of the four samples that were misclassified by RF represent a single technical replicate of a sample, while the remaining two replicates were correctly classified. A PCA biplot of the four discriminatory volatiles as variables and the human BALF samples as observations showed volatile 46 is positively associated with Coccidioides infection and 255 is, again, associated with non-Cocci Infected samples, this time specifically bacterial infections (FIG.7). Summary of Examples

[0132] The disclosed study identified VOCs that could distinguish Coccidioides pneumonia fromnon-Cocci pneumonia. HS-SPME and GC×GCTOFMS were used to examine the BALF from patients treated at Mayo Clinic, Phoenix, Arizona who presented with symptoms consistent with community-acquired pneumonia. 69 named volatile organic compounds were detected in the headspace of BALF. Four compounds were found in the in vitro cultures of at least one strain of Coccidioides, one of which, 2-butyl-1-octanol, was detected in the host specific (spherule) life 48 100243977.1Atty. Docket No.055743-824202 Client’s Ref.: M23-284L-WO1-a cycle of both C. immitis and C. posadasii for 11 of the 12 strains investigated. An additional 15 named compounds were identified in a Coccidioides mouse lung infection model, six of which (2- butanone, 2-methyl-1-propanol, 2-methyl-2-propanol, amylene hydrate, nonanal, and undecane) were found to be significantly correlated to immune response to Valley fever infection, as measured by cytokine production. While no compounds were identified in all three systems – in vitro cultures, mouse lung infection model and human BALF samples – that is not unexpected, as only one compound, decanal was identified in both the in vitro and murine model studies and was not detected in the human BALF samples. Further, many prior studies have shown that the VOCs of microbial cultures are rarely among the discriminatory in vivo or ex vivo volatile biomarkers for pneumonia detected in breath or other lung specimens. As a confirmation that relevant VOCs were detected from the ex vivo specimens, propofol (VOC 317; Table 3A) was identified in 53 of the 55 BALF samples, which is an anesthetic and sedative used during bronchoscopy.

[0133] When comparing infected versus negative classes, two infected samples (B136 and B165)completely misclassified as negative (Table 1). Sample B165 was categorized as a bacteria sample; however, it is infected with Mycobacterium kansasii, which is a slow-growing, non-tuberculous mycobacterium (NTM) that often presents as chronic pulmonary cavity disease. B136 was also categorized as a bacteria sample as the BALF for this patient was positive for Pseudomonas aeruginosa and Mycobacterium avium complex. Additionally, it was noted in the patient’s chart that they had indeterminate lung nodules with a history of abnormal CT chest scans, and a differential diagnosis would include a granulomatous infection. In this case, it may be incidental that P. aeruginosa was cultured out of the BALF. In these cases, the host may be responding differently to an NTM infection than it would another bacterial infection, leading to a different volatile profile. Two out of three replicates for a third sample, M16, also misclassified as negative, although it is categorized as infected with multiple organisms, bacterial (P. aeruginosa) and viral (human rhinovirus / enterovirus and human parainfluenza virus). Patient M16 also had a personal history of tuberculosis with no information regarding the current status of their tuberculous, but may be presenting with a different volatile profile due to the presence of tuberculosis. Patient N41 was categorized as negative; however, the patient’s chart noted “organisms consistent with Coccidioides sp; positive GMS”. GMS is a Grocott methenamine silver stain used for the identification of fungi on cytosmears and tissue sections. Unfortunately, the lab results were not associated with the patient’s chart, so for this study the patient was classified as negative. However, it is interesting that the patient’s volatile profile was classifying as Coccidioides infected, and thus may be corroborating the clinical suspicion of Coccidioides. 49 100243977.1Atty. Docket No.055743-824202 Client’s Ref.: M23-284L-WO1-a

[0134] One of two replicates for C61 and one of three replicates for C75 misclassified in both theCoccidioides versus non-Cocci infected and Coccidioides versus bacteria samples. C61 was Coccidioides positive by culture and had a complement fixation (CF) reading of 1:512; additionally, the patient’s chart notes that they suffer from Coccidioides meningitis. C75 was Coccidioides positive by enzyme immunoassay (EIA) for IgG with a CF reading of 1:64, the patient’s chart notes they suffer from chronic pulmonary Coccidioides infection. It is possible that Coccidioides meningitis and / or chronic Coccidioides infection would yield a different volatile profile than someone suffering from a strictly primary pulmonary Coccidioides infection. Thus, future development of Valley fever breath tests should investigate the sensitivity of volatile biomarkers in acutely versus chronically-infected patients, with the possibility that a separate suite of breath biomarkers may need to be identified to accurately diagnose the latter group.

[0135] Between the models, a total of 14 discriminatory VOCs were identified. Seven VOCs (161,199, 215, 254, 368, 377, and 515) were able to distinguish infected samples from non-infected samples with a model error rate of 12%, five of these [161, 215, 368, 377, and 515] were most strongly associated with the presence vs. absence of infection. Five VOCs were able to distinguish Coccidioides-positive samples from non-Cocci infected samples with a model error rate less than 7%, while four VOCs were able to distinguish Coccidioides infection from a bacterial infection with a model error rate less than 5%. Two VOCs are shared between these latter two models (150 and 255). It is likely that the high proportion of bacterial infections among the non-Cocci infected samples is driving the repeated selection of these two biomarkers during modeling.

[0136] The study disclosed herein demonstrated the development of VOC biomarkers for a Valleyfever breath test, with the strongest evidence supporting the development of a breath test that discriminates Valley fever from bacterial etiologies of pneumonia.

[0137] In summary, the BALF of patients with primary pulmonary Valley fever can bedistinguished from the BALF of patients with other forms of CAP using their volatile profiles. Seven VOCs were identified that could distinguish Coccidioides pneumonia from both non-Cocci infected and specifically bacterial pneumonia. The results disclosed herein showed that a breath test for Valley fever is a viable option, especially a test that discriminates Coccidioides or perhaps fungal infections from non-mycobacterial bacterial infections. 50 100243977.1

Claims

Atty. Docket No.055743-824202 Client’s Ref.: M23-284L-WO1-a CLAIMS What is claimed is:

1. A minimally invasive or non-invasive computer-implemented method for identifying apulmonary infection in a subject, the method comprising: (a) from a test sample from a lung of the subject, determining or having determined the relative abundance of each of a plurality of volatile organic compounds (VOCs) to prepare a sample VOC dataset; and (b) generating, by a processor, a machine learning prediction of a diagnosis selected from a pneumonia, a bacterial pneumonia, and a fungal pneumonia infection afflicting the subject using the sample VOC dataset as input to a machine learning model in view of a plurality of VOC abundances predetermined to be predictive of the pneumonia, bacterial pneumonia or fungal pneumonia.

2. The method of claim 1, wherein the machine learning model includes Random Forest(RF) classification and is trained, by the processor, by steps including: (a) accessing at least one sample VOC dataset and at least one VOC training dataset associated with the pneumonia, bacterial pneumonia or fungal pneumonia, wherein the VOC training dataset comprises the plurality of VOCs predetermined to be predictive of the pneumonia, bacterial pneumonia or fungal pneumonia; and (b) analyzing the relative abundance VOC of each sample VOC dataset; and (c) ranking an ability of each VOC to classify a sample for a diagnosis of pneumonia, bacterial pneumonia or fungal pneumonia.

3. The method of claim 2, wherein the training further comprises: (d) building a firstplurality of classifier trees; (e) selecting a first set of top VOCs from the first plurality of classifier trees; (f) building a second plurality of classifier trees using the set of top VOCs from (e); (g) selecting and saving a second set of top VOCs from the second plurality of classifier trees, the second set of top classifiers defining selected VOCs; and (h) performing multiple times to select the best set of top VOCs.

4. The method of claim 2, wherein the Random Forest classification includes building anensemble of a plurality of classifier trees, where the final prediction for a test sample is obtained by majority vote on the combination of predictions of all of the plurality of classifier trees. 51 100243977.1Atty. Docket No.055743-824202 Client’s Ref.: M23-284L-WO1-a5. The method of claim 1, wherein the fungal pneumonia is coccidioidomycosis, caused bythe fungus Coccidioides.

6. The method of claim 1 or 2, further comprising generating, by the processor, a pluralityof VOC predictors for pneumonia, bacterial pneumonia and fungal pneumonia.

7. The method of claim 6, wherein the plurality of VOC predictors comprises [161, 215,368, 377, and 515].

8. The method of claim 7, wherein the plurality of VOC predictors comprises [161, 199,215, 254, 368, 377 and 515].

9. The method of claim 8, wherein the VOC predictors correctly predict pneumoniainfection in the subject with an error rate of no more than 12%.

10. The method of claim 6, wherein the plurality of VOC predictors comprises [46, 150, 255,and 369].

11. The method of claim 6, wherein the plurality of VOC predictors comprises [150, 182,255, 352, 374].

12. The method of claim 11, wherein the VOC predictors correctly predict fungal pneumoniainfection in the subject with an error rate of no more than 7%.

13. A method for distinguishing a fungal pneumonia from a bacterial pneumonia, the methodcomprising: (a) from a test sample from a lung of the subject, determining or having determined the relative abundance of each of a plurality of volatile organic compounds (VOCs) to prepare a sample VOC dataset; and (b) generating, by a processor, a machine learning prediction of fungal pneumonia infection afflicting the subject using the sample VOC dataset as input to a machine learning model in view of a plurality of VOC abundances predetermined to be predictive of the fungal pneumonia.

14. The method of claim 13, wherein the machine learning model includes Random Forest(RF) classification and is trained, by the processor, by steps including: (a) accessing at 52 100243977.1Atty. Docket No.055743-824202 Client’s Ref.: M23-284L-WO1-a least one sample VOC dataset and at least one VOC training dataset associated with the fungal pneumonia, wherein the VOC training dataset comprises the plurality of VOC abundances predetermined to be predictive of the fungal pneumonia; and (b) analyzing the relative abundance of VOC of each sample VOC dataset; and (c) ranking an ability of each VOC to classify a sample for a diagnosis of fungal pneumonia.

15. The method of claim 14, wherein the training further comprises: (d) building a firstplurality of classifier trees; (e) selecting a first set of top VOCs from the first plurality of classifier trees; (f) building a second plurality of classifier trees using the set of top VOCs from (e); (g) selecting and saving a second set of top VOCs from the second plurality of classifier trees, the second set of top classifiers defining selected VOCs, and (h) performing multiple times to select the best set of top VOCs.

16. The method of claim 14, wherein the Random Forest classification includes building anensemble of a plurality of classifier trees, where the final prediction for a test sample is obtained by majority vote on the combination of predictions of all of the plurality of classifier trees.

17. The method of claim 13, wherein the fungal pneumonia is coccidioidomycosis, causedby the fungus Coccidioides.

18. The method of claim 13 or 14, further comprising generating, by the processor, aplurality of VOC predictors for fungal pneumonia.

19. The method of claim 18, wherein the plurality of VOC predictors comprises [150, 182,255, 352, 374], wherein the VOC predictors correctly predict fungal pneumonia infection in the subject with an error rate of no more than 7%.

20. The method of claim 19, wherein the plurality of VOC predictors further comprises [46and 369], wherein the VOC predictors correctly predict fungal pneumonia infection in the subject with an error rate of no more than 5%.

21. The method of any preceding claim, wherein the sample is a bronchoalveolar lavagesample, a sputum sample, or a breath sample. 53 100243977.1Atty. Docket No.055743-824202 Client’s Ref.: M23-284L-WO1-a22. The method of any preceding claim, wherein the relative abundance of each of pluralityof VOCs is determined using a mass spectrometer.

23. The method of claim 22, wherein the mass spectrometry comprises two-dimensional gaschromatography-time-of-flight mass spectrometry (GC×GC-TOFMS).

24. A method for treating a subject having or suspected of having a pneumonia, the methodcomprising: selecting the subject’s treatment from a fungal infection treatment and an antibiotic, by distinguishing or having distinguished a suspected pneumonia as a bacterial pneumonia or a fungal pneumonia by the method of any one of claims 1-4 or claims 13-17.

25. A sensor platform for identifying a pulmonary infection, distinguishing a fungalpneumonia from a bacterial pneumonia, or any combination thereof, in a subject comprising: (a) a receiver configured to receive a sample from a subject in need thereof; (b) a sensor configured to detect and / or assess relative abundance of one or more VOCs of any preceding claim ; (c) a processor configured to analyze, process and / or identify the one or more of the detected VOCs; (d) an output module configured to transmit a result generated by the processor.

26. The sensor platform of claim 25, wherein the processor comprises a machine learningalgorithm configured to analyze, process, and / or identify the VOCs contained in the sample.

27. The sensor platform of claim 25 or 26, wherein the sample is a bronchoalveolar lavagesample, a sputum sample, or a breath sample. 54 100243977.1

Images