Methods and compositions for detecting and characterizing drug toxicity in lung

Abstract

Disclosed are methods and systems for detecting lung injury, including interstitial lung disease, using lung tissue-specific methylation patterns in circulating cell-free DNA. The methods utilize differentially methylated genomic regions to distinguish lung-derived DNA signals and enable detection of treatment-induced toxicity. Applications include diagnostics, companion diagnostics, and drug safety monitoring.

Description

METHODSAND COMPOSITIONS FOR DETECTING AND CHARACTERIZING DRUG TOXICITY IN LUNG CROSS-REFERENCE TO RELATED APPLICATIONS

[0001] This application claims the benefit of U.S. Provisional Application No 63 / 791,026, filed April 18, 2025, which is incorporated herein by reference in its entirety.FIELD OF THE INVENTION

[0002] The present disclosure relates to molecular diagnostics, and more specifically to methods, systems, and compositions for detecting lung injury, interstitial lung disease (ILD), and treatment-induced pulmonary toxicity using tissue-specific methylation patterns in circulating cell-free DNA (cfDNA).BACKGROUND

[0003] Interstitial lung disease (ILD) includes a group of pulmonary disorders characterized by inflammation and fibrosis of lung tissue. Early detection remains challenging due to lack of sensitive, non-invasive biomarkers. It has been reported that that Interstitial lung disease (ILD) samples revealed higher un / methylation cfDNA from the lung-tissue specific methyl markers than healthy control. Certain therapeutic agents, including antibody-drug conjugates (ADCs) such as trastuzumab deruxtecan (Enhertu) and sacituzumab govitecan (Trodelvy), are associated with pulmonary toxicity including ILD and pneumonitis. Current monitoring approaches rely on imaging and clinical symptoms, which often detect disease only after significant progression.

[0004] Circulating cell-free DNA (cfDNA) has emerged as a promising biomarker source. However, existing approaches — such as cancer signal-of-origin (CSO) classifiers — are confounded by tumor-derived DNA and lack specificity for non-malignant lung injury.Indeed, such approaches may misclassify lung-derived cfDNA as cancer-associated, limiting their utility for ILD detection. Accordingly, there remains a need for improved methods that can sensitively and specifically detect lung injury, including treatment-induced ILD, using non-invasive biomarkers.

[0005] Described herein is a genomic and epigenomic methylation-based detection platform to identify ILD, other related fibrotic condition, lung damage. This includes methods and compositions for measuring methylation at lung tissue-specific regions, computing a methylation-based score or classification and identifying lung injury based on the score.These techniques support methods for monitoring drug-induced lung toxicity, methods for companion diagnostics for ADC therapies and systems and panels for detecting lung-derived cfDNA . This may use of at least 10-30 lung-specific DMRs.SUMMARY OF THE INVENTION

[0006] Described herein is a method for detecting lung injury in a subject, including: obtaining a biological sample including cell-free DNA; detection methylation at a plurality of genomic regions; generating a methylation profile; and classifying the subject as having lung injury based on the methylation profile. In other embodiments, the lung injury includes interstitial lung disease (ILD). In other embodiments, the lung injury is induced by a therapeutic agent. In other embodiments, the therapeutic agent is an antibody-drug conjugate. In other embodiments, the antibody-drug conjugate includes trastuzumab deruxtecan. In other embodiments, the method includes differentially methylated regions in lung tissue compared to non-lung tissues. In other embodiments, the differentially methylated regions in injured lung tissue compared to non-injured lung tissues. In other embodiments, the plurality of genomic regions includes at least 10 lung-specific DMRs. In other embodiments, the plurality includes 10, 20, 30, 40, 50, 60, 70, 80, 90 or more DMRS,. In other embodiments, the plurality includes between 10 and 30 DMRs, optionally 12- 25 DMRs. For example, genomic regions includes those exhibiting differential methylation specific to lung tissue, are hypermethylated or hypomethylated relative to non-lung tissues, and are detectable in circulating cfDNA. In some embodiments, this includes regions include CpG-dense loci with >8 CpGs, regions exhibit methylation differences >0.2 between lung and non-lung tissues and / or regions show low background methylation in blood cells In various embodiments, markers are selected using multi-tissue methylation datasets, including differential methylation analysis (e.g., Limma), filtering criteria including one or more of: FDR thresholds (e.g., 0.05), methylation thresholds (>0.6 vs <0.5), noise filtering and / or TCGA cancer filtering

[0007] In other embodiments, the method includes measuring methylation includes bisulfite sequencing. In other embodiments, the method includes measuring methylation includes a targeted sequencing panel. In other embodiments, the method includes measuring methylation includes a methylation-sensitive restriction enzyme assay. In other embodiments, classifying includes applying a machine learning model. In other embodiments, the model includes logistic regression. In other embodiments, the model distinguishes between high and low methylation clusters. In other embodiments, the plurality of genomic regions includesregions having at least 8 CpG sites. In other embodiments, the genomic regions exhibit a methylation difference of at least 0.2 between lung tissue and non-lung tissues. In other embodiments, the genomic regions exhibit a methylation level of at least 0.6 in lung tissue and at most 0.5 in non-lung tissues. In other embodiments, the genomic regions are selected to have background methylation levels of less than 5% in circulating DNA from healthy subjects. In other embodiments, the genomic regions are selected to exclude regions associated with methylation patterns of non-lung cancer tissues. In other embodiments, the biological sample includes plasma-derived cell-free DNA. In other embodiments, the method includes quantifying a total amount of cell-free DNA and incorporating the total amount into the classification. In other embodiments, the elevated total cell-free DNA level relative to a baseline is indicative of lung injury. In other embodiments, measuring methylation includes bisulfite conversion followed by next-generation sequencing. In other embodiments, measuring methylation includes amplification using primers specific to the plurality of genomic regions. In other embodiments, measuring methylation includes digestion using one or more methylation-sensitive restriction enzymes. In other embodiments, the genomic regions are selected to include at least one methylation-sensitive restriction enzyme cut site. In other embodiments, generating the methylation profile includes determining a peak count or signal intensity for each genomic region. In other embodiments, classifying includes assigning the subject to a high methylation cluster or a low methylation cluster. In other embodiments, the assignment to the high methylation cluster is indicative of lung injury. In other embodiments, the machine learning model is trained using cross-validation. In other embodiments, the machine learning model applies a classification threshold selected to achieve a specificity of at least 90%. In other embodiments, the increase in lung-derived methylation signal is detected relative to a pre-treatment baseline sample. In other embodiments, the method further includes determining a rate of change of methylation signal over time. In other embodiments, the lung injury is associated with progression toward interstitial lung disease. In other embodiments, the therapeutic agent is selected from an antibody-drug conjugate, a chemotherapeutic agent, or an immunotherapy agent. In other embodiments, the antibody-drug conjugate is selected from trastuzumab deruxtecan or sacituzumab govitecan. In other embodiments, the subject does not have lung cancer.

[0008]

[0009] Further described herein is a method for monitoring treatment-induced lung toxicity, including performing any of the aforementioned methods at multiple time points and identifying an increase in lung-derived methylation signal. Further described herein is amethod for ILD detection performing any of the aforementioned methods. Further described herein is a method for identification of patients with lung disease by performing any of the aforementioned methods. Further described herein is a method for drug safety monitoring by performing any of the aforementioned methods. Further described herein is a method for detection of treatment-induced lung injury by performing any of the aforementioned methods. Further described herein is a method for companion diagnostics, including selection and monitoring of patients receiving ADC therapies by performing any of the aforementioned methods.

[0010] Also described herein is a system including: a sequencing or methylation detection module; a processor configured to analyze methylation at lung-specific genomic regions; and a classifier configured to detect lung injury. The system of claim 16, wherein the processor is configured to perform clustering analysis using at least one of k-means clustering, Gaussian mixture modeling, principal component analysis, or UMAP. In other embodiments, the classifier is configured to output a probability score indicative of lung injury. In various embodiments, the system is configured to performed any of the aforementioned methods.

[0011] Also described herein is a panel including probes or primers targeting a plurality of genomic regions that are differentially methylated in lung tissue relative to non-lung tissues. In other embodiments, the plurality of genomic regions includes between 15 and 30 lung tissue-specific DMRs. In other embodiments, the panel is configured for use in a targeted sequencing assay or a methylation-sensitive restriction enzyme assay.

[0012] It will be understood that the foregoing description is illustrative and not limiting. Variations and modifications of the disclosed embodiments will be apparent to those of ordinary skill in the art and are intended to fall within the scope of the appended claims.BRIEF DESCRIPTION OF FIGURES

[0013] Figure 1. ILD samples after ADC treatment. Figures from Magenheim et al. (2022) outlines a set of 17 lung-specific methylation markers to differentiate lung disease from healthy plasma, including 12 alveolar-specific markers combined with 5 additional markers (3 bronchial and 2 common lung).

[0014] Figure 2. Increased prevalence of lung-associated cfDNAin treatment samples vs baseline Suspicious samples with lung disease were identified by using a genomic and epigenomic methylation-based detection platform. Prevalence was observed in treatment samples than in baseline. Baseline samples: 349 / 2742 (12.7%); Non-baseline samples:856 / 4878 (17.5%; Treatment samples: 453 / 2339 (19.4%). A higher percentage lung samples in treatment than baseline group indicates a potential treatment effect for lung damage.

[0015] Figure 3. 34 healthy-lung-tissue-specific DMRs were identified based on filtering rule described and panel detection overlap. 19 healthy-lung-tissue-specific DMRs can be used as candidate features for the ILD model development.

[0016] Figure 4. Clustering of samples using lung-specific methylation. Negative treatment samples in were analyzed. Clustering methods with linear algorithm such as Kmeans++, GMM and PCA ended up with similar classification of two groups. More samples were categorized into a high peak counter cluster group using UMAP based on 18 DMRs. Given the methylation level (peak count difference), two groups are named as High peak count cluster and Low peak count cluster, respectively. Two clusters can be distinguished well by these healthy-lung-tissue-specific regions.

[0017] Figure 5A., Figure 5B. Lung-tissue-specific regions associate with methyl score. Several regions showed strong association (Spearman p >=0.7) with sample level methylation score (LR breast) on clinical breast cancer samples with EnhertuTM treatment.

[0018] Figure 6. ILD prediction on clinical samples by lung-tissue DMRs. The methods and compositions described herein identified 18 regions that are healthy-lung-tissue specific hyper methylated DMRs. Without being bound by any particular theory, in non-cancerous lung diseases (ILD) situation, it’s possible that the methylation status of these regions may become more methylated. If so, the samples with high methylation level on these 18 healthy-lung-tissue specific DMRs might subject to the lung damage caused by treatment. Samples with high peak counts were identified for ILD samples potentially caused by ADC treatment (EnhertuTM). cfDNA showed significantly higher in predicted ILD samples. Predicted ILD showed distinguishable methylation pattern (heatmap). Significant elevated cfDNA and distinguishable methyl pattern by the lung-tissue DMRs showed good predictability with the aforementioned DMRs on clinical data.DETAILED DESCRIPTION

[0019] As described, the present disclosure provides a genomic and epigenomic detection platform based on lung tissue-specific methylation signatures in cfDNA. The methods and compositions described herein take advantage of observations and phenomena including lung injury and ILD are associated with increased cfDNA derived from lung tissue, lung-derived cfDNA exhibits distinct methylation patterns at lung-specific differentially methylatedregions (DMRs) and that these patterns can be used to distinguish lung injury from baseline states and from cancer-derived signals.

[0020] The present disclosure provides methods, systems, and compositions for detecting lung injury, including interstitial lung disease (ILD), based on tissue-specific epigenomic signatures present in circulating cell-free DNA (cfDNA). The invention is founded on the recognition that cfDNA released from lung tissue into circulation carries characteristic methylation patterns that can be distinguished from cfDNA originating from other tissues, including blood cells and tumor tissue. By leveraging these lung-specific methylation signatures, the disclosed methods enable sensitive and non-invasive detection of lung injury, including injury induced by therapeutic interventions.

[0021] In various embodiments, a biological sample is obtained from a subject. The biological sample may comprise any sample containing cell-free DNA, including but not limited to plasma, serum, whole blood, or other bodily fluids. In some embodiments, plasma is preferred due to its established utility in cfDNA-based assays. The cfDNA is then isolated or otherwise made available for downstream methylation analysis using methods known in the art.

[0022] The methods disclosed herein involve the measurement of DNA methylation at a plurality of genomic regions that are functionally defined as lung tissue-specific differentially methylated regions (DMRs). These regions are not limited to specific genomic coordinates but are instead defined by their functional properties. In particular, such regions exhibit differential methylation patterns in lung tissue relative to non-lung tissues, including blood-derived cells and other organ systems. In certain embodiments, the regions are characterized by relatively high methylation levels in lung tissue and comparatively lower methylation levels in non-lung tissues, or vice versa.

[0023] In an embodiment, the population of nucleic acids is one obtained from a serum, plasma or blood sample from a subject suspected of having neoplasia, a tumor, or cancer or previously diagnosed with neoplasia, a tumor, or cancer. The population of nucleic acids includes nucleic acids having varying levels of methylation. Methylation can occur from any one or more post-replication or transcriptional modifications. Post-replication modifications include modifications of the nucleotide cytosine, particularly at the 5-position of the nucleobase, e.g., 5-methylcytosine, 5-hydroxymethylcytosine, 5-formylcytosine and 5-carboxylcytosine. The affinity agents can be antibodies with the desired specificity, natural binding partners or variants thereof (Bock et al., Nat Biotech 28: 1106-1114 (2010); Song etal., Nat Biotech 29: 68-72 (2011)), or artificial peptides selected e.g., by phage display to have specificity to a given target.

[0024] Examples of capture moieties contemplated herein include methyl binding domain (MBDs) and methyl binding proteins (MBPs) as described herein, including proteins such as MeCP2 and antibodies preferentially binding to 5-methylcytosine. Likewise, partitioning of different forms of nucleic acids can be performed using histone binding proteins which can separate nucleic acids bound to histones from free or unbound nucleic acids. Examples of histone binding proteins that can be used in the methods disclosed herein include RBBP4, RbAp48 and SANT domain peptides. Although for some affinity agents and modifications, binding to the agent may occur in an essentially all or none manner depending on whether a nucleic acid bears a modification, the separation may be one of degree. In such instances, nucleic acids overrepresented in a modification bind to the agent at a greater extent that nucleic acids underrepresented in the modification. Alternatively, nucleic acids having modifications may bind in an all or nothing manner. But then, various levels of modifications may be sequentially eluted from the binding agent.

[0025] For example, in some embodiments, partitioning can be binary or based on degree / level of modifications. For example, all methylated fragments can be partitioned from unmethylated fragments using methyl-binding domain proteins (e.g., MethylMiner Methylated DNA Enrichment Kit (ThermoFisher Scientific)). Subsequently, additional partitioning may involve eluting fragments having different levels of methylation by adjusting the salt concentration in a solution with the methyl-binding domain and bound fragments. In some embodiments, partitioning of the nucleic acids is performed by contacting the nucleic acids with a methylation binding domain (“MBD”) of a methylation binding protein (“MBP”). MBD binds to 5-methylcytosine (5mC). MBD is coupled to paramagnetic beads, such as Dynabeads® M-280 Streptavidin via a biotin linker. Partitioning into fractions with different extents of methylation can be performed by eluting fractions by increasing the NaCl concentration.

[0026] An exemplary method for molecular tag identification of MBD-bead partitioned libraries through NGS is as follows:

[0027] Physical partitioning of an extracted DNA sample (e.g., extracted blood plasma DNA from a human sample) using a methyl-binding domain protein-bead purification kit, saving all elutions from process for downstream processing.

[0028] Parallel application of differential molecular tags and NGS-enabling adapter sequences to each partition. For example, the hypermethylated, residual methylation ('wash'), and hypomethylated partitions are ligated with NGS-adapters with molecular tags.

[0029] Re-combining all molecular tagged partitions, and subsequent amplification using adapter-specific DNA primer sequences.

[0030] Enrichment / hybridization of re-combined and amplified total library, targeting genomic regions of interest (e.g., cancer-specific genetic variants and differentially methylated regions).

[0031] Re-amplification of the enriched total DNA library, appending a sample tag. Different samples are pooled and assayed in multiplex on an NGS instrument.

[0032] Bioinformatics analysis of NGS data, with the molecular tags being used to identify unique molecules, as well deconvolution of the sample into molecules that were differentially MBD-partitioned. This analysis can yield information on relative 5-methylcytosine for genomic regions, concurrent with standard genetic sequencing / variant detection.

[0033] Examples of MBPs contemplated herein include, but are not limited to:(a) MeCP2 is a protein preferentially binding to 5-methyl-cytosine over unmodified cytosine.(b) RPL26, PRP8 and the DNA mismatch repair protein MHS6 preferentially bind to 5- hydroxymethyl-cytosine over unmodified cytosine.(c) FOXK1, FOXK2, FOXP1, FOXP4 and FOXI3 preferably bind to 5-formyl- cytosine over unmodified cytosine (lurlaro et al., Genome Biol. 14: R119 (2013)). (d) Antibodies specific to one or more methylated nucleotide bases.

[0034] In some embodiments, the lung-specific DMRs are identified through a multi-step selection process. This process may include analyzing methylation data across multiple tissue types, including lung tissue and non-lung tissues such as blood cells, to identify candidate regions exhibiting tissue-specific methylation patterns. Statistical methods, including differential methylation analysis using false discovery rate thresholds and minimum methylation differences, may be employed to identify regions that are significantly differentially methylated. Additional filtering steps may be applied to remove regions with high background methylation in circulating DNA or regions associated with unrelated cancer types, for example by using reference datasets such as TCGA. In certain embodiments, regions are selected that contain a minimum number of CpG sites and satisfy predefined methylation thresholds in target versus non-target tissues.

[0035] Using such approaches, a set of lung-specific DMRs may be identified and used as features for downstream analysis. In some embodiments, at least 10 such regions are used. In other embodiments, the number of regions may range from approximately 15 to 30. In certain specific embodiments, a subset of approximately 19 lung-specific DMRs is used as a feature set for detecting lung injury. These regions may include hypermethylated regions associated with lung tissue, which have been observed to exhibit increased methylation signals in subjects with lung injury.

[0036] Once methylation levels at the selected regions are measured, a methylation profile is generated for the sample. Measurement of methylation may be performed using a variety of platforms. In some embodiments, bisulfite sequencing is employed, including whole-genome bisulfite sequencing or targeted bisulfite sequencing approaches. In other embodiments, targeted sequencing panels are used to selectively interrogate the regions of interest. In still other embodiments, methylation-sensitive restriction enzyme (MSRE)-based assays are employed, in which enzymatic digestion patterns are used to infer methylation states at specific genomic loci. The invention is not limited to any particular methylation detection platform, and any suitable method capable of distinguishing methylated from unmethylated DNA may be used.

[0037] The resulting methylation profile may then be analyzed to determine whether the subject exhibits lung injury. In some embodiments, the analysis involves unsupervised learning approaches. For example, clustering algorithms such as k-means clustering, Gaussian mixture models (GMM), principal component analysis (PCA), or uniform manifold approximation and projection (UMAP) may be used to group samples based on similarities in methylation patterns. In certain embodiments, samples segregate into distinct clusters corresponding to high and low methylation states across the lung-specific DMRs. Samples exhibiting higher methylation signals at these regions have been observed to correspond to increased levels of lung-derived cfDNA and are associated with lung injury.

[0038] In other embodiments, supervised machine learning models are used to classify samples. Such models may include, but are not limited to, logistic regression models, support vector machines, random forests, or neural networks. In some embodiments, a logistic regression model is trained using labeled samples to distinguish between subjects with and without lung injury. The model may be trained using cross-validation techniques, such as 10-fold cross-validation, to ensure robustness and generalizability. Performance metrics including sensitivity, specificity, and area under the receiver operating characteristic curve (AUC) may be used to evaluate the model. In certain embodiments, models trained using thedisclosed lung-specific DMR features achieve high performance, including AUC values of approximately 0.96 and sensitivity and specificity values exceeding 90%.

[0039] In addition to methylation features, other parameters may be incorporated into the analysis. For example, total cfDNA concentration or yield may be used as an additional feature, as increased cfDNA levels have been observed in subjects with lung injury. In certain embodiments, subjects classified as having lung injury exhibit both elevated methylation signals at lung-specific DMRs and increased cfDNA levels relative to baseline or control subjects.

[0040] The methods described herein are particularly useful for detecting treatment-induced lung injury. Certain therapeutic agents, including antibody-drug conjugates (ADCs), are known to cause pulmonary toxicity, including ILD and pneumonitis. The disclosed methods enable monitoring of such toxicity by analyzing changes in lung-derived methylation signals over time. In some embodiments, samples are collected from a subject at multiple time points, including prior to treatment (baseline) and during or after treatment. An increase in lung-specific methylation signal relative to baseline may indicate the onset or progression of lung injury.

[0041] In some embodiments, the methods are used as companion diagnostics to guide therapeutic decision-making. For example, patients receiving ADC therapies may be monitored for early signs of lung toxicity, enabling timely intervention, dose adjustment, or discontinuation of treatment. In other embodiments, the methods may be used more broadly for detecting non-treatment-related lung diseases, including fibrotic conditions and other forms of pulmonary injury. Examples of issues with particular ADCs include those listed in Table 1.Table 1. ADC related lung conditionsADC Treatment Descriptionfam-trastuzumab ADC treatment ENHERTU™ can cause Lung problems, deruxtecan-nxki like interstitial lung disease / pneumonitissacituzumab ADC treatment Trodelvy™ ILD / pneumonitis aregovitecan-hziy possible adverse reactionsAZD0901 ADC treatment AZD0901. No reported lung-damagerelated adverse effect of AZD0901

[0042] The present disclosure also contemplates systems and compositions for implementing the described methods. In some embodiments, a system is provided that includes a methylation detection module, such as a sequencing instrument or MSRE-based assay system, and a computational module configured to analyze methylation data and generate a classification output. In certain embodiments, the system includes a panel of probes or primers designed to target the lung-specific DMRs described herein. Such panels may be configured for use in targeted sequencing or other methylation detection assays.

[0043] Importantly, the disclosed approach overcomes limitations associated with prior methods that rely on cancer-derived signals, such as cancer signal-of-origin classifiers. Such methods have been shown to produce false-positive lung cancer signals in the context of lung injury, thereby confounding interpretation. By focusing on tissue-specific methylation signatures rather than tumor-associated signals, the present invention provides improved specificity for detecting non-malignant lung injury.

[0044] While the invention has been described with respect to lung tissue, the principles described herein may be extended to other tissues and organs. In general, tissue-specific methylation signatures in cfDNAmay be used to detect injury or disease in a variety of organ systems. Accordingly, the disclosed methods may be adapted to detect tissue injury in organs other than the lung by selecting appropriate tissue-specific DMRs.

[0045]

[0046] Described herein are methods and compositions are described herein, where data associated with a plurality of subjects are obtained from one or more sources. Such data may include clinical information, such as treatment history, therapeutic regimens, disease stage, and demographic variables. In addition, genomic data may be obtained from sequencing assays and may include, for example, somatic mutations, copy number variations, structural variants, or other genomic features. Epigenomic data may also be obtained, including measurements of DNA methylation across genomic regions. These data may be derived from tissue samples, blood samples, or other biological specimens, and may be generated using any suitable assay or sequencing technology.

[0047] Further described is a scalable computational framework for integrating multi-modal biological data to identify disease subtypes associated with clinically meaningful outcomes.In various embodiments, the disclosed methods integrate clinical, genomic, and epigenomic data to generate composite representations that capture underlying biological heterogeneity.

[0048] In some embodiments, data are obtained from one or more sources including clinical records (e.g., treatment history, demographics), genomic sequencing data (e.g., somatic mutations, copy number alterations), and / or epigenomic data (e.g., DNA methylation profiles). In various embodiments, the data may be derived from tissue samples, blood samples, or other biological specimens.

[0049] Following acquisition, the data are processed to generate features suitable for downstream analysis. Clinical variables may be encoded as categorical or continuous variables, while genomic data may be represented as binary indicators reflecting the presence or absence of alterations in particular genes or genomic regions. Epigenomic data, which may comprise high-dimensional measurements, are transformed to generate derived features that capture latent structure within the data. In some embodiments, dimensionality reduction techniques such as principal component analysis are applied to methylation data to generate a set of principal components that summarize variation across samples. In other embodiments, alternative techniques such as matrix factorization methods or machine learning models are employed to generate latent representations.

[0050] Further described herein are clustering algorithms that serve to capture distinct aspects of structure within the data. Accordingly, in certain embodiments, multiple clustering methods are applied to the generated features. These methods may include partition-based approaches, hierarchical clustering techniques, density -based clustering, and probabilistic mixture models. Each clustering method produces a candidate assignment of subjects into clusters. Various unsupervised learning techniques may be employed to partition a set of subjects into subgroups, including, for example, partition-based algorithms, hierarchical clustering approaches using different linkage criteria, density-based methods, probabilistic or statistical mixture models, or other machine learning-based clustering techniques. In some embodiments, each technique independently generates a candidate assignment of subjects to one or more clusters based on underlying feature representations. These candidate assignments may be generated using different parameterizations, initialization schemes, distance metrics, or feature subsets, thereby providing multiple alternative views of the underlying data structure. In certain implementations, the plurality of candidate clusterings may be further compared, aggregated, or refined to produce a consensus or optimized clustering, although in other embodiments individual clustering outputs may be selected or weighted based on predefined criteria or performance metrics. These assignments are thencombined to generate a consensus clustering, for example by constructing a co-association matrix that reflects the frequency with which pairs of subjects are assigned to the same cluster across methods. As illustrated in Figure 2, such consensus approaches enable assessment of clustering stability and identification of robust subgroup structures.

[0051] For example, a stability representation may be generated to evaluate the consistency of subject clustering across multiple clustering approaches. In certain embodiments, clustering assignments for a plurality of subjects are obtained using two or more distinct clustering algorithms, which may include, for example, partition-based methods, hierarchical clustering methods employing different linkage criteria, density -based clustering techniques, probabilistic mixture models, or other unsupervised learning approaches.

[0052] A consensus matrix or similar structure may then be constructed to quantify agreement between clustering results. In some implementations, each entry of the matrix corresponds to a pairwise comparison between subjects and represents a measure of coassignment, such as the frequency, proportion, probability, or weighted score indicating how often the subjects are assigned to the same cluster across the plurality of clustering methods or across multiple runs of a given method. The values may be normalized to a defined range, for example between zero and one, or expressed using alternative scales or similarity metrics.

[0053] The consensus matrix may be visualized using a graphical representation, such as a heatmap or other visual encoding, in which varying intensities, colors, or patterns reflect levels of agreement. In certain embodiments, the presence of well-defined block structures within the matrix may indicate stable subgrouping of subjects, whereas diffuse or non-distinct patterns may indicate reduced clustering stability, ambiguity in subgroup boundaries, or heterogeneity within the dataset.

[0054] In some embodiments, measures of clustering stability may be further quantified using statistical or computational metrics, including but not limited to consensus scores, silhouette scores, variation of information, or resampling-based stability indices. Such analyses may be used to select among clustering methods, determine an optimal number of clusters, assess robustness of identified subgroups, or guide downstream analyses.

[0055] In alternative implementations, consensus or stability assessments may be performed across different feature sets, data modalities, or preprocessing strategies, thereby enabling evaluation of the sensitivity of clustering results to variations in input data or modeling assumptions. Additionally, stability information may be incorporated into downstream processes, such as weighting cluster assignments, filtering unreliable subgroupings, or informing confidence scores associated with subject classification.

[0056]

[0057] Once cluster assignments are established, the resulting subgroups are evaluated with respect to clinical outcomes. In certain embodiments, survival analysis techniques are applied to assess differences in outcomes between clusters. For example, Kaplan-Meier survival curves may be generated to visualize differences in survival probabilities over time, as shown in Figure 4. Statistical tests such as log-rank tests may be used to determine whether observed differences are significant. In addition, regression-based survival models, such as Cox proportional hazards models, may be fitted to quantify the contribution of individual features to risk.

[0058] Through such modeling, specific genomic and epigenomic features may be identified as prognostic biomarkers. For example, mutations in certain genes may be associated with increased risk, while derived epigenomic features may be associated with protective effects. In particular embodiments, a latent epigenomic feature derived from methylation data may correspond to coordinated methylation changes in genes involved in developmental or differentiation pathways, thereby reflecting underlying biological mechanisms that influence disease progression. These findings demonstrate that integration of epigenomic features provides additional prognostic information beyond genomic data alone.

[0059] The disclosed methods may be applied in a variety of clinical contexts. In some embodiments, a subject’s data are processed according to the methods described herein, and the subject is assigned to a cluster corresponding to a disease subtype. This subtype assignment may then be used to inform prognosis, guide treatment selection, or identify eligibility for particular therapies. For example, subjects assigned to high-risk clusters may be directed toward more aggressive treatment regimens, while subjects in lower-risk clusters may receive less intensive therapies.

[0060] The methods described herein may be implemented using one or more computing systems. Such systems may include processors configured to execute instructions stored in memory, as well as interfaces for receiving input data and providing output results. The systems may be deployed in clinical laboratories, research environments, or cloud-based platforms, and may be integrated with electronic health record systems or laboratory information systems.

[0061] Although particular embodiments have been described, it will be appreciated that numerous variations and modifications may be made without departing from the scope of the present disclosure. For example, additional data modalities such as transcriptomic, proteomic, or metabolomic data may be incorporated into the analysis. Similarly, alternative featureengineering, clustering, or modeling techniques may be employed. The disclosed framework is therefore broadly applicable to a wide range of diseases and analytical settings.

[0062] More information is found in PCT Pub. No. WO 2017062867, 2023 / 034453, each of which are fully incorporated by reference herein.

[0063]

[0064] Further described herein are methods and compositions related to feature engineering based on the detection platform and data analysis processes. In various embodiments, features may include: clinical variables (e.g., treatment lines, regimens), genomic indicators (e.g., binary mutation presence for selected genes), and / or epigenomic measurements (e.g., methylation levels across genomic regions).

[0065] In certain embodiments, epigenomic data are transformed into derived features using dimensionality reduction techniques, including: Principal component analysis (PCA), Nonnegative matrix factorization (NMF), and / or autoencoders or other machine learning models. In various embodiments, these derived features capture latent biological signals not readily observable from raw measurements. In various embodiments, the methods and compositions described herein are utilized with any cancer type, non-cancer diseases (e.g., autoimmune, neurological). In various embodiments, additional data modalities may include: transcriptomics, proteomics, and / or metabolomics.

[0066] Feature generation, clustering, and modeling techniques may be substituted with equivalent methods without departing from the scope of the invention. For example, when analyzing genomic regions in an individual sample, a computing system can determine a number of sequence representations that correspond to polynucleotide molecules having at least the threshold number of methylated cytosines in the genomic region. The computing system can perform one or more normalization operations using the counts of polynucleotide molecules or sequence reads that correspond to the genomic region and have at least the threshold number of methylated cytosines to generate normalized metrics. To illustrate, the computing system can divide the counts of polynucleotide molecules or reads that correspond to the genomic region and have at least the threshold number of methylated cytosines by the number of molecules or sequencing reads that correspond to a control region, such as a positive control region. In another instance, the computing system can perform the normalized metrics by dividing the counts of polynucleotide molecules or reads that correspond to the genomic region and have at least the threshold number of methylated cytosines by the number of molecules or sequencing reads in a control dataset (i.e., thecontrol dataset includes of tumor not-detected samples) corresponding to the same genomic region and have at least the same threshold number of methylated cytosines.

[0067] The normalized metrics can be analyzed with respect to a threshold value. The threshold value can correspond to a given genomic region, such as a given promoter region. In various examples, the threshold value can be different for different promoter regions. In these scenarios, a first promoter region can have a first threshold value and a second promoter region can have a second threshold value. In situations where the normalization metric is at least the threshold value, the computing system can determine that the genomic region has a first methylation status. In scenarios where the normalization metric is less than the threshold value, the computing system can determine that the genomic region has a second methylation status. In one or more illustrative examples, the first methylation status can be labeled as “methylated” and the second methylation status can be labeled as “not methylated.”

[0068] The threshold value for a given genomic region can be determined based on training data obtained from samples of individuals in which cancer is not detected. In one or more examples, sequence representations obtained from the training samples can be analyzed to determine a z-score with respect to the number of polynucleotide molecules that correspond to the genomic region and that have at least the threshold amount of methylated cytosines. In one or more illustrative examples, the threshold value for a promoter region that is used to determine the normalization metrics for the promoter region can be derived from the z-score calculated based on the training samples with respect to the promoter region.

[0069] In various embodiments, sequencing data can be analyzed by the computing system to determine indicators of the presence of cancer without training specific models. In one or more examples, the computing system can determine a tumor fraction value based on sequencing data generated from one or more samples obtained from a single subject in which it is unknown whether or not cancer is present in the subject. In various examples, the plurality of samples obtained from a given subject can be produced by performing a number of titrations on a single sample. In one or more additional examples, a first sample can be obtained from a subject prior to or at onset of at least one administration of a treatment or a procedure related to cancer and one or more second samples can be obtained from the subject after at least one of administration of a treatment or a procedure related to cancer. In one or more illustrative examples, the one or more second samples can be obtained at least one week, at least two weeks, at least three weeks, at least four weeks, at least five weeks, at least six weeks, at least eight weeks, or at least ten weeks after administration of the treatment orprocedure. In at least some examples, first sample and the second sample can be derived from at least one of a bodily fluid obtained from the subject or tissue obtained from the subject.

[0070] In one or more examples, one or more samples can be obtained from a given subject. The sequencing data generated from the one or more samples can be analyzed by the computing system to determine quantitative measures for a number of classification regions. In one or more examples, the quantitative measures can correspond to an amount of sequence representations that have at least a threshold amount of overlap with one or more classification regions. In one or more additional examples, the quantitative measures can correspond to sequence representations having at least a threshold amount of methylated cytosines in CpG regions having at least a threshold amount of CG content. In various examples, the indication of cancer status in the subject can include tumor fraction. In one or more additional examples, the indication of cancer status in the subject can include mutant allele fraction. In at least some examples, the quantitative measures can correspond to a number of sequencing reads that correspond to a given classification region in relation to a total number of sequencing reads across a plurality of positive control regions. In one or more further examples, the indicators of cancer status can be used to determine an output that corresponds to cancer status or not being present in a given individual in response to analyzing the one or more indicators of cancer status with respect to one or more thresholds. In one or more illustrative examples, tumor fraction determined from one or more samples obtained from a subject can be analyzed with respect to one or more thresholds. In instances where tumor fraction is greater than a threshold level, the computing system can determine that the probability of cancer status in the subject is at least 80%, at least 85%, at least 90%, at least 95%, or at least 99%. Further, in situations where multiple samples are obtained from a subject, first quantitative measures generated from a first sample obtained from the subject can be analyzed with respect to second quantitative measures generated from a second sample obtained from the subject. In at least some examples, differences between the first quantitative measures and the second quantitative measures can be analyzed to determine an indication of treatment response in the subject.

[0071] The quantitative measures used to determine an indication of cancer status in a subject can be determined by analyzing quantitative measures of a subset of classification regions. In at least some examples, the subset of classification regions can be different for different subjects. In one or more illustrative examples, values of quantitative measures for a number of classification regions can be analyzed with respect to one another and ranked according to the magnitude of the value of the quantitative measures. In various examples, theclassification regions for a given sample can be ranked in descending order from the one or more classification regions having the greatest value of a quantitative measure to the one or more classification region having the least value of the quantitative measure.

[0072] In various examples, after ranking the quantitative measures of the classification regions, quantitative measures that correspond to a group of the classification regions can be removed before determining the indication of cancer status in the subject. For example, the group of classification regions that are not used to determine the indication of cancer status in the subject can include the 1% of classification regions having the greatest quantitative measure values, the 2% of classification regions having the greatest quantitative measure values, 3% of classification regions having the greatest quantitative measure values, 4% of classification regions having the greatest quantitative measure values, 5% of classification regions having the greatest quantitative measure values, or the 6% of classification regions having the greatest quantitative measure values. In at least some examples, a number of classification regions having relatively high quantitative measure values can be excluded from the group of classification regions used to determine the indication of cancer status in the subject because, in at least some cases, classification regions corresponding to quantitative measure values at or near the top of the ranked list can have non-tumor origins and / or be related to sequencing artifacts. Thus, by removing the quantitative measures that correspond to these classification regions from the analysis used to determine the indication of cancer status in the subject, the accuracy with which the indication of cancer status in the subject can increase.

[0073] In one or more examples, after determining the group of classification regions to be used to determine an indication of cancer status in the subject, a subset of classification regions of the group can then be determined by identifying at least 10 classification regions of the group, at least 25 classification regions of the group, at least 50 classification regions of the group, at least 75 classification regions of the group, at least 100 classification regions of the group, at least 150 classification regions of the group, at least 200 classification regions of the group, at least 250 classification regions of the group, at least 300 classification regions of the group, at least 350 classification regions of the group, at least 400 classification regions of the group, at least 450 classification regions of the group, or at least 500 classification regions of the group having the greatest values for the respective quantitative measure.

[0074] In at least some examples, one or more statistical measures, such as at least one of mean, median, or mode, can be applied to the quantitative measures of the subset of the classification regions of the group to generate an initial indication of cancer status in thesubject. In various examples, the initial indication of cancer can be modified according to a scaling factor. The scaling factor can be applied to the initial indication of cancer status in the subject because, in at least some scenarios, the positive control regions can have different amounts of methylated CpGs. For example, at least a portion of the positive control regions can have fully methylated CpGs while other positive control regions may not be fully methylated. Additionally, in various situations, some classification regions can correspond to a high value of an indication of cancer status in subjects, such as 90% tumor fraction, 95% tumor fraction, 99% tumor fraction, or 100% tumor fraction, but nucleic acid molecules that correspond to these classification regions may not be fully methylated. To account for these cases, the scaling factor can be applied to the initial indication of cancer status in the subject to provide a more accurate determination of the indication. In one or more illustrative examples, the scaling factor can be determined by analyzing indications of cancer status in subjects determined using one or more techniques described herein in relation to additional data that corresponds to additional indications of cancer status in subjects, such as validation data or other techniques that generate data orthogonal to the indications of tumors being present in subjects described herein.

[0075] In various examples, the classification regions used to determine the quantitative measures can correspond to classification regions that correspond to one or more portions of differentially methylated regions. In one or more examples, the differentially methylated regions can include promoter regions that correspond to one or more classifications of cancer. For example, the classification regions can be determined by analyzing a number of sequencing representations across a differentially methylated region. In these scenarios, one or more portions of the differentially methylated regions that overlap with at least at threshold number of sequencing representations can be included in the classification regions. In one or more examples, the quantitative measures of the one or more portions of the differentially methylated regions can be determined based on the molecule count distribution of the differentially methylated region. For example, the quantitative measures can be determined based on the molecule count within one or more peaks of the molecule distribution of the differentially methylated region. To illustrate, in various examples, the distribution of molecules across a differentially methylated region can indicate one or more peaks where greater amounts of molecules overlap with one or more subregions within the differentially methylated region. In various examples, the one or more genomic regions that correspond to the one or more subregions of the differentially methylated regions that correspond to the highest amounts of sequence representations for a sample can be defined as classificationregions. In at least some examples, the distribution of sequence representations can have a peak that corresponds to a subregion of the differentially methylated region having a higher number of sequence representations than other subregions of the differentially methylated region. In these scenarios, the subregion can be identified as a classification region. By determining subregions of at least a portion of the differentially methylated regions used to determine the indication of cancer status in the subject, the amount of computing resources and memory resources used to determine the indication of cancer status in the subject can be decreased.

[0076] To illustrate, a classification region can include one or more portions of a differentially methylated region in which at least 50% of the sequencing representations obtained from a sample overlap, at least 55% of the sequencing representations obtained from a sample overlap, at least 60% of the sequencing representations obtained from a sample overlap, at least 65% of the sequencing representations obtained from a sample overlap, at least 70% of the sequencing representations obtained from a sample overlap, at least 75% of the sequencing representations obtained from a sample overlap, at least 80% of the sequencing representations obtained from a sample overlap, at least 85% of the sequencing representations obtained from a sample overlap, at least 90% of the sequencing representations obtained from a sample overlap, at least 95% of the sequencing representations obtained from a sample overlap, or at least 99% of the sequencing representations obtained from a sample overlap. In one or more illustrative examples, the one or more portions of the differentially methylated region that comprise a classification region can be contiguous with respect to a reference sequence. Further information is found in PCT Pub. No. WO 2024 / 211717, which is fully incorporated by reference herein.

[0077] In various embodiments, the method includes clustering and subtype identification. In some embodiments, multiple clustering algorithms are applied, including. K-means clustering, hierarchical clustering (e.g., Ward linkage, average linkage), density -based clustering (e.g., DBSCAN, HDBSCAN), Gaussian mixture models

[0078] In various embodiments, cluster assignments from multiple methods are combined to generate a consensus clustering, for example using: co-association matrices, voting schemes and / or stability analysis. Consensus clustering may reveal stable or unstable subgroup structures.

[0079] In various embodiments, methods and compositions described herein are utilized to determination association with clinical outcomes, subjects assigned to clusters are evaluated with respect to clinical outcomes, including: overall survival, progression-free survival,treatment response. In various embodiments, statistical methods may include: Kaplan-Meier survival analysis, Log-rank tests and / or Cox proportional hazards modeling

[0080] In various embodiments, clusters exhibiting statistically significant differences in outcomes may define clinically relevant subtypes. Further information is found in PCT Pub. No. WO 2025 / 019254 and 2025019297, each of which are fully incorporated by reference herein.

[0081] Described herein are methods and compositions related to biomarker identification. In various embodiments, the method includes determination of feature importance may be provided using survival models such as penalized Cox regression. In certain embodiments, specific genomic and epigenomic features are associated with increased or decreased risk. In various embodiments, certain epigenomic-derived features may be associated with reduced risk and correspond to methylation patterns affecting developmental transcription factors. In various embodiments, these features may be used as biomarkers for prognosis or therapeutic targeting.

[0082] Described herein are methods and compositions for clinical applications, which may include one or more of the following: Stratify patients into risk groups, predict disease progression, guide treatment selection, identify candidates for targeted therapies, and / or develop companion diagnostics. In various embodiments, are methods may be implemented on one or more computing systems including, one or more processors, memory storing executable instructions, interfaces for receiving biological and clinical data.EXAMPLESExample 1 - Identification of Lung Tissue-Specific Differentially Methylated Regions (DMRs)

[0083] In this example, lung tissue-specific differentially methylated regions (DMRs) were identified using a multi-stage bioinformatic and statistical filtering pipeline designed to enrich for regions exhibiting high specificity for lung tissue relative to other tissue types.

[0084] Input data comprised genome-wide methylation profiles from a plurality of human tissues, including at least 24 distinct cell and tissue types, of which at least two were lung-related cell types (e.g., alveolar and bronchial cells). Additional reference datasets included methylation profiles from blood-derived cell populations and publicly available epigenomic resources. Candidate regions were initially defined as genomic loci containing a minimum number of CpG dinucleotides, for example at least eight CpG sites per region, to ensure robustness of methylation signal measurement.

[0085] Differential methylation analysis was performed across tissue types using statistical frameworks such as linear modeling (e.g., Limma), applying thresholds including a false discovery rate (FDR) of 0.05 or lower and a minimum methylation difference (delta) of approximately 0.2 between lung tissue and non-lung tissues. Candidate regions were further filtered to retain those exhibiting high methylation levels (e.g., >0.6) in lung tissue and comparatively lower methylation levels (e.g., <0.5) in non-lung tissues, or vice versa depending on the directionality of the signal.

[0086] Additional filtering steps were applied to improve specificity. These included removal of regions with high background methylation in blood-derived DNA, exclusion of regions lacking sufficient methylation-sensitive restriction enzyme (MSRE) cut sites (for compatibility with certain assay formats), and elimination of regions with excessive variability or noise (e.g., greater than approximately 5% background signal). Regions were also evaluated against cancer methylation datasets, such as TCGA, to remove loci that exhibited elevated methylation in unrelated cancer types, thereby reducing the likelihood of confounding tumor-derived signals.

[0087] Following this multi-stage filtering process, an initial set of approximately 34 candidate lung-specific DMRs was identified. From this set, a subset of approximately 19 regions was selected based on criteria including reproducibility, signal strength, and compatibility with downstream detection platforms. These regions were designated as lung tissue-specific features suitable for use in ILD detection models.

[0088] In certain embodiments, the identified DMRs may be further refined or expanded based on additional datasets or clinical validation. In other embodiments, subsets of these regions may be selected to optimize assay performance for particular applications, such as targeted sequencing panels or enzyme-based assays.Example 2 - Detection of Increased Lung-Derived cfDNA in Treatment Samples

[0089] This example demonstrates that lung-derived cfDNA signals, as measured by lungspecific methylation markers, are elevated in samples obtained from subjects undergoing therapeutic treatment relative to baseline samples.

[0090] A cohort of clinical samples was analyzed, including both baseline (pre-treatment) samples and samples collected during or after treatment with therapeutic agents, including antibody-drug conjugates (ADCs). Methylation levels at lung-specific DMRs were measured using a methylation detection platform, and samples were classified as exhibiting lung-derived cfDNA signals based on predefined thresholds.

[0091] The prevalence of samples exhibiting lung-associated methylation signatures was calculated for different groups. Baseline samples showed a prevalence of approximately 12.7% (349 out of 2742 samples), whereas non-baseline samples showed a higher prevalence of approximately 17.5% (856 out of 4878 samples). Notably, treatment samples exhibited the highest prevalence, approximately 19.4% (453 out of 2339 samples).

[0092] The increased prevalence of lung-associated cfDNA signals in treatment samples relative to baseline indicates that therapeutic intervention is associated with increased lung tissue damage or turnover, resulting in the release of lung-derived cfDNA into circulation. These findings are consistent with known adverse effects of certain therapies, including ADCs, which have been reported to cause interstitial lung disease and pneumonitis.

[0093] In some embodiments, these prevalence differences may be used to define thresholds for classifying samples as indicative of lung injury. In other embodiments, longitudinal changes in individual subjects may be monitored, with increases in lung-derived cfDNA signal over time serving as an early indicator of treatment-induced lung toxicity.Example 3 - Unsupervised Clustering of Samples Based on Lung-Specific Methylation Profiles

[0094] In this example, unsupervised learning techniques were applied to classify samples based on methylation patterns at lung-specific DMRs, demonstrating the ability to distinguish subpopulations of samples with differing levels of lung-derived cfDNA.

[0095] A subset of clinical samples, including approximately 101 treatment samples from a defined cohort, was analyzed after exclusion of outlier samples identified through quality control procedures. Methylation levels at a set of lung-specific DMRs (e.g., 18-19 regions) were quantified and used as input features for clustering algorithms.

[0096] Multiple unsupervised clustering methods were applied, including k-means clustering (e.g., k-means++ initialization), Gaussian mixture models (GMM), principal component analysis (PCA), and UMAP-based dimensionality reduction. Across these methods, samples consistently segregated into two primary clusters, indicating robust underlying structure in the methylation data.

[0097] The two clusters were designated as a “high peak count” cluster and a “low peak count” cluster based on relative methylation signal intensity across the selected DMRs. Samples in the high peak count cluster exhibited elevated methylation signals, suggesting increased contribution of lung-derived cfDNA.

[0098] Further analysis revealed that samples in the high methylation cluster also exhibited a trend toward higher total cfDNA yield compared to samples in the low methylation cluster. This observation supports the hypothesis that increased lung tissue damage leads to increased release of cfDNA into circulation, accompanied by elevated methylation signals at lungspecific regions.

[0099] In some embodiments, clustering results may be used directly to classify samples without requiring labeled training data. In other embodiments, clustering outputs may be used to define training labels for subsequent supervised model development.Example 4 - Identification of ILD-Associated Samples in Clinical Cohorts

[0100] This example illustrates the application of lung-specific DMR features to identify samples associated with interstitial lung disease (ILD) in a clinical cohort.

[0101] A dataset including approximately 132 clinical samples from subjects undergoing treatment with an ADC therapy was analyzed. Methylation levels at approximately 19 lungspecific DMRs were measured and used as input features for clustering and classification.

[0102] Using clustering approaches such as k-means, GMM, and PC A, a subset of samples (e.g., 11 out of 132) was identified as belonging to a high methylation cluster consistent with elevated lung-derived cfDNA signals. These samples were designated as potential ILD-positive cases.

[0103] Comparative analysis showed that the identified ILD-associated samples exhibited significantly higher cfDNA levels relative to other samples in the cohort. Additionally, heatmap visualization of methylation patterns demonstrated distinct profiles for the ILD-associated group, indicating that the lung-specific DMR features capture biologically meaningful differences.

[0104] It was noted that a substantial proportion of samples (e.g., approximately 85.6%) were positive for a breast cancer methylation signal, which may confound interpretation of cfDNA levels. Despite this potential confounding factor, the lung-specific methylation features were still able to distinguish a subset of samples with characteristics consistent with lung injury.

[0105] In certain embodiments, orthogonal validation methods (e.g., imaging or clinical diagnosis) may be used to confirm ILD status. In other embodiments, the methylation-based classification may be used as an early screening tool to identify subjects at risk of developing ILD.Example 5 - Development and Validation of a Predictive Model for ILD Detection

[0106] In this example, a supervised machine learning model was developed and evaluated for the purpose of predicting ILD status based on lung-specific methylation features.

[0107] A logistic regression (LR) model was trained using methylation data from lungspecific DMRs as input features. The training dataset included labeled samples representing ILD-positive and ILD-negative cases, derived from clustering results and / or clinical annotations. Model training was performed using cross-validation techniques, such as 10-fold cross-validation, to ensure robustness and to minimize overfitting.

[0108] Model performance was evaluated using standard metrics. The LR model achieved an average area under the receiver operating characteristic curve (AUC) of approximately 0.964, with corresponding high sensitivity (approximately 1.0) and specificity (greater than approximately 0.93). These results demonstrate that the lung-specific DMR features provide strong predictive power for distinguishing ILD-associated samples.

[0109] The trained model was further applied to an independent test dataset including approximately 102 samples that were negative for breast cancer methylation signals. In this dataset, no samples were classified as ILD-positive, which was attributed to lower overall methylation levels at the selected DMRs in the test cohort.

[0110] Analysis of training versus test data indicated that the training dataset exhibited higher methylation values, suggesting a potential overfitting effect or differences in sample composition. These findings highlight the importance of selecting representative training data and may inform future model refinement.

[0111] In some embodiments, model thresholds may be adjusted to achieve desired trade-offs between sensitivity and specificity. In other embodiments, additional features, such as cfDNA concentration or alternative methylation markers, may be incorporated to improve model performance.Example 6- Correlation of Lung-Specific DMRs with Global Methylation Signals

[0112] This example examines the relationship between individual lung-specific DMR features and overall methylation signal in clinical samples.

[0113] A set of approximately 19 lung-specific DMRs was analyzed across a cohort of clinical samples, including subjects undergoing treatment with ADC therapies. For each region, the correlation between methylation level and a sample-level methylation score (e.g., derived from a separate classifier such as a breast cancer methylation score) was calculated using statistical measures such as Spearman correlation.

[0114] Results indicated that approximately 10 out of the 19 regions exhibited strong correlation with the global methylation score, with correlation coefficients (p) of approximately 0.7 or greater. These findings suggest that certain lung-specific regions may also be influenced by broader methylation patterns, including those associated with tumor-derived DNA.

[0115] The presence of such correlations highlights the importance of careful feature selection to minimize confounding effects. In some embodiments, regions exhibiting strong correlation with non-lung signals may be excluded or down-weighted in model training. In other embodiments, multivariate models may be used to disentangle overlapping signals and isolate lung-specific contributions.

[0116] Despite potential confounding, the overall set of lung-specific DMRs was shown to provide sufficient discriminatory power for detecting ILD-associated samples. In certain embodiments, subsets of the DMRs may be selected to optimize specificity for lung-derived signals while maintaining sensitivity.

Claims

CLAIMS1. A method for detecting lung injury in a subject, comprising:obtaining a biological sample comprising cell-free DNA;detection methylation at a plurality of genomic regions;generating a methylation profile; andclassifying the subject as having lung injury based on the methylation profile.

2. The method of claim 1, wherein the lung injury comprises interstitial lung disease (ILD).

3. The method of claim 1, wherein the lung injury is induced by a therapeutic agent.

4. The method of claim 3, wherein the therapeutic agent is an antibody-drug conjugate.

5. The method of claim 4, wherein the antibody-drug conjugate comprises trastuzumab deruxtecan.

6. The method of claim 1, comprising differentially methylated regions in lung tissue compared to non-lung tissues.

7. The method of claim 1, comprising differentially methylated regions in injured lung tissue compared to non-injured lung tissues.

8. The method of claim 1, wherein the plurality of genomic regions comprises at least 10 lung-specific DMRs.

9. The method of claim 8, wherein the plurality comprises 10, 20, 30, 40, 50, 60, 70, 80, 90 or more DMRS,.

10. The method of claim 9, wherein the plurality comprises between 10 and 30 DMRs, optionally 12- 25 DMRs.

11. The method of claim 1, wherein measuring methylation comprises bisulfite sequencing.

12. The method of claim 1, wherein measuring methylation comprises a targeted sequencing panel.

13. The method of claim 1, wherein measuring methylation comprises a methylationsensitive restriction enzyme assay.

14. The method of claim 1, wherein classifying comprises applying a machine learning model.

15. The method of claim 14, wherein the model comprises logistic regression.

16. The method of claim 15, wherein the model distinguishes between high and low methylation clusters.

17. The method of claim 1, wherein the plurality of genomic regions comprises regions having at least 8 CpG sites.

18. The method of claim 1, wherein the genomic regions exhibit a methylation difference of at least 0.2 between lung tissue and non-lung tissues.

19. The method of claim 1, wherein the genomic regions exhibit a methylation level of at least 0.6 in lung tissue and at most 0.5 in non-lung tissues.

20. The method of claim 1, wherein the genomic regions are selected to have background methylation levels of less than 5% in circulating DNAfrom healthy subjects.

21. The method of claim 1, wherein the genomic regions are selected to exclude regions associated with methylation patterns of non-lung cancer tissues.

22. The method of claim 1, wherein the biological sample comprises plasma-derived cell-free DNA.

23. The method of claim 1, further comprising quantifying a total amount of cell-free DNA and incorporating the total amount into the classification.

24. The method of claim 23, wherein an elevated total cell-free DNA level relative to a baseline is indicative of lung injury.

25. The method of claim 1, wherein measuring methylation comprises bisulfite conversion followed by next-generation sequencing.

26. The method of claim 1, wherein measuring methylation comprises amplification using primers specific to the plurality of genomic regions.

27. The method of claim 1, wherein measuring methylation comprises digestion using one or more methylation-sensitive restriction enzymes.

28. The method of claim 27, wherein the genomic regions are selected to include at least one methylation-sensitive restriction enzyme cut site.

29. The method of claim 1, wherein generating the methylation profile comprises determining a peak count or signal intensity for each genomic region.

30. The method of claim 1, wherein classifying comprises assigning the subject to a high methylation cluster or a low methylation cluster.

31. The method of claim 30, wherein assignment to the high methylation cluster is indicative of lung injury.

32. The method of claim 14, wherein the machine learning model is trained using cross-validation.

33. The method of claim 14, wherein the machine learning model applies a classification threshold selected to achieve a specificity of at least 90%.

34. The method of claim 16, wherein the increase in lung-derived methylation signal is detected relative to a pre-treatment baseline sample.

35. The method of claim 34, wherein the method further comprises determining a rate of change of methylation signal over time.

36. The method of claim 15, wherein the lung injury is associated with progression toward interstitial lung disease.

37. The method of claim 3, wherein the therapeutic agent is selected from an antibody-drug conjugate, a chemotherapeutic agent, or an immunotherapy agent.

38. The method of claim 37, wherein the antibody-drug conjugate is selected from trastuzumab deruxtecan or sacituzumab govitecan.

39. The method of claim 1, wherein the subject does not have lung cancer.

40. A method for monitoring treatment-induced lung toxicity, comprising performing the method of claim 1 at multiple time points and identifying an increase in lung-derived methylation signal.

41. A system comprising:a sequencing or methylation detection module;a processor configured to analyze methylation at lung-specific genomic regions; and a classifier configured to detect lung injury. The system of claim 16, wherein the processor is configured to perform clustering analysis using at least one of k-means clustering, Gaussian mixture modeling, principal component analysis, or UMAP.

42. The system of claim 41, wherein the classifier is configured to output a probability score indicative of lung injury.

43. A panel comprising probes or primers targeting a plurality of genomic regions that are differentially methylated in lung tissue relative to non-lung tissues.

44. The panel of claim 43, wherein the plurality of genomic regions comprises between 15 and 30 lung tissue-specific DMRs.

45. The panel of claim 43, wherein the panel is configured for use in a targeted sequencing assay or a methylation-sensitive restriction enzyme assay.

Images