Classifier Scores Research Articles

Abstract B058: Novel machine learning lung cancer classifier shows significant correlation with cancer-specific fragmentomics features

Abstract Lung cancer is the second most prevalent cancer, and the leading cause of cancer-related deaths, in the US. Since the survival rate of patients diagnosed with lung cancer at earlier stages is dramatically higher than those diagnosed at later stages, technologies that shift lung cancer diagnosis to an earlier stage would be expected to improve survival rates. Fragmentomics, the study of short circulating cell-free DNA (cfDNA) fragments in the blood, is an emerging field in cancer diagnostics that has the potential to catalyze such a shift. Studies have shown that fragmentomics features, such as fragment sizes and end motifs, can be used to discriminate normal cfDNA from circulating tumor DNA (ctDNA) generated through tumor necrosis and apoptosis. Using this knowledge, machine learning tools, trained to identify ctDNA-specific features, are being leveraged in novel liquid biopsy assays to improve early-stage lung cancer detection. To create a machine learning classifier that detects discrete signals of ctDNA in individuals with lung cancer, Genece Health utilized low pass (∼3x coverage) whole genome sequencing (Illumina) of plasma extracted cfDNA from >400 lung cancer samples and >1,000 negative controls. This data was used to generate our novel FEMS (Fragment End Motifs and Sizes) dataframe, containing >30,000 features, by combining fragment lengths with their corresponding 5’ 4bp end motifs. The FEMS dataframe was then used to train a neural network machine learning classifier. To evaluate the performance of our machine learning classifier, we investigated whether the classifier scores correlated with well-known hallmarks of ctDNA. First, consistent with published ctDNA characteristics, we observed significant (p<0.01) enrichment of fragment sizes ≤167bp in lung cancer samples. Concordantly, our classifier scores are significantly correlated with this fragment size enrichment. Also, both fragment size enrichment and our classifier scores are correlated with lung cancer stage, with higher enrichment and higher scores observed in later stages. Next, we show that our classifier scores are correlated with specific end motif enrichments or depletions in lung cancer samples. For example, in lung cancer samples, CCCA end motifs were significantly enriched (p<0.01) in fragment sizes 120-140bp (max at 128bp), while significantly depleted in fragment sizes 190-210bp (min at 204bp). Once again, our classifier scores were significantly correlated with both the enrichment of this end motif in short fragments and its depletion in long fragments. Last, we assessed the classifier’s performance in an independent cohort (46 lung cancer and 132 non-cancer samples) and show 89.2% sensitivity with 90.4% specificity (auROC=0.957). The correlation observed between our classifier scores and published ctDNA characteristics, combined with the strong performance in an independent cohort, suggests that Genece’s novel machine learning classifier is effectively leveraging ctDNA-specific fragmentomics features for the identification of lung cancer in patient samples. Citation Format: Carlos Guzman, Michael L Salmans, Mengchi Wang, Byung In Lee, Andrew R Carson. Novel machine learning lung cancer classifier shows significant correlation with cancer-specific fragmentomics features [abstract]. In: Proceedings of the AACR Special Conference: Liquid Biopsy: From Discovery to Clinical Implementation; 2024 Nov 13-16; San Diego, CA. Philadelphia (PA): AACR; Clin Cancer Res 2024;30(21_Suppl):Abstract nr B058.

Enhancing the convenience of frailty index assessment for elderly Chinese people with machine learning methods.

Frailty is a state that is closely associated with adverse health outcomes in the aging process. The frailty index (FI), which measures frailty in terms of cumulative deficits, has been widely used for frailty assessment in elderly people, and its advantage of self-reported information collection makes it applicable to a broader group of elderly people. Our study aims to simplify the Frailty Index Assessment Scale, while maintaining its reliability and accuracy, to easily and quickly assess frailty in elderly people. In this study, participants (age ≥ 65 years) from the Chinese Longitudinal Healthy Longevity Survey (CLHLS), which had 13,339, 372 and 1214 participants in 2008, 2011, and 2014, respectively, were used. The 2008 dataset was split into 80% for training and 20% for internal validation, and the data from 2011 to 2014 as external validation. In order to obtain effective predictors, we used Lasso regression, Boruta algorithm and random forest classifier score for feature selection. We used six models for predictive model construction and evaluated the models in the validation dataset. Model performance was measured by area under the curve (AUC), accuracy and F1 score. Logistic regression was found to be the best performing and most interpretable algorithm with AUC, accuracy and F1 of 0.974, 0.932 and 0.880 for the validation dataset, respectively. The AUCs for the external independent validation dataset were 0.963 and 0.977, respectively. Subgroup analysis showed that the model had good predictive power in both males and females. The predictive power was stronger among the elderly people over 80 years old, with AUC, accuracy and F1 of 0.973,0.914, and 0.893, respectively. The model also obtained good predictive power in the case of FI measured by different indicators. The model showed good robustness in the follow-up assessment of frailty status in elderly people, with the AUC remaining above 0.95 and accuracy above 0.9 over the long-term follow-up. Using machine learning techniques, we have successfully developed a simple frailty assessment prediction model based on 10 key features to shorten the frailty assessment scale with near full-scale accuracy. A user-friendly website was created to facilitate the application of this prediction model ( https://healthy-aging.shinyapps.io/Frailty_Assessment/ ).

Sequential label shift detection in classification data: An application to dengue fever.

Classifiers have been developed to help diagnose dengue fever in patients presenting with febrile symptoms. However, classifier predictions often rely on the assumption that new observations come from the same distribution as training data. If the population prevalence of dengue changes, as would happen with a dengue outbreak, it is important to raise an alarm as soon as possible, so that appropriate public health measures can be taken and also so that the classifier can be re-calibrated. In this paper, we consider the problem of detecting such a change in distribution in sequentially-observed, unlabeled classification data. We focus on label shift changes to the distribution, where the class priors shift but the class conditional distributions remain unchanged. We reduce this problem to the problem of detecting a change in the one-dimensional classifier scores, leading to simple nonparametric sequential changepoint detection procedures. Our procedures leverage classifier training data to estimate the detection statistic, and converge to their parametric counterparts in the size of the training data. In simulated outbreaks with real dengue data, we show that our method outperforms other detection procedures in this label shift setting.

Transcriptomic Profiling of Primary Prostate Cancers and Nonlocalized Disease on Prostate-Specific Membrane Antigen Positron Emission Tomography/Computed Tomography: A Multicenter Retrospective Study.

To characterize the relationship between Decipher genomic classifier scores and prostate-specific membrane antigen (PSMA) positron emission tomography/computed tomography (PET/CT)-based metastatic spread. We identified patients from four institutions who underwent PSMA PET/CT scans pretreatment for primary staging or postradical prostatectomy (RP) for suspected recurrence and had Decipher transcriptomic data available from biopsy or RP specimens. PSMA PET/CT-based patterns of spread were classified as localized (miT + N0M0) or nonlocalized (miN1M0 or miM1a-c). We calculated the association between Decipher scores and the risk of nonlocalized disease on PSMA PET/CT using multivariable logistic regression for pretreatment patients and multivariable Cox regression for post-RP patients. We also compared select transcriptomic signatures between patients with localized and nonlocalized diseases. Five hundred eighty-six patients were included (pretreatment: n = 329; post-RP: n = 257). Higher Decipher scores were associated with nonlocalized disease on PSMA PET/CT both pretreatment (odds ratio, 1.18 [95% CI, 1.03 to 1.36] per 0.1 increase in Decipher score, P = .02) and post-RP (hazard ratio, 1.15 [95% CI, 1.05 to 1.27] per 0.1 increase in Decipher score, P = .003). In the pretreatment setting, nonlocalized disease was associated with higher rates of TP53 mutations and lower rates of PAM50 luminal A subtype compared with localized disease. In the post-RP setting, overexpression of signatures related to metabolism, DNA repair, and androgen receptor signaling were associated with higher rates of nonlocalized disease. Higher Decipher scores were associated with nonlocalized disease identified on PSMA PET/CT both pretreatment and post-RP. There were several transcriptomic differences between localized and nonlocalized diseases in both settings.

All thresholds barred: direct estimation of call density in bioacoustic data

Passive acoustic monitoring (PAM) studies generate thousands of hours of audio, which may be used to monitor specific animal populations, conduct broad biodiversity surveys, detect threats such as poachers, and more. Machine learning classifiers for species identification are increasingly being used to process the vast amount of audio generated by bioacoustic surveys, expediting analysis and increasing the utility of PAM as a management tool. In common practice, a threshold is applied to classifier output scores, and scores above the threshold are aggregated into a detection count. The choice of threshold produces biased counts of vocalizations, which are subject to false positive/negative rates that may vary across subsets of the dataset. In this work, we advocate for directly estimating call density: The proportion of detection windows containing the target vocalization, regardless of classifier score. We propose a validation scheme for estimating call density in a body of data and obtain, through Bayesian reasoning, probability distributions of confidence scores for both the positive and negative classes. We use these distributions to predict site-level call densities, which may be subject to distribution shifts (when the defining characteristics of the data distribution change). These methods may be applied to the outputs of any binary detection classifier operating on fixed-size audio input windows. We test our proposed methods on a real-world study of Hawaiian birds and provide simulation results leveraging existing fully annotated datasets, demonstrating robustness to variations in call density and classifier model quality.

DNA Methylation Profiles Are Stable in H3 K27M-Mutant Diffuse Midline Glioma Neurosphere Cell Lines.

Diffuse midline gliomas are among the deadliest human cancers and have had little progress in treatment in the last 50 years. Cell cultures of these tumors have been developed recently, but the degree to which such cultures retain the characteristics of the source tumors is unknown. DNA methylation profiling offers a powerful tool to look at genome-wide epigenetic changes that are biologically meaningful and can help assess the similarity of cultured tumor cells to their in vivo progenitors. Paraffinized diagnostic tissue from three diffuse intrinsic pontine gliomas with H3 K27M mutations was compared with subsequent passages of neurosphere cell cultures from those tumors. Each cell line was passaged 3-4 times and analyzed with DNA methylation arrays and standard algorithms that provided a comparison of diagnostic classification and cluster analysis. All samples tested maintained high classifier scores and clustered within the reference group of H3 K27M-mutant diffuse midline gliomas. There was a gain of 1q in all cell lines, with two cell lines initially manifesting the gain of 1q only during culture. In vitro cell cultures of H3 K27M-mutant gliomas maintain high degrees of similarity in DNA methylation profiles to their source tumor, confirming their fidelity even with some chromosomal changes.

Molecular refinement of pilocytic astrocytoma in adult patients.

Pilocytic astrocytomas (PA) in adults are rare and may be challenging to identify based only on histomorphology. Compared to their paediatric counterparts, they are reportedly molecularly more diverse and associated with a worse prognosis. We aimed to describe the characteristics of adult PAs more precisely by comprehensively profiling a series of 79 histologically diagnosed adult cases (≥18 years). We performed global DNA methylation profiling and DNA and RNA panel sequencing, and integrated the results with clinical data. We further compared the molecular characteristics of adult and paediatric PAs that had a significant match to one of the established PA methylation classes in the Heidelberg brain tumour classifier. The mean age in our cohort was 33 years, and 43% of the tumours were located supratentorially. Based on methylation profiling, only 39% of the cases received a significant match to a PA methylation class. Sixteen per cent matched a different tumour type and 45% had a Heidelberg classifier score <0.9 with an affiliation to diverse established methylation classes in t-SNE analyses. Although the KIAA1549::BRAF fusion was found in 98% of paediatric PAs, this was true for only 27% of histologically defined and 55% of adult PAs defined by methylation profiling. A particularly high fraction of adult tumours with histological features of PA do not match current PA methylation classes, indicating ambiguous histology and an urgent need for molecular profiling. Moreover, even in adult PAs with a match to a PA methylation class, the distribution of genetic drivers differs significantly from their paediatric counterparts (p<0.01).

Label-Aware Dual Graph Neural Networks for Multi-Label Fundus Image Classification.

Fundus disease is a complex and universal disease involving a variety of pathologies. Its early diagnosis using fundus images can effectively prevent further diseases and provide targeted treatment plans for patients. Recent deep learning models for classification of this disease are gradually emerging as a critical research field, which is attracting widespread attention. However, in practice, most of the existing methods only focus on local visual cues of a single image, and ignore the underlying explicit interaction similarity between subjects and correlation information among pathologies in fundus diseases. In this paper, we propose a novel label-aware dual graph neural networks for multi-label fundus image classification that consists of population-based graph representation learning and pathology-based graph representation learning modules. Specifically, we first construct a population-based graph by integrating image features and non-image information to learn patient's representations by incorporating associations between subjects. Then, we represent pathologies as a sparse graph where its nodes are associated with pathology-based feature vectors and the edges correspond to probability of the co-occurrence of labels to generate a set of classifier scores by the propagation of multi-layer graph information. Finally, our model can adaptively recalibrate multi-label outputs. Detailed experiments and analysis of our results show the effectiveness of our method compared with state-of-the-art multi-label fundus image classification methods.

Distinct Molecular Processes Mediate Donor-derived Cell-free DNA Release From Kidney Transplants in Different Disease States.

Among all biopsies in the Trifecta-Kidney Study ( ClinicalTrials.gov NCT04239703), elevated plasma donor-derived cell-free DNA (dd-cfDNA) correlated most strongly with molecular antibody-mediated rejection (AMR) but was also elevated in other states: T cell-mediated rejection (TCMR), acute kidney injury (AKI), and some apparently normal biopsies. The present study aimed to define the molecular correlates of plasma dd-cfDNA within specific states. Dd-cfDNA was measured by the Prospera test. Molecular rejection and injury states were defined using the Molecular Microscope system. We studied the correlation between dd-cfDNA and the expression of genes, transcript sets, and classifier scores within specific disease states, and compared AMR, TCMR, and AKI to biopsies classified as normal and no injury (NRNI). In all 604 biopsies, dd-cfDNA was elevated in AMR, TCMR, and AKI. Within AMR biopsies, dd-cfDNA correlated with AMR activity and stage. Within AKI, the correlations reflected acute parenchymal injury, including cell cycling. Within biopsies classified as MMDx Normal and archetypal No injury (NRNI), dd-cfDNA still correlated significantly with rejection- and injury-related genes. TCMR activity (eg, the TCMR Prob classifier) correlated with dd-cfDNA, but within TCMR biopsies, top gene correlations were complex and not the top TCMR-selective genes. In kidney transplants, elevated plasma dd-cfDNA is associated with 3 distinct molecular states in the donor tissue: AMR, recent parenchymal injury (including cell cycling), and TCMR, potentially complicated by parenchymal disruption. Moreover, subtle rejection- and injury-related changes in the donor tissue can contribute to dd-cfDNA elevations in transplants considered to have no rejection or injury.

Quality of T2-weighted MRI re-acquisition versus deep learning GAN image reconstruction: A multi-reader study

PurposeTo evaluate CycleGAN's ability to enhance T2-weighted image (T2WI) quality. MethodA CycleGAN algorithm was used to enhance T2WI quality. 96 patients (192 scans) were identified from patients who underwent multiple axial T2WI due to poor quality on the first attempt (RAD1) and improved quality on re-acquisition (RAD2). CycleGAN algorithm gave DL classifier scores (0–1) for quality quantification and produced enhanced versions of QI1 and QI2 from RAD1 and RAD2, respectively. A subset (n = 20 patients) was selected for a blinded, multi-reader study, where four radiologists rated T2WI on a scale of 1–4 for quality. The multi-reader study presented readers with 60 image pairs (RAD1 vs RAD2, RAD1 vs QI1, and RAD2 vs QI2), allowing for selecting sequence preferences and quantifying the quality changes. ResultsThe DL classifier correctly discerned 71.9 % of quality classes, with 90.6 % (96/106) as poor quality and 48.8 % (42/86) as diagnostic in original sequences (RAD1, RAD2). CycleGAN images (QI1, QI2) demonstrated quantitative improvements, with consistently higher DL classifier scores than original scans (p < 0.001). In the multi-reader analysis, CycleGAN demonstrated no qualitative improvements, with diminished overall quality and motion in QI2 in most patients compared to RAD2, with noise levels remaining similar (8/20). No readers preferred QI2 to RAD2 for diagnosis. ConclusionDespite quantitative enhancements with CycleGAN, there was no qualitative boost in T2WI diagnostic quality, noise, or motion. Expert radiologists didn’t favor CycleGAN images over standard scans, highlighting the divide between quantitative and qualitative metrics.

Abstract C139: The robustness of a targeted methylation-based multi-cancer early detection (MCED) test to population differences in self-reported ethnicity

Abstract Background: DNA methylation (Me) is known to vary by race and ethnicity; cell-free DNA-based MCED tests must detect cancer-specific Me in the presence of this variation to be useful as a screening tool. Clinical study cohorts, however, often do not reflect the screening population. Additionally, underserved populations who may benefit most from improved cancer screening are often underrepresented. We sought to evaluate the robustness of an MCED test to detect cancer-specific Me patterns across self-reported ethnicity (SRE) in the training (N = 4487) and validation (N = 5309) datasets from the 15254-participant (pts) Circulating Cell-free Genome Atlas (CCGA; NCT02889978) study. Methods: To understand MCED test performance across SRE, we assessed concordance between ancestry and SRE by genotype in CCGA. We then conducted two in silico experiments to assess the robustness of the MCED test to differences in SRE. First, MCED test classifier scores from 1254 non-cancer pts from the validation cohort were regressed against covariates of sex and SRE, with smoking status and age included as potential confounders. Two SRE groups were considered for the analysis: White, non-Hispanic (W; 996 pts) and all other non-White SRE (NW; 258 pts). Second, to evaluate classifier generalization across SRE, two experimental MCED test classifiers (single SRE, S; multiple SRE, M) were trained similarly to the MCED test with intentionally skewed SRE compositions: S with 100% W pts and M with all available NW pts (20% NW). The training sets for both classifiers further included pts from STRIVE (NCT03085888) and matched on demographics (age, sex, smoking status), number of pts, samples, cancer type and stage. Sensitivity (at 99.4% specificity) and accuracy of cancer signal origin (CSO) prediction for the two classifiers were assessed in a separate held out population of 507 NW cancer pts. Results: Ancestry and SRE by genotype were in good agreement with &amp;gt;95% concordance in CCGA study pts. Neither sex, smoking status, nor SRE were significantly correlated with MCED classifier scores; age was weakly (r2 &amp;lt; 0.01) associated. In the classifier training experiment, both S and M performed nearly identically for sensitivity (S: 57% [95% CI: 53-62], 290/507 vs M: 57% [52-61], 288/507; McNemar test, P &amp;gt; 0.5) and CSO prediction accuracy (S: 90% [86-93] 262/290 vs M: 91% [88-94] 263/288) in a diverse population of NW cancer pts, despite the relatively large difference in SRE training composition. Neither analysis detected a significant contribution from Me patterns associated with SRE on classifier performance. Conclusions: Though DNA Me differences are known to occur across SRE, MCED classifier scores were comparable across non-cancer pts in SRE groups. Similarly, cancer signal detection in a NW population was comparable regardless of classifier training exposure. These findings indicate that detection of invasive cancer by the MCED test is not strongly influenced by SRE. Ongoing interventional studies of the MCED test are targeting enrollment of population-representative SRE. Citation Format: Oliver Venn, Joerg Bredno, Alexis Thornton, Earl Hubbell, Kathryn Kurtzman, John Beausang, Charles Swanton. The robustness of a targeted methylation-based multi-cancer early detection (MCED) test to population differences in self-reported ethnicity [abstract]. In: Proceedings of the 16th AACR Conference on the Science of Cancer Health Disparities in Racial/Ethnic Minorities and the Medically Underserved; 2023 Sep 29-Oct 2;Orlando, FL. Philadelphia (PA): AACR; Cancer Epidemiol Biomarkers Prev 2023;32(12 Suppl):Abstract nr C139.

Predictive Value of Genomic Classifier Scores and Transcriptomic Data for Prostate Cancer Distant Metastasis Risk: A Multicenter Retrospective Study.

Prostate-specific membrane antigen (PSMA) positron emission tomography (PET) has a greater specificity and sensitivity for detection of extraprostatic prostate cancer than conventional imaging. The Decipher genomic classifier is an established prognostic biomarker being evaluated for its ability to predict systemic treatment intensification. The relationship between Decipher scores and PSMA-based spread remains unknown, as do differences in transcriptomic patterns of PSMA PET-based spread in the de novo vs. recurrent setting. We retrospectively identified patients who (a) had undergone staging with a PSMA PET prior to treatment or for evaluation of recurrence post-radical prostatectomy (RP) at two institutions and (b) had transcriptomic data available from the Genomics Resource for Intelligent Discovery (GRID) database from either biopsy or RP specimens. We classified the PSMA PET pattern of spread using molecular imaging (mi) staging as localized (miT+N0M0), node-positive (miN1M0), distant metastasis (miM1a-c), or negative/non-diagnostic. We used logistic regression to calculate the odds ratios (OR) with 95% confidence intervals (CI) for distant metastasis risk based on Decipher score both pre-treatment and post-RP. As an exploratory analysis, we compared each of the staging groups for differences in important transcriptomic signatures. Kruskal-Wallis and Pearson chi-squared tests were used for continuous and categorical variables, respectively. A total of 315 patients were included in this analysis (n = 164 pre-treatment, n = 151 post-RP). Eighty PSMA PET scans were negative, while 147 were miT+N0M0, 45 were miN1M0, and 43 were miM1a-c. A higher Decipher score was associated with distant metastasis (miM1a-c) on PSMA PET both pre-treatment (OR 1.3 [95% CI: 1.0-1.7] per 0.1 increase in Decipher score, P = 0.05) and post-RP (OR 1.2 [1.0-1.4] per 0.1 increase in Decipher score, P = 0.04). There were higher TP53 mutation (P = 0.01) and cell cycle progression (P = 0.04) signature scores in miM1a-c patients compared to miN1M0 or miT+N0M0 patients. Basal subtype was more prevalent per PAM50 in miM1a-c or miN1M0 patients (36%) than miT+N0M0 patients (19%, P=0.01). Patients with de novo miN1M0 or miM1a disease (n = 19) had higher Decipher scores (0.85 vs 0.57, P = 0.10) and IFNa response (P = 0.08) than patients with recurrent miN1M0 or miM1a disease (n = 35). Higher Decipher scores were associated with distant metastasis on PSMA PET in both the de novo and recurrent setting. Transcriptomic differences in pathways related to proliferation, p53 status, and PAM50 classification were seen when comparing localized, node-positive, and distant metastatic disease. Patients with de novo miN1M0 or miM1a disease may harbor more aggressive disease than those with miN1M0 or miM1a disease at recurrence.

Dynamic Changes of Molecular Subtype Classification and Genomic Classifier Scores in High-Risk Prostate Cancer Patients Undergoing Pre-Operative Stereotactic Body Radiation Therapy

The radiobiology of irradiated human prostate cancer (PCa) remains poorly understood as irradiated tissues often remain in situ. PREPARE-SBRT (NCT03663218) is a clinical trial testing the safety of neoadjuvant MRI-guided stereotactic body radiation therapy (SBRT) for men with high-risk PCa. We leveraged paired samples from pre-treatment biopsy and irradiated prostatectomy (RP) specimens to evaluate transcriptomic changes in irradiated tumors at acute timepoints following SBRT. Tumor RNA expression profiles were generated using Decipher GRID by Veracyte on 12 subjects enrolled on NCT03663218 with paired pre/post-SBRT tissues (n = 24). Descriptive statistics using Decipher Genomic Classifier (GC) Score [0-1] and GC risk group (low/int/high) were generated from a validated 22-gene GC. Tumor biology signatures reported as dichotomous variables evaluated changes in androgen receptor (AR) activity [higher v lower] and cell cycle progression (CCP) [lower v higher]. Decipher prostate subtype classifier [PSC, luminal differentiated (LD), luminal proliferating (LP), basal immune (BI), basal neuroendocrine like (BN)] classification and PAM50 molecular subtype [luminal A (lumA), luminal B (lumB), basal] at pre/post-SBRT timepoints were reported. A control cohort of transcriptomic profiles of 803 matched untreated biopsy and RP samples from the same patients were used to control for signature differences attributable to sample type. The median pre- and post-SBRT GC scores were 0.55 and 0.72 (Δ+0.17), respectively. By comparison, median GC scores in a control cohort (n = 803) of biopsy and RP specimens were 0.50 and 0.56 (Δ+0.06), respectively. SBRT increased GC score in 9/12 subjects (75%) with a median increase of 0.3. Changes in GC score resulted in reclassification of baseline GC risk in 7 of 12 subjects with 71% of reclassified subjects (5/7) transitioning to a higher genomic risk. 92% of subjects (11/12) had higher AR activity scores at baseline. Of this subgroup, 5/11 (45%) converted to lower AR activity score after SBRT. CCP signatures remained stable in 9 of 12 subjects (75%) with 7/12 subjects exhibiting lower CCP score at baseline and only 1 subject transitioning from lower to higher CCP score. Distribution of PAM50 molecular subtype at baseline and after SBRT was lumA (33>53%), lumB (25>17%), basal (42>25%) resulting in 83% of subjects (10/12) annotated to a different PAM50 molecular subtype at pre/post-SBRT assessment. PSC subtype distribution at baseline was LD (33%), LP (25%), BI (33%) and BN (8%). Strikingly, after SBRT, 92% (11/12) of subjects were classified as BI molecular subtype with several immune activation signatures also increased after SBRT. A majority of subjects demonstrate a post-SBRT increase in GC score with reclassification of genomic-prognostic risk group in 58%. An enrichment of the basal-immune molecular subtype was observed following SBRT suggesting a convergence towards this biology in irradiated tumors.

Synthetic accessibility-informed designing of efficient organic semi-conductors for organic solar cells

Modern computer-assisted synthesis offers solid support for easy virtual screening of easily synthesizable compounds. By using the synthetic accessibility score this restriction may be bypassed. In the present work, a chemical space of organic semi-conductors is generated. Synthetic accessibility of generated organic semi-conductors is studied using Synthetic accessibility (SA) score and Synthetic Bayesian classifier (SYBA) score. These methods are very fast. A significance difference is observed between results obtained from two groups. The distribution of predicted scores using two methods is also different. More research is required in this field. Generated organic semi-conductors are screened on the basis of calculated scores. This study is indicating the potential of synthetic accessibility scores in the virtual screening. These faster methods are ideal for large scale screening.

P07.05.B DNA-METHYLATION PROFILING IS AN EFFECTIVE ASSET FOR IDENTIFICATION OF TUMORS IN SUSPECTED, YET IMMUNOHISTOCHEMICALLY UNSPECIFIABLE, NEURO-ONCOLOGICAL CASES

Abstract BACKGROUND Patients with a suspected neurooncological disease on radiographic images and no histopathological evidence of a tumor on the surgically retrieved tissue, pose a great challenge for clinicians and neuropathologists. DNA methylation-based classifications and copy number variation (CNV) analysis allow robust brain tumor classification. Here, we aim to evaluate their diagnostic properties in histopathologically unspecifiable or negative cases. MATERIAL AND METHODS We screened all neurosurgical cases, which underwent surgery at our institution between 2009 and 2021, with suspected neurooncological, but negative or unspecific histopathological diagnosis. We differentiated two groups: cases with cell-enriched, reactive tissue (+/- single tumor cells), insufficient to fully classify the lesion according to criteria of WHO for CNS tumors (group 1) and cases without histological evidence of tumor cells (group 2). In both groups, DNA methylation profiling was applied. Endpoints were diagnostic results of DNA methylation profiling (classifier Score ≥0.9) and evidence/no evidence of tumorigenic CNV alterations and eventually feasibility to make a diagnosis according to WHO 2021 criteria for brain tumors. RESULTS 23 cases with unspecifiable histopathological diagnosis were identified, out of whom 5 (21.7%) were confirmed as healthy brain tissue via DNA-methylation profiling, whereas the eventual diagnosis of tumorigenic tissue was provided in 18 (78.3%) cases. 14 of these were assigned to group 1, 4 cases to group 2. In 6 cases of group 1 (42.9%) DNA-methylation profiling resulted in a successful tumor classification (score ≥0.9) compared to 1 case (25%) in group 2. CNV indicated tumorigenic alterations in 9 (64.3%) patients in group 1 and 2 (50%) patients in group 2. Specific tumor alterations in the CNV were found in 8 (57.1%) patients in group 1 and in 1 (25%) patient in group 2. Combining CNV and DNA-methylation profiling, a final tumor diagnosis according to WHO 2021 criteria was feasible in 9 (64.3%) cases in group 1 and 3 (75%) cases in group 2. Median time from surgery to histopathological diagnosis was 6 and 7 days and 32.5 and 31 days for integrated diagnosis (group 1 vs. group 2). CONCLUSION Molecular diagnostics with DNA-methylation profiling and CNV analysis substantially increases the likelihood of a definitive diagnosis according to 2021 WHO criteria for these challenging cases.

Grapes Quality Prediction Using Iot &amp; Machine Learning Based on Pre Harvesting

Minimizing pesticide use, preserving water, as well as enhancing soil health are just a few of the sustainable farming techniques that must be carefully considered while growing grapes of a high calibre. These practices can help preserve the environment and ensure the longevity of the vineyard. However, it is difficult for the farmers to find the suitability of the soil and its environment to cultivate grapes with high quality. Thus this research aims to evaluate the fitness of the soil for the fitness of growing quality grapes with the aid of machine learning algorithm. The research was done on Nasik region which is called as the “Grape Capital of India” situated in Maharashtra. Total of 154 villages were considered for the examination and soil specimens were collected and sent to the government testing lab in Maharashtra. The soil characteristics by considering both micro and macro nutrients, and the water characteristics were obtained from the lab. Also the climatic features, quality of the petiole and fruit characteristics were included for creating the dataset. These data was given to six different machine learning algorithm to classify the soil by defining whether the soil is fit for grapes or not. Moreover, this research proposed to analyze the correlation between the nutrients by which the relationship and dependency between the different nutrients and features were considered for defining the grapes quality. Also both the micro and macro nutrients were given equal importance in defining the soil quality suitable for obtaining high quality grapes. Based on the results obtained, Pimpalas Ramche contains more nutrients for the grape to grow more successfully based on samples gathered from different vine yards and the decision tree classifier scores better than any other classifiers among the machine learning algorithms employed in terms of accuracy.

EPEN-10. SPINAL EPENDYMOMA WITH MYCN-AMPLIFICATION – A DISEASE OF CHILDHOOD AND YOUNG ADULTHOOD WITH DISMAL PROGNOSIS

Abstract Spinal ependymoma (EPN) with MYCN-amplification (SP-EPN-MYCN) was recently recognized as a distinct tumor type within the 2021 WHO classification of CNS tumors. To date, information about the epidemiological, clinical, and biological features of SP-EPN-MYCN remain sparse. Using DNA-methylation profiling, we collected a cohort of n = 71 (54 primaries, 17 relapses) SP-EPN-MYCNs. Based on the Heidelberg Methylation Brain Tumor Classifier V12.5, 95.6% (68/71) reached a classifier score of ≥ 0.9. In 68/71 cases, focal amplifications of MYCN were detected by copy number variation (CNV) analysis. Two tumors of pediatric patients harbored MYC-amplifications. Other repeated CNVs amongst the primary cases included gains of chromosomes 5 (5/54) and 18 (5/54), loss of chromosome 10 (14/51), and combined 17p loss and 17q gain (5/54). FISH (n = 17) and immunohistochemistry (n = 19) provided positive results for the detection of the MYCN-amplification. Epidemiological data was available for 63% (34/54) of all patients, showing an even sex distribution (17 female patients). Twenty-four percent (8/34) were children and adolescents younger than 20 years. The rest of the cohort consisted of an AYA-population with an overall median age of 31.5 years. Primary lesions were predominantly located in the thoracic (72%) and/or cervical (55%) spinal cord. Two patients showed extra-CNS metastasis to the humerus and the paraspinal musculature, respectively. At the point of analysis, 38% (10/26) of all patients had died, and 10/16 alive patients had already developed disseminated disease. Of patients with available information, 90% (18/20) had relapsed, whereas 10% (2/20) showed stable disease with leptomeningeal spread. Median PFS (n = 17) and OS (n = 25) were 15,5 (12 – 34) and 85 (61 – NA) months. In summary, SP-EPN-MYCN is a disease with dismal prognosis that can disseminate outside of the CNS and affects mainly children and AYA. Currently, comprehensive multi-omics analyses are ongoing.

Methylation classifier array and classification of solitary fibrous tumors.

e23522 Background: Solitary fibrous tumors (SFTs) are fibroblastic tumors which carry a characteristic NAB2::STAT6 gene rearrangement. SFTs represent a heterogenous group that may arise in multiple locations and have variable risk of progression. Despite a consistent oncogenic driver, there is little biological data explaining this variability. Recently, the German Cancer Research Network (DKFZ) have modified their brain tumor methylation classifier for use in sarcomas. We subjected solitary fibrous tumors diagnosed at our institution to a validated, in-house sarcoma methylation array classifier modeled after that of the DKFZ in order to assess for differences in methylation patterns that may account for biological or clinical variances. Methods: Twenty-eight primary site SFTs from 2011-2021 were subjected to the sarcoma methylation classifier. Histologic diagnosis was confirmed by STAT6 immunohistochemistry or by the presence of the NAB2::STAT6 fusion. The methylation classifier was run on the Illumina iScan platform and the resulting IDAT files were analyzed via a custom bioinformatics pipeline. The sarcoma classifier was trained using the same samples and normalization strategies used by the DKFZ group and validated using the same publicly available samples. Results: Methylation data for each tumor was collected, given a classification score, and visualized by t-distributed stochastic neighbor embedding (t-SNE). We noted three distinct groups of SFTs on the t-SNE plot. Group 1 was comprised of all eight intracranial SFTs and were correctly classified as SFT (classifier score ≥0.9). Group 2 contained nine pleural-based tumors and one SFT each from the mediastinum and bladder wall. Classification was successful in only seven of eleven tumors. Group 3 SFTs predominantly arose in the soft tissues or bone (two pelvis, four extremity, and one mediastinum) and all failed accurate classification. Finally, two orbital SFTs stratified separately from all groups on t-SNE plot but did achieve an appropriate classification. There were no significant differences in grade or risk stratification among the three groups. Conclusions: Our preliminary data show differential clustering based on tumor location, suggesting biological differences between intracranial, pleural/visceral, and connective tissue locations. These results highlight the heterogeneity of these tumors and the inadequacy of the current methylation classification scheme for SFT. Future directions will include analysis of additional cases, gene expression patterns, and outcome data.

Methylation-based prediction of myelodysplastic syndrome survival outcomes.

7058 Background: Myelodysplastic syndromes (MDS) are a heterogeneous group of myeloid malignancies associated with a myriad of deleterious outcomes and a 5-year relative survival of 37%. MDS risk stratification is key for optimal treatment decisions. DNA methylation is associated with MDS biology due to frequent somatic mutations in genes (i.e. TET2, DNMT3A, IDH1, IDH2, and WT1) that affect DNA methylation. We hypothesized that methylation-based markers may help stratify risk and improve IPSS-R (the Revised International Prognostic Scoring System). Here, we evaluated the potential of both Bone Marrow (BM) Whole-genome Bisulfite Sequencing (WGBS) and Serum-based Targeted Methylation (TM) Sequencing to predict survival of MDS patients. Methods: We conducted paired BM WGBS and Serum-based TM sequencing on a cohort of 127 patients with MDS (N = 104) and secondary AML (N = 23) treated at the Columbia University Medical Center. We categorized patients based on an overall survival of &gt; 3 (“long”, N = 50) or &lt; 3 years (“short”, N = 77). We then identified differentially methylated regions (DMRs) differing between long and short survivor groups, and analyzed the biological pathways and functions enriched in these regions. To assess survival association, we trained a random forest classifier on principal components (“methylation classifier”) on the methylation fractions from a subset of 96 Serum and 98 BM MDS patient samples to predict long vs. short survival, and compared this against a logistic regression IPSS-R model. We optimized the hyperparameters and assessed classifier performance using six-fold nested cross-validation. Finally, we conducted a multivariable cox regression with predicted scores from the methylation classifier, IPSS-R scores (recalculated at sample collection), age, and gender. Results: We identified a total of 7,742 DMRs in the BM WGBS and 14,093 DMRs in the Serum TM as significantly different between long and short survivor groups. Signaling pathways for calcium, cAMP, MAPK, Rap1, and PI3K-Akt were significantly enriched in DMRs. We also identified a previously-reported CpG island hypermethylation signature associated with AML progression in BM and Serum. The methylation classifier achieved a comparable AUROC compared to IPSS-R logistic regression for both BM WGBS (0.80 vs 0.78) and Serum TM (0.77 vs 0.73) samples. Multivariable Cox regression demonstrated that both predicted scores from the methylation classifier (Serum p =0.002; BM p =0.016) and IPSS-R scores (Serum p =0.003; BM p =0.004) are significant predictors of survival. Conclusions: We demonstrated methylation-based sequencing represents a promising new tool for MDS patient risk stratification. As IPSS-R relies on bone marrow aspirates, the Serum-based TM assay offers a less-invasive method for MDS risk stratification.

Longitudinal transcriptome profiling of localized hormone-sensitive tumors in treatment-naïve ENACT patients with prostate cancer with and without enzalutamide (ENZA).

5026 Background: The ENACT trial (NCT02799745) randomized men with low- or intermediate-risk prostate cancer to 12 months of ENZA monotherapy followed by active surveillance (AS) for 12 months or to AS alone. Men were followed longitudinally, including by serial genome-wide expression profiling. Of 258 genomic signatures evaluated from baseline samples, increased Decipher genomic classifier score was prognostic for progression on AS; androgen receptor activity (AR-A) and PAM50 subtypes were associated with clinical benefit from ENZA, as previously reported. Here we present transcriptomic changes over time for men undergoing AS and those treated transiently with ENZA. Methods: Gene-expression profiling was performed via the Decipher GRID platform (Veracyte, Inc.) on 131 ENACT patient samples (ENZA, n=57; AS, n=74) collected at prespecified time points (screening, 12 months, and 24 months). Predefined GRID signatures were assessed for correlation with biological and clinical characteristics, including immune activity, treatment sensitivity, metastatic risk, and molecular subtypes. Statistical analysis was conducted by Cox proportional hazards models and logistic/linear regression; longitudinal changes were assessed by linear mixed-effects models. Results: Analysis of 12-month longitudinal samples showed significant transcriptomic changes in patients treated with ENZA, including markers of AR-A downregulation, decreased regulatory T-cell suppression, and increased immune activation. Many of the transcriptomic signature changes that occurred during ENZA treatment, including immune marker changes, were transient and returned to baseline values by 24 months following ENZA treatment discontinuation at 12 months. Conclusions: This exploratory biomarker analysis provides valuable data on transcriptomic changes in patients undergoing AS and those exposed to ENZA followed by cessation of treatment. Longitudinal analysis data provided insights into progression of early disease in the AS arm and possible strategies for extending ENZA activity beyond progression. Immune modulation with ENZA treatment may provide further insights for exploration of immunotherapy strategies. Clinical trial information: NCT02799745 .