Beam selection for joint transmission in cell-free massive multi-input multi-output systems faces the problem of extremely high training overhead and computational complexity. The traffic-aware quality of service additionally complicates the beam selection problem. To address this issue, we propose a traffic-aware hierarchical beam selection scheme performed in a dual timescale. In the long-timescale, the central processing unit collects wide beam responses from base stations (BSs) to predict the power profile in the narrow beam space with a convolutional neural network, based on which the cascaded multiple-BS beam space is carefully pruned. In the short-timescale, we introduce a centralized reinforcement learning (RL) algorithm to maximize the satisfaction rate of delay w.r.t. beam selection within multiple consecutive time slots. Moreover, we put forward three scalable distributed algorithms including hierarchical distributed Lyapunov optimization, fully distributed RL, and centralized training with decentralized execution of RL to achieve better scalability and better tradeoff between the performance and the execution signal overhead. Numerical results demonstrate that the proposed schemes significantly reduce both model training cost and beam training overhead and are easier to meet the user-specific delay requirement, compared to existing methods.

We report a personalized tumor-informed technology, Patient-specific pROgnostic and Potential tHErapeutic marker Tracking (PROPHET) using deep sequencing of 50 patient-specific variants to detect molecular residual disease (MRD) with a limit of detection of 0.004%. PROPHET and state-of-the-art fixed-panel assays were applied to 760 plasma samples from 181 prospectively enrolled early stage non-small cell lung cancer patients. PROPHET shows higher sensitivity of 45% at baseline with circulating tumor DNA (ctDNA). It outperforms fixed-panel assays in prognostic analysis and demonstrates a median lead-time of 299 days to radiologically confirmed recurrence. Personalized non-canonical variants account for 98.2% with prognostic effects similar to canonical variants. The proposed tumor-node-metastasis-blood (TNMB) classification surpasses TNM staging for prognostic prediction at the decision point of adjuvant treatment. PROPHET shows potential to evaluate the effect of adjuvant therapy and serve as an arbiter of the equivocal radiological diagnosis. These findings highlight the potential advantages of personalized cancer techniques in MRD detection.

Background: Through parallel testing and comparison of personalized and fixed panel minimal residual disease (MRD) assays, to establish the best technique and application strategy of dynamic MRD detection for prognosis prediction and disease assessment among non-small cell lung cancer (NSCLC) patients. Method: We analyzed 760 plasma samples from prospectively enrolled 181 patients with NSCLC recruited to the MEDAL study (NCT03634826), with disease stage I (63%), II (19%) and III (18%). 80% were adenocarcinomas. Plasma samples were collected at baseline (n=157), landmark 3-day and 1-month (n=334), and longitudinal points (n=248) were analyzed. Additional plasma was collected after relapse for 14 patients (n=21). Median follow-up was 1092 days, and 48 patients progressed. We employed a novel personalized tumor-informed technology named PROPHET using deep sequencing of 50 patient-specific variants. The PROPHET was developed to detect MRD with a limit of detection (LoD) of 0.004% and sample-level specificity of greater than 99% in the analytical validation. Detection and quantification of MRD through tumor-informed (TI) and tumor-agnostic (TA) fixed panel assays in the same samples were conducted for a head-to-head comparison. Results: ctDNA was detected by PROPHET prior to treatment in 45% of samples (83%, 75% and 23% for disease stage III, II and I), and showed a higher positive rate than the TI and TA assays (22% and 19%). PROPHET identified 30 more ctDNA positive patients with a median ctDNA fraction of 0.01% at baseline. From the landmark single test, the sensitivity was 45%; integrating longitudinal time points increased the sensitivity to 85%. Landmark PROPHET status was the only risk factor other than clinical features to predict the clinical relapse (p<0.001, multivariate Cox model). MRD positive patients defined by non-canonical variants (n=8) had similar disease-free survival (DFS) as MRD positive patients defined by canonical variants (n=9). Landmark PROPHET-based MRD status combined with clinical TNM stage outperformed TNM stage for prediction of prognosis (p<0.001). Longitudinal MRD achieved negative predictive value (NPV) of 99% with an interval of 150 days, and demonstrated 299 days of longer lead-time than other state-of-the-art fixed-panel assays. Among 16 patients with equivocal radiological diagnosis, all the MRD positive patients relapsed (n=6). Among relapsed patients received next-line treatments, 7 ctDNA negative patients survived or died other disease, 67% (2/3) ctDNA positive patients died from cancer. Sensitivity for bone and brain metastasis was 100% (11/11) and 50% (2/4), respectively. Conclusion: The sensitive tumor-informed personalized MRD approach could provide advantages in prognosis prediction at landmark and disease assessment during surveillance for NSCLC patients. Citation Format: Kezhong Chen, Chenyang Wang, Haifeng Shen, Xi Li, Yichen Jin, Shuailai Wu, Fujun Qiu, Qiang Lu, Di Peng, Shuai Fang, Bing Li, Juan Lv, Jinlei Song, Yang Wang, Shannon Chuai, Zhihong Zhang. Individualized tumor-informed circulating tumor DNA (ctDNA) analysis for postoperative monitoring of non-small cell lung cancer (NSCLC) - the MEDAL study [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2023; Part 1 (Regular and Invited Abstracts); 2023 Apr 14-19; Orlando, FL. Philadelphia (PA): AACR; Cancer Res 2023;83(7_Suppl):Abstract nr 1039.

Background The feasibility of DNA methylation-based assays in detecting minimal residual disease (MRD) and postoperative monitoring remains unestablished. We aim to investigate the dynamic characteristics of cancer-related methylation signals and the feasibility of methylation-based MRD detection in surgical lung cancer patients. Methods Matched tumor, tumor-adjacent tissues, and longitudinal blood samples from a cohort (MEDAL) were analyzed by ultra-deep targeted sequencing and bisulfite sequencing. A tumor-informed methylation-based MRD (timMRD) was employed to evaluate the methylation status of each blood sample. Survival analysis was performed in the MEDAL cohort ( n = 195) and validated in an independent cohort (DYNAMIC, n = 36). Results Tumor-informed methylation status enabled an accurate recurrence risk assessment better than the tumor-naïve methylation approach. Baseline timMRD-scores were positively correlated with tumor burden, invasiveness, and the existence and abundance of somatic mutations. Patients with higher timMRD-scores at postoperative time-points demonstrated significantly shorter disease-free survival in the MEDAL cohort (HR: 3.08, 95% CI: 1.48–6.42; P = 0.002) and the independent DYNAMIC cohort (HR: 2.80, 95% CI: 0.96–8.20; P = 0.041). Multivariable regression analysis identified postoperative timMRD-score as an independent prognostic factor for lung cancer. Compared to tumor-informed somatic mutation status, timMRD-scores yielded better performance in identifying the relapsed patients during postoperative follow-up, including subgroups with lower tumor burden like stage I, and was more accurate among relapsed patients with baseline ctDNA-negative status. Comparing to the average lead time of ctDNA mutation, timMRD-score yielded a negative predictive value of 97.2% at 120 days prior to relapse. Conclusions The dynamic methylation-based analysis of peripheral blood provides a promising strategy for postoperative cancer surveillance. Trial registration This study (MEDAL, ME thylation based D ynamic A nalysis for L ung cancer) was registered on ClinicalTrials.gov on 08/05/2018 (NCT03634826). https://clinicaltrials.gov/ct2/show/NCT03634826 .

213 Background: Identifying molecular residual disease (MRD) with tailored tumor-informed ctDNA-based next-generation sequencing (NGS) assays after curative surgery could facilitate the individualized management of resected colorectal cancer (CRC) patients. Here, we prospectively evaluated the clinical performance of tumor-informed ctDNA mutation analysis using a novel Burning Rock Patient-specific Prognostic and Potential Therapeutic Marker Tracking (brPROPHET) approach for assessing MRD in resected CRC patients. Methods: The brPROPHET assay was designed to track patient-specific somatic variants based on whole-exome sequencing (WES) of the tumor tissue and matched white blood cells (WBCs). Fixed panel with informed calling (FI) and fixed panel with agnostic calling (FA) assays were performed in a subset of patients with a 168-gene panel spanning 273 kb of the human genome for a head-to-head comparison. Results: A total of 117 patients (stage II/III 53[45.0%]/41[35.0%]) were analyzed. 60 (51.0%) patients were treated with adjuvant therapy after surgery. brPROPHET assay was designed to target up to 55 variants per patient. Only 6% (344/5835) of designed variants were included in the fixed panels. 75% (2908/3886) of genes selected for panel design were private to a specific patient. Preoperative ctDNA was detected in 97% (113/117) of the patients with 88% (14/16), 98% (52/53), 98% (40/41), and 100% (7/7) in stage I, II, III and IV, respectively. The median ctDNA levels were observed to be higher in patients with advanced stages and significantly correlated with tumor volume. MRD status was tested postoperatively on day 7 and day 30 with a positivity rate of 18% (21/117) and 15% (14/93), respectively. Due to a short follow-up period, only two patients had recurred, and ctDNA was detected prior to radiological relapse, with a lead time of 1 and 2 months, respectively. Among 74 patients enrolled for parallel comparing three MRD assays, preoperative ctDNA was detected in 97.3%, 75.7%, and 68.9% of patients with brPROPHET, FI and FA fixed panel assays, respectively. 20.3% (15/74) of patients had baseline ctDNA captured by brPROPHET assay only. The ctDNA levels of these patients (median:0.3 mean tumor molecules [MTM] / milliliter [mL]) were lower than those captured by FI/FA fixed panels (median:3.0 MTM/mL, p<0.05). A total of 135 postoperative blood samples were tested by all three assays, the positive rates with brPROPHET, FI and FA fixed panel assays were 14.8%, 8.1%, and 6.7% respectively. Conclusions: This initial study reported the clinical performance of the patient-specific brPROPHET assay in CRC, demonstrating superior sensitivity in detecting preoperative and postoperative ctDNA than the fixed panel assays. The ongoing study including correlation with clinical outcomes and serial testing will be presented.

Background: Circulating tumor DNA (ctDNA)-based molecular residual disease (MRD) detection has been shown to predict postoperative relapse in several cancers. Currently, fixed and personalized panel-based tumor-informed (TI) ctDNA testing strategies are available for MRD assessment. Here, we aim to evaluate the prognostic value of MRD in resectable gastric cancer (GC) and assess MRD with fixed and personalized panels for comparison. Methods: We prospectively enrolled patients with stage I-III GC who underwent R0 resection at Ruijin Hospital. Blood samples were collected on the day of surgery preoperatively and during the second-fourth week postoperatively. Tissue samples were collected shortly thereafter at the time of curative-intent resection. Both the Patient-specific pROgnostic and Potential tHErapeutic marker Tracking (brPROPHET, a personalized panel-based TI assay, Burning Rock Biotech) and a 168-gene panel-based TI assay (Burning Rock Biotech) were used to quantify ctDNA. Results: A total of 55 patients (19 females and 36 males) with a median age of 65 years (range: 35-87) were enrolled. Precise 5 and 41 patients were treated with neoadjuvant therapy and adjuvant therapy, respectively. The median duration of follow-up was 784 days (range: 582-853). All patients underwent the fixed panel-based TI assay. Of which, 19 patients were collected with postoperative blood samples. The MRD status of postoperative blood by the fixed panel showed a significant association with recurrence-free survival (RFS, MRD-positive [MRD+] vs. MRD-negative [MRD-]: 484 days vs. not reached [NR], hazard ratio [HR] = 8.0, p = 0.02). Of the 55 enrolled patients, 19 patients underwent both personalized and fixed panel-based TI assays. For the personalized panel, preoperative ctDNA was detected in 94.7% (18/19) of the patients with 100% (4/4), 66.7% (2/3), and 91.7% (11/12) in stage I, II, and III, respectively. The numbers were 36.8% (7/19), 0% (0/4), 0% (0/3), and 58.3% (7/12) with the fixed panel, respectively. Among the 19 patients, three of the 13 patients who had postoperative blood samples developed tumor recurrence during the follow-up visit. The sensitivity of the personalized panel for predicting tumor recurrence achieved 100% (3/3) with 90% (9/10) specificity. The sensitivity of the fixed panel was 33.3% (1/3) with 100% (10/10) specificity. In addition, patients with MRD+ by the personalized panel had a significantly shorter RFS than those with MRD- (566 days vs. NR, HR > 100, p = 0.003). Conclusion: Our study indicates that MRD+ by either the personalized or fixed panel was associated with an unfavorable survival outcome. ctDNA-based MRD detection by the personalized panel-based TI assay might be a powerful prognostic biomarker to identify GC patients who underwent radical resection at a higher risk of relapse or with an unfavorable survival outcome. Citation Format: Pei Xue, Yanfei Shao, Xueliang Zhou, Haiyan Li, Yang Wang, Chenyang Wang, Hao Zhang, Bing Li, Shuo Shi, Haiwei Du, Jing Sun. Circulating tumor DNA-based molecular residual disease predicts relapse in patients with resectable gastric cancer [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2023; Part 1 (Regular and Invited Abstracts); 2023 Apr 14-19; Orlando, FL. Philadelphia (PA): AACR; Cancer Res 2023;83(7_Suppl):Abstract nr 1037.

Facial age estimation aims to rank the face aging data by taking in the correlation among age categories. Conventional age estimation models are trained based on assumed high-quality training annotations in a totally-supervised manner. However, noisy data in a sparse distribution collected from unconstrained environment usually account for the corruption of produced gradients and ordinal relationships, which may fail to fairly describe the correlated face aging data. In this paper, we propose a meta-set learning (MSL) approach for exploiting the unfairness of face aging datasets, thus achieving unbiased age classification in unconstrained conditions. To address this, we elaborately create an unfairness filtration network under the meta-learning paradigm, which exceeds a reliable margin-reweighting initialization suffering from class variance, simultaneously exploiting the meta-reweighting intervention to minimize the training bias caused by class imbalance. Moreover, our proposed model leverages the learned instance-level margin between logits and develops a unimodal constrained logits loss, further surviving age regression models from unfairness. Experimental results on multiple in-the-wild datasets demonstrate that our proposed method achieves superior results compared to existing state-of-the-art methods.

Background: Identifying molecular residual disease (MRD) with tailored tumor-informed ctDNA based next-generation sequencing (NGS) assays after curative surgery could facilitate the individualized management of resected CRC patients. Here, we investigated the clinical utility of tumor-informed ctDNA mutation analysis using a novel Patient-specific pROgnostic and Potential tHErapeutic marker Tracking (PROPHET) approach for accessing MRD in resected CRC patients. Using the same set of baseline and post-operative blood samples, we compared the performance of PROPHET assay with tumor-naïve fixed panel for detecting MRD and predicting recurrence in resected CRC patients. Methods: The prospective study recruited 42 patients diagnosed with stage I-III CRC from May 2019 to Jun 2020 at the First Affiliated Hospital of Soochow University. Tumor tissue samples were collected at surgery. Blood samples collected before surgery (baseline), 8-day post-operative time point before any adjuvant therapy were analyzed. The detection and quantification of ctDNA for MRD assessment was investigated using PROPHET, a personalized, tumor-informed ctDNA assay designed to track up to 50 top-ranked patient-specific somatic variants based on whole-exome sequencing (WES) of the tumor tissue and matched white blood cells (WBCs). Tumor- naïve fixed assay was performed using targeted NGS panel, containing 41 gastrointestinal cancer-related genes. Results: Baseline ctDNA status was detected in 95.23% (40/42) of the patients with PROPHET assay, and 69.05% (29/42) of the patients with fixed panel. Of 42 patients included in the analysis, 1, 25, and 16 patients had pathological stages I, II, and III CRC with baseline ctDNA detected in 100% (1/1), 92% (23/25), 100% (16/16) patients with PROPHET assay, and 100% (1/1), 64% (16/25), 75% (12/16) patients with fixed panel. Post-operative ctDNA-positive status with PROPHET assay was associated with 3-year DFS, compared with ctDNA-negative group (hazard ratio [HR], 16.57, 95% confidence interval [CI]:3.01-91.36, p=0.014). 15% (6/40) patients were identified to be MRD- positive and 83.33% (5/6) patients eventually relapsed at 3-years follow-up. Although fixed panel also showed high performance in predicting relapse HR, 4.48, 95% CI:1.9-10.9; p< 0.001, 27.5% (11/40) patients were identified to be MRD-positive and only 45.45% (5/11) patients eventually relapsed. 3-year prognostication with PROPHET assay at 8-day post-operation yielded higher positive predictive value (83.33% vs 45.45%), negative predictive value (91.18% vs 89.66%), and specificity (96.88% vs 81.25%) as compared with tumor-naïve fixed panel. Conclusion: Patient-specific PROPHET assay based on unique somatic mutation profiles detects patients with high-risk of recurrence, which achieved higher specificity than tumor-naïve fixed panel. Citation Format: Jian Zhou, Jian Yang, Zixiang Zhang, Ye Li, Bin Yi, Yuchen Tang, Dechun Li, Xiaozhe Li, Di Peng, XI Li, Yang Wang, Haiyan Li, Bing Li, Chenyang Wang, Pengfei Zhu, Longfei Chen, Shuailai Wu, Shuai Fang, Chenxi Li, Fujun Qiu, Shannon Chuai, Zhihong Zhang. Patient-specific tumor-informed circulating tumor DNA (ctDNA) analysis for postoperative monitoring of patients with stages I-III colorectal cancer (CRC) [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2022; 2022 Apr 8-13. Philadelphia (PA): AACR; Cancer Res 2022;82(12_Suppl):Abstract nr 5917.

Background: Patient-specific flexible gene panels designed based on whole-exome sequencing (WES) of resected tumor tissues is a promising strategy for ctDNA-based detection of molecular residual disease (MRD) in early-stage NSCLC. Flexible gene panels could potentially overcome the limitations of fixed panels by incorporating more unique genomic regions that might be absent in fixed panels; however, no study has reported a head-to-head comparison of these two approaches in postoperative disease monitoring. In this study, we investigated the clinical utility of a novel Patient-specific pROgnostic and Potential tHErapeutic marker Tracking (PROPHET) tumor-informed ctDNA assay. Using the same set of longitudinal blood samples, we further compared the performance of PROPHET assay with tumor-informed (TI) and tumor-naïve (TN) fixed panels for predicting MRD and prognosis in surgical NSCLC patients. Methods: Fifty-three patients with stage I-III resected NSCLC from the MEDAL study (NCT03634826) with adequate samples and median follow-up of 647 days were analyzed. Matched surgical tumor tissue and blood samples collected at various time points, including before surgery (baseline), 3-days (B) and 1-month (C) postoperative time points before any adjuvant therapy and subsequent follow-up time points (F) were analyzed. PROPHET assay involved four major steps: identify somatic mutations using WES, customized design of a patient-specific panel consisting of 50 single nucleotide variants, ultra-deep unique molecular-identifier-based next-generation sequencing (UMI-NGS) of serial blood samples using the patient-specific panel, and MRD risk prediction. Fixed panel assay of serial blood samples was performed using UMI-NGS with 168 gene panel spanning 273 kb of human genome. Results: At 1-month post-surgery, PROPHET assay accurately predicted MRD-positive cases among relapsed patients (50%, 13/26), whereas all disease-free patients were MRD-negative (100%, 21/21). Three-year prognostication with PROPHET assay at B+C yielded higher sensitivity (59% vs 26% vs 22%), negative predictive value (66% vs 51% vs 50%), and hazard ratio (7.15, 95%CI [3.2-15.9] vs 4.48 [1.9-10.9] vs 5.58 [2.1-14.7]) as compared with TI and TN fixed panel assays. Disease monitoring using PROPHET assay at B/C/F accurately predicted the MRD risk in 70% (21/30) of relapsed patients at a median lead time of 318 days (range: 20-751), whereas TI assay predicted 43% (13/30) at 282 days (range 20-716) and TN assay predicted 37% (11/30) at 282 days (range: 20-634). Conclusion: Patient-specific tumor-informed ctDNA-based postoperative monitoring enables risk stratification at early postoperative settings better than fixed panel, which paves an alternative strategy in the individualized management of surgical NSCLC patients. Citation Format: Kezhong Chen, Haifeng Shen, Shuailai Wu, Pengfei Zhu, Chenyang Wang, Analyn Lizaso, Guannan Kang, Yang Wang, Juan Lv, Shuai Fang, Wenjun Wu, Fujun Qiu, Yuan Sun, Qiang Lu, Heng Zhao, Shannon Chuai, Fan Yang, Zhihong Zhang. Tumor-informed patient-specific panel outperforms tumor-naïve and tumor-informed fixed panel for circulating tumor DNA (ctDNA)-based postoperative monitoring of non-small cell lung cancer (NSCLC) [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2022; 2022 Apr 8-13. Philadelphia (PA): AACR; Cancer Res 2022;82(12_Suppl):Abstract nr 5916.

Background Aberrant DNA methylation may offer opportunities in revolutionizing cancer screening and diagnosis. We sought to identify a non-invasive DNA methylation-based screening approach using cell-free DNA (cfDNA) for early detection of hepatocellular carcinoma (HCC). Methods Differentially, DNA methylation blocks were determined by comparing methylation profiles of biopsy-proven HCC, liver cirrhosis, and normal tissue samples with high throughput DNA bisulfite sequencing. A multi-layer HCC screening model was subsequently constructed based on tissue-derived differentially methylated blocks (DMBs). This model was tested in a cohort consisting of 120 HCC, 92 liver cirrhotic, and 290 healthy plasma samples including 65 hepatitis B surface antigen-seropositive (HBsAg+) samples, independently validated in a cohort consisting of 67 HCC, 111 liver cirrhotic, and 242 healthy plasma samples including 56 HBsAg+ samples. Results Based on methylation profiling of tissue samples, 2321 DMBs were identified, which were subsequently used to construct a cfDNA-based HCC screening model, achieved a sensitivity of 86% and specificity of 98% in the training cohort and a sensitivity of 84% and specificity of 96% in the independent validation cohort. This model obtained a sensitivity of 76% in 37 early-stage HCC (Barcelona clinical liver cancer [BCLC] stage 0-A) patients. The screening model can effectively discriminate HCC patients from non-HCC controls, including liver cirrhotic patients, asymptomatic HBsAg+ and healthy individuals, achieving an AUC of 0.957(95% CI 0.939–0.975), whereas serum α-fetoprotein (AFP) only achieved an AUC of 0.803 (95% CI 0.758–0.847). Besides detecting patients with early-stage HCC from non-HCC controls, this model showed high capacity for distinguishing early-stage HCC from a high risk population (AUC=0.934; 95% CI 0.905–0.963), also significantly outperforming AFP. Furthermore, our model also showed superior performance in distinguishing HCC with normal AFP (< 20ng ml −1 ) from high risk population (AUC=0.93; 95% CI 0.892–0.969). Conclusions We have developed a sensitive blood-based non-invasive HCC screening model which can effectively distinguish early-stage HCC patients from high risk population and demonstrated its performance through an independent validation cohort. Trial registration The study was approved by the ethic committee of The Second Xiangya Hospital of Central South University (KYLL2018072) and Chongqing University Cancer Hospital (2019167). The study is registered at ClinicalTrials.gov(# NCT04383353 ).

e16153 Background: Biliary tract cancer (BTC) is an aggressive malignancy with poor prognosis and limited treatment options. The epigenome-associated multi-dimensional studies for BTC are limited. Here, we proposed a DNA methylation-based classification scheme and investigated the associations between methylation and multi-dimensional data including clinicopathological features, genetic aberrations, and prognosis. Methods: Multi-dimensional data concerning mutation, DNA methylation, and clinical data from 105 BTC patients who received surgical resection (gallbladder cancer [GBC], n=48, cholangiocarcinoma [CCA], n=57) were analyzed. Differentially methylated blocks (DMBs) in GBCs and CCAs were identified by comparing tumors and adjacent tissues. Methylation-based subtyping was performed via non-supervised consensus clustering. Results: The differentially methylated blocks (DMBs) in GBCs and CCAs are highly overlapping. Based on the common DMBs of GBCs and CCAs, the classifier yielded high sensitivity of 95.2% and specificity of 96.0%. Hypomethylated genes were enriched in pathways of transmemberane receptor and ion binding, while hypermethylation occurred in genes concerning DNA binding transcription activity. By non-supervised clustering of common DMBs, 6 clusters with different degrees of methylation change were identified. Higher methylation alteration (cluster risk-high) was associated with more copy number variation and shorter OS. By LASSO regression, a 12-gene prognostic model was trained and validated (Table). As for immune characteristics, the methylation of CD8A gene was lower in the cluster risk-high group, and this association was validated in the TCGA-cholangiocarcinoma cohort. In addition, lower methylation change was associated with higher BCR/TCR diversity, immune cell infiltration, and PD-L1 and CTLA4 mRNA expression. Conclusions: In BTC, methylation patterns may serve as a robust indicator of immune-related features and prognosis. Our integrative analysis highlights the association between genomic and epigenomic features, and provides insights into the molecular heterogeneity, and the potential therapeutic significance in biliary tract malignancies. Included methylation sites in the LASSO model.[Table: see text]

The low abundance of circulating tumour DNA (ctDNA) in plasma samples makes the analysis of ctDNA biomarkers for the detection or monitoring of early-stage cancers challenging. Here we show that deep methylation sequencing aided by a machine-learning classifier of methylation patterns enables the detection of tumour-derived signals at dilution factors as low as 1 in 10,000. For a total of 308 patients with surgery-resectable lung cancer and 261 age- and sex-matched non-cancer control individuals recruited from two hospitals, the assay detected 52–81% of the patients at disease stages IA to III with a specificity of 96% (95% confidence interval (CI) 93–98%). In a subgroup of 115 individuals, the assay identified, at 100% specificity (95% CI 91–100%), nearly twice as many patients with cancer as those identified by ultradeep mutation sequencing analysis. The low amounts of ctDNA permitted by machine-learning-aided deep methylation sequencing could provide advantages in cancer screening and the assessment of treatment efficacy.

Background: An effective blood-based method for hepatocellular carcinoma (HCC) early detection has not yet been developed. Aberrant DNA methylation may offer opportunities in revolutionizing cancer screening and diagnosis. We sought to identified a non-invasive DNA methylation-based diagnostic approach using cell-free DNA for HCC early detection in high risk population.Methods: Differentially methylated DNA methylation blocks were determined by comparing methylation profiles of biopsy-proven HCC, cirrhotic and healthy liver tissues using targeted bisulfite sequencing. A multi-layer HCC screening model was subsequently constructed based on tissue-derived differentially methylated makers using plasma samples from 120 HCC patients, 92 cirrhotic patients and 290 healthy individuals. This model was independently validated in a cohort consisting of 67 HCC patients, 115 cirrhotic patients and 242 healthy individuals. The clinical parameters of misclassified samples were also investigated. Findings: Based on methylation profiling of tissue samples, a total of 2,321 DNA methylation markers were derived, which were subsequently used to construct a cfDNA-based HCC screening model, achieving a sensitivity of 86% and specificity of 97% in the training cohort and a sensitivity of 82% and specificity of 96% in the independent validation cohort. The HCC screening model can effectively discriminate HCC patients from non-HCC patients, including patients with cirrhosis, HBV infection and healthy individuals, achieving an AUC of 0.956; whereas serum alpha-fetoprotein (AFP) achieved an AUC of 0.803.Interpretation: We have developed and validated an HCC screening model which can effectivelydistinguish early stage HCC patients from cirrhotic patients and symptomatic HBsAg-seropositiveindividuals.Funding: Key Research and Development Program of Hunan Province, Natural ScienceFoundation of Hunan Province, China.Declaration of Interest: None to declare. Ethical Approval: The study was approved by the ethic committee of The Second Xiangya Hospital of Central South University (KYLL2018072) and Chongqing University Cancer Hospital (2019167). All collection and usage of human sample and clinical data were in accordance with the principles of the Declaration of Helsinki. Written informed content was obtained from every participant for the use of their tissue or plasma samples.

In this paper, we propose a progressive margin loss (PML) approach for unconstrained facial age classification. Conventional methods make strong assumption on that each class owns adequate instances to outline its data distribution, likely leading to bias prediction where the training samples are sparse across age classes. Instead, our PML aims to adaptively refine the age label pattern by enforcing a couple of margins, which fully takes in the in-between discrepancy of the intra-class variance, inter-class variance and class center. Our PML typically incorporates with the ordinal margin and the variational margin, simultaneously plugging in the globally-tuned deep neural network paradigm. More specifically, the ordinal margin learns to exploit the correlated relationship of the real-world age labels. Accordingly, the variational margin is leveraged to minimize the influence of head classes that misleads the prediction of tailed samples. Moreover, our optimization carefully seeks a series of indicator curricula to achieve robust and efficient model training. Extensive experimental results on three face aging datasets demonstrate that our PML achieves compelling performance compared to state of the art. Code will be made publicly.

Early detection of colorectal carcinoma (CRC) would help to identify tumors when curative treatments are available and beneficial. However, current screening methods for CRC, e.g., colonoscopy, may affect patients’ compliance due to the uncomfortable, invasive and time-consuming process. In recent decades, methylation profiles of blood-based circulating tumor DNA (ctDNA) have shown promising results in the early detection of multiple tumors. Here we conducted a study to investigate the performance of ctDNA methylation markers in early detection of CRC. In total, 742 participants were enrolled in the study including CRC (n = 332), healthy control (n = 333), benign colorectal disease (n = 65) and advanced adenoma (n = 12). After age-matched and randomization, 298 participants (149 cancer and 149 healthy control) were included in training set and 141 (67 cancer and 74 healthy control) were in test set. In the training set, the specificity was 89.3% (83.2–93.7%) and the sensitivity was 88.6% (82.4–93.2%). In terms of different stages, the sensitivities were 79.4% (62.1–91.2%) in patients with stage I, 88.9% (77.3–95.8%) in patients with stage II, 91.4% (76.9–98.2%) in patients with stage III and 96.2% (80.3–99.9%) in patients with stage IV. Similar results were validated in the test set with the specificity of 91.9% (83.1–97.0%) and sensitivity of 83.6% (72.5–91.6%). Sensitivities for stage I-III were 87.0% (79.7–92.4%) in the training set and 82.5% (70.2–91.3%) in the test set, respectively. In the unmatched total population, the positive ratios were 7.8% (5.2–11.2%) in healthy control, 30.8% (19.9–43.5%) in benign colorectal disease and 58.3% (27.5–84.7%) in advanced adenoma, while the sensitivities of stage I–IV were similar with training and test sets. Compared with methylated SEPT9 model, the present model had higher sensitivity (87.0% [81.8–91.2%] versus 41.2% [34.6–48.1%], P < 0.001) under comparable specificity (90.1% [85.4–93.7%] versus 90.6% [86.0–94.1%]). Together our findings showed that ctDNA methylation markers were promising in the early detection of CRC. Further validation of this model is warranted in prospective studies.

Current NCCN guidelines do not recommend the use of adjuvant chemotherapy for stage IA lung adenocarcinoma patients with R0 surgery. However, 25% to 40% of patients with stage IA disease experience recurrence. Stratifying patients according to the recurrence risk may tailor adjuvant therapy and surveillance imaging for those with a higher risk. However, prognostic markers are often identified by comparing high-risk and low-risk cases which might introduce bias due to the widespread interpatient heterogeneity. Here, we developed a scoring system quantifying the degree of field cancerization in adjacent normal tissues and revealed its association with disease-free survival (DFS). Methods: We recruited a cohort of 44 patients with resected stage IA lung adenocarcinoma who did not receive adjuvant therapy. Both tumor and adjacent normal tissues were obtained from each patient and subjected to capture-based targeted genomic and epigenomic profiling. A novel methylome-based scoring system namely malignancy density ratio (MD ratio) was developed based on 39 patients by comparing tumor and corresponding adjacent normal tissues of each patient. A MD score was then obtained by Wald statistics. The correlations of MD ratio, MD score, and genomic features with clinical outcome were investigated. Results: Patients with a high-risk MD ratio showed a significantly shorter postsurgical DFS compared with those with a low-risk MD ratio (HR=4.47, P=0.01). The MD ratio was not associated with T stage (P=1), tumor cell fraction (P=0.748) nor inflammatory status (p=0.548). Patients with a high-risk MD score also demonstrated an inferior DFS (HR=4.69, P=0.039). In addition, multivariate analysis revealed EGFR 19 del (HR=5.39, P=0.012) and MD score (HR= 7.90, P=0.01) were independent prognostic markers. Conclusion: The novel methylome-based scoring system, developed by comparing the signatures between tumor and corresponding adjacent normal tissues of individual patients, largely minimizes the bias of interpatient heterogeneity and reveals a robust prognostic value in patients with resected lung adenocarcinoma.

e15076 Background: Colorectal cancer (CRC) develops as a result of neoplastic progression, which often takes decades, providing a window for early detection. Unfortunately, there has been little success in developing blood-based screening method due to the low amount of ctDNA present in the circulation, especially in patients with early stage disease. The role of aberrant DNA methylation, occurring very early in tumorigenesis, has been well elucidated. In this prospective study, we evaluated the potentiality of DNA methylation status obtained from ctDNA as an early detection method. Methods: Panel Design: Methylation data of tumor samples (12 types, n = 4,772), adjacent normal (8 types, n = 411), and normal white blood cells (n = 656) from TCGA and GSE were compared. Differentially methylated sites were extracted using modified wald-test with an adjusted p-value < 0.05 and fold-change > 2. Our panel covers 80,672 CpG sites, spanning 1.05Mb of human genome. We performed targeted bisulfite sequencing on plasma samples of 67 (stage I: 13, II:29, III: 23, IV: 2) Chinese CRC patients and 144 healthy individuals to construct a model for deriving markers that are differentially methylated and their associated weight. The model was validated in 2 independent cohorts. Results: We constructed a model using a support vector machine (SVM)-based machine learning classifier based on top 4,000 differentially methylated regions (DMRs) selected by random forest between tumor and normal plasma samples. Subsequently, 5-fold cross-validation with 100-time repeats were performed to gain a robust estimation of model performance, achieving a sensitivity of 91%, specificity of 98% and area under curve (AUC) of 98.6%. The model was subsequently validated in 2 independent cohorts: one consisted of 57 stage I-III CRC patients and 74 healthy individuals and another one with 47 stage IV patients and the same 74 healthy individuals. The model yielded a sensitivity of 83% and 95% for the early and late stage cohorts, respectively. A specificity of 95% was obtained for both cohorts. Conclusions: Our findings demonstrated the potential of profiling DNA methylation, which can effectively distinguish cancerous from healthy, for the purpose of screening. This method has potential to serve as a supplementary or alternative approach in early detection.

Introduction: Lung cancer found at an early stage carries much-improved prognosis. Unfortunately, there has been little success in developing blood-based diagnostic method. Profiling somatic mutations from ctDNA has shown great promise for cancer diagnosis, prognosis, and surveillance. Despite the substantial advances made in ctDNA detection techniques, the detection rate remains low for early stage disease. The role of aberrant DNA methylation in the process of tumorigenesis both at individual genes and a genome-wide scale has been well elucidated. It occurs very early in cancer development, thus capable of serving as a diagnostic biomarker. In this prospective study, we evaluated the potentiality of DNA methylation status obtained from ctDNA as an early diagnostic marker for NSCLC.Methods: Panel Design: Methylation data of tumor samples (12 types, n=4772), adjacent normal (8 types, n=411), and normal white blood cells (WBC, n=656) from TCGA and GSE were compared. Differentially methylated sites were extracted using modified wald-test with an adjusted p-value <0.05 and fold-change>2. Our panel covers 80,672 CpG sites, spanning 1.05Mb of human genome. We performed targeted bisulfite sequencing on plasma samples of 359 early stage Chinese NSCLC patients and 144 healthy individuals to interrogate their methylation statuses with an average sequencing depth of 1,000x. TheResults: The training cohort consisted of 359 early stage Chinese lung cancer patients (266 stage IA, 25 stage IB, 43 stage II and 25 stage III) with a median age of 51 and 144 Chinese healthy individuals with a median age of 59. We constructed a diagnostic classification model using a support vector machine (SVM)-based machine learning classifier based on top 3,000 differentially methylated regions (DMRs) selected by random forest between tumor and normal plasma samples. Subsequently, 5-fold cross-validation with 100-time repeats were performed to gain a robust estimation of model performance, achieving a sensitivity of 79.4%, specificity of 95.4% and area under curve (AUC) of 95.2%. The model was subsequently validated in an independent cohort, consisting of 79 early stage Chinese NSCLC patients and 74 healthy individuals with comparable clinical characteristics as the patients in the training cohort. The model yielded a sensitivity of 77% and specificity of 90% in the validation cohort, suggesting its robustness.Conclusions: Overall, our findings demonstrated in a large clinical cohort the potential of profiling DNA methylation from ctDNA, which can effectively distinguish cancerous from healthy, for the purpose of diagnosis. This method has potential to serve as a supplementary or alternative approach in lung cancer early detection. The general concept can be further extended to other types of cancer diagnostics.Citation Format: Naixin Liang, Bingsi Li, Chenyang Wang, Tao Zheng, Jiayue Xu, Shuai Fang, Fujun Qiu, Jing Su, Lichen Zhang, Xin Lu, Miaomiao Song, Lingjian Yang, Han Han-Zhang, Xinru Mao, Hao Liu, Shanqing Li, Ke Ma, Zhihong Zhang. DNA methylation profiling from circulating tumor DNA for early diagnosis of non-small cell lung cancer [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2019; 2019 Mar 29-Apr 3; Atlanta, GA. Philadelphia (PA): AACR; Cancer Res 2019;79(13 Suppl):Abstract nr 816.

e15605 Background: Gallbladder cancer (GBC), an uncommon malignancy with a high mortality rate, is often diagnosed late due to lack of early symptoms and the relative hidden nature of the gallbladder. Despite the advancements in imaging technologies, there is no reliable screening test for GBC. The role of aberrant DNA methylation in the process of tumorigenesis both at individual genes and a genome-wide scale has been well elucidated. It occurs very early in cancer development, thus capable of serving as a screening marker. Methods: Panel Design: Methylation data of tumor samples (12 types, n = 4,772), adjacent normal (8 types, n = 411), and normal white blood cells (n = 656) from TCGA and GSE were compared. Differentially methylated sites were derived using a Bayesian hierarchical model-DSS with an adjusted p-value < 0.05. Our panel covers 80,672 CpG sites, spanning 1.05Mb of human genome. This panel contains 12,196 GBC relevant CpG sites. We performed targeted bisulfite sequencing on 23 GBC patients (6 stage II-III, 17 stage IV) and 13 patients with non-malignant gallbladder diseases (cholecystitis and gallstones). Of the 23 GBC patients, we obtained adjacent normal tissue from 7 of them. Basic clinical features such as age, gender, of patients with GBC and patients with non-malignant gallbladder diseases were comparable. Results: Among the 12,196 GBC relevant CpG sites, 10,216 sites were statistically significantly hypermethylated and 275 sites were statistically significantly hypomethylated comparing to patients with non-malignant gallbladder disease as well as adjacent normal gallbladder tissues. Subsequently, we used the derived differentially methylated CpG sites to construct a linear regression model, achieving an area under curve of 99%. Collectively, the methylation levels were comparable between tissues with non-malignant disease and adjacent normal. Interestingly, when considering the 275 hypomethylated markers alone, we observed that the methylation level of adjacent normal tissues is significantly higher than tissues with non-malignant disease. Conclusions: Collectively, our panel can effectively distinguish GBC samples from non-cancerous samples, demonstrating the potential of DNA methylation in GBC screening. Furthermore, hypomethylation markers can be used to distinguish non-malignant disease from the healthy.