Expression-based Classification Research Articles

Gene expression signature for predicting homologous recombination deficiency in triple-negative breast cancer

Triple-negative breast cancers (TNBCs) are a subset of breast cancers that have remained difficult to treat. A proportion of TNBCs arising in non-carriers of BRCA pathogenic variants have genomic features that are similar to BRCA carriers and may also benefit from PARP inhibitor treatment. Using genomic data from 129 TNBC samples from the Malaysian Breast Cancer (MyBrCa) cohort, we developed a gene expression-based machine learning classifier for homologous recombination deficiency (HRD) in TNBCs. The classifier identified samples with HRD mutational signature at an AUROC of 0.93 in MyBrCa validation datasets and 0.84 in TCGA TNBCs. Additionally, the classifier strongly segregated HRD-associated genomic features in TNBCs from TCGA, METABRIC, and ICGC. Thus, our gene expression classifier may identify triple-negative breast cancer patients with homologous recombination deficiency, suggesting an alternative method to identify individuals who may benefit from treatment with PARP inhibitors or platinum chemotherapy.

Hybridexpress: an R/Bioconductor package for comparative transcriptomic analyses of hybrids and their progenitors.

Hybridization, the process of crossing individuals from diverse genetic backgrounds, plays a pivotal role in evolution, biological invasiveness, and crop breeding. At the transcriptional level, hybridization often leads to complex nonadditive effects, presenting challenges for understanding its consequences. Although standard transcriptomic analyses exist to compare hybrids to their progenitors, such analyses have not been implemented in a software package, hindering reproducibility. We introduce hybridexpress, an R/Bioconductor package designed to facilitate the analysis, visualization, and comparison of gene expression patterns in hybrid triplets (hybrids and their progenitors). hybridexpress provides users with a user-friendly and comprehensive workflow that includes all standard comparative analyses steps, including data normalization, calculation of midparent expression values, sample clustering, expression-based gene classification into categories and classes, and overrepresentation analysis for functional terms. We illustrate the utility of hybridexpress through comparative transcriptomic analyses of cotton allopolyploidization and rice root trait heterosis. hybridexpress is designed to streamline comparative transcriptomic studies of hybrid triplets, advancing our understanding of evolutionary dynamics in allopolyploids, and enhancing plant breeding strategies. hybridexpress is freely accessible from Bioconductor (https://bioconductor.org/packages/HybridExpress) and its source code is available on GitHub (https://github.com/almeidasilvaf/HybridExpress).

Abstract PO4-14-08: A gene expression-based classifier for HER2-low breast cancer

Abstract Background: The definition of HER2-low breast cancer from clinical trials of antibody-conjugated drugs (ADCs) relies on immunohistochemistry scoring (IHC). However, in daily-practice the accuracy of IHC is hindered by inter-observer variability in assessing HER2 non-overexpressing status. Here, we aimed to identify breast cancer primary tumors with low HER2 expression by leveraging gene expression profiling. Materials and methods: A discovery approach was applied to gene expression profile of internal INT1 (n= 125) and INT2 (n= 84) datasets. We identified specific differently expressed genes (DEGs) according to HER2 IHC categories 0, 1+, 2+ and 3+. Principal Component Analysis (PCA) was used to generate a HER2-low signature whose performance was confirmed in the independent INT3 (n= 95), and TCGA and GSE20194 publicly available datasets. The association between the HER2-low signature and HER2 IHC categories was evaluated by non-parametric Kruskal-Wallis (KW) test with post hoc pair wise comparisons (i.e. contrasts); the HER2-low signature discriminatory capability was assessed by estimating the area under the receiver operating characteristic (ROC) curve (AUC) with its corresponding 95% Confidence Interval (CI). Gene Ontology and KEGG analyses were performed to enrich the DEGs for functional information. Results: A 20-gene HER2-low signature, consisting of both up-regulated (n=11) and down-regulated (n=9) genes, was computed based on the DEGs according to HER2 IHC categories as follows: 11 genes specific for the 1+ category, 8 genes for the 2+ category, and 1 gene for the HER2-low category. HER2-low signature genes were significantly enriched with lipid and steroid metabolism pathways, peptidase regulation, and humoral immune response. The HER2-low signature levels demonstrated a significant association with HER2 IHC categories (KW p-value < 0.001) and were distributed in a bell-shaped pattern across IHC categories (low values 0 and 3+; high values 1+ and 2+), effectively distinguishing HER2-low from 0 (contrast p-value < 0.001) and 3+ (contrast p-value < 0.001). Notably, the signature levels were significantly higher in tumors scored with 1+ as compared to 0 (contrast p =0.002). The HER2-low signature association with IHC categories and discriminatory capability was confirmed in the independent INT3 and TCGA datasets with higher values in HER2-low compared to both 0 and 3+. The HER2-low signature achieved an AUC value of 0.72 (95%CI 0.62-0.83) in differentiating HER2 0/3+ from HER2 1+/2+ categories, which is worth noting in light of the individual ERBB2 mRNA AUC value of 0.48 (95%CI 0.34-0.62). Conclusions: The 20-gene HER2-low signature was generated by maximizing the differences in gene expression between tumors with different HER2 status according to IHC. In contrast to the previously published single ERBB2 gene expression assessment, our data presents compelling evidence for effectively distinguishing HER2-low tumors, including those scored as 1+ from HER2-0 tumors. Our signature holds potential in selecting novel candidates for ADC therapy. Citation Format: Serena Di Cosimo, Sara Pizzamiglio, Chiara Maura Ciniselli, Valeria Duroni, Vera Cappelletti, Loris de Cecco, Maria Carmen De Santis, Rosaria Orlandi, Marilena Iorio, Marco Silvestri, Giancarlo Pruneri, Paolo Verderio. A gene expression-based classifier for HER2-low breast cancer [abstract]. In: Proceedings of the 2023 San Antonio Breast Cancer Symposium; 2023 Dec 5-9; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2024;84(9 Suppl):Abstract nr PO4-14-08.

Machine Learning–Supported Diagnosis of Small Blue Round Cell Sarcomas Using Targeted RNA Sequencing

Small blue round cell sarcomas (SBRCSs) are a heterogeneous group of tumors with overlapping morphologic features but markedly varying prognosis. They are characterized by distinct chromosomal alterations, particularly rearrangements leading to gene fusions, whose detection currently represents the most reliable diagnostic marker. Ewing sarcomas (ESs) are the most common SBRCSs, defined by gene fusions involving EWSR1 and transcription factors of the ETS gene family, while the most frequent non-EWSR1-rearranged SBRCSs harbor a CIC rearrangement. Unfortunately, the identification of CIC::DUX4 translocation events, the most common CIC rearrangement, is challenging with current methods. Here, a machine-learning approach to support SBRCS diagnosis relying on gene expression profiles measured via targeted sequencing is presented. The analyses on a curated cohort of 69 soft tissue tumors (STT) showed markedly distinct expression patterns for SBRCS subgroups. A random forest (RF) classifier trained on ES and CIC-rearranged cases predicted probabilities of being CIC-rearranged > 0.9 for CIC-rearranged-like sarcomas and < 0.6 for other SBRCS. Testing on a retrospective cohort of 1335 routine diagnostic cases identified 15 candidate CIC-rearranged tumors with a probability > 0.75, all of which supported by expert histopathological reassessment. Furthermore, the multi-gene RF classifier appeared advantageous over using high ETV4 expression alone, previously proposed as surrogate to identify CIC rearrangement. Taken together, the expression-based classifier can offer valuable support for SBRCS pathological diagnosis.

Gene Expression-Based Cancer Classification for Handling the Class Imbalance Problem and Curse of Dimensionality.

Cancer is a leading cause of death globally. The majority of cancer cases are only diagnosed in the late stages of cancer due to the use of conventional methods. This reduces the chance of survival for cancer patients. Therefore, early detection consequently followed by early diagnoses are important tasks in cancer research. Gene expression microarray technology has been applied to detect and diagnose most types of cancers in their early stages and has gained encouraging results. In this paper, we address the problem of classifying cancer based on gene expression for handling the class imbalance problem and the curse of dimensionality. The oversampling technique is utilized to overcome this problem by adding synthetic samples. Another common issue related to the gene expression dataset addressed in this paper is the curse of dimensionality. This problem is addressed by applying chi-square and information gain feature selection techniques. After applying these techniques individually, we proposed a method to select the most significant genes by combining those two techniques (CHiS and IG). We investigated the effect of these techniques individually and in combination. Four benchmarking biomedical datasets (Leukemia-subtypes, Leukemia-ALLAML, Colon, and CuMiDa) were used. The experimental results reveal that the oversampling techniques improve the results in most cases. Additionally, the performance of the proposed feature selection technique outperforms individual techniques in nearly all cases. In addition, this study provides an empirical study for evaluating several oversampling techniques along with ensemble-based learning. The experimental results also reveal that SVM-SMOTE, along with the random forests classifier, achieved the highest results, with a reporting accuracy of 100%. The obtained results surpass the findings in the existing literature as well.

A gene expression-based classifier for HER2-low breast cancer

In clinical trials evaluating antibody-conjugated drugs (ADCs), HER2-low breast cancer is defined through protein immunohistochemistry scoring (IHC) 1+ or 2+ without gene amplification. However, in daily practice, the accuracy of IHC is compromised by inter-observer variability. Herein, we aimed to identify HER2-low breast cancer primary tumors by leveraging gene expression profiling. A discovery approach was applied to gene expression profile of institutional INT1 (n = 125) and INT2 (n = 84) datasets. We identified differentially expressed genes (DEGs) in each specific HER2 IHC category 0, 1+, 2+ and 3+. Principal Component Analysis was used to generate a HER2-low signature whose performance was evaluated in the independent INT3 (n = 95), and in the publicly available TCGA and GSE81538 datasets. The association between the HER2-low signature and HER2 IHC categories was evaluated by Kruskal–Wallis test with post hoc pair-wise comparisons. The HER2-low signature discriminatory capability was assessed by estimating the area under the receiver operating characteristic curve (AUC). Gene Ontology and KEGG analyses were performed to evaluate the HER2-low signature genes functional enrichment. A HER2-low signature was computed based on HER2 IHC category-specific DEGs. The twenty genes included in the signature were significantly enriched with lipid and steroid metabolism pathways, peptidase regulation, and humoral immune response. The HER2-low signature values showed a bell-shaped distribution across IHC categories (low values in 0 and 3+; high values in 1+ and 2+), effectively distinguishing HER2-low from 0 (p < 0.001) to 3+ (p < 0.001). Notably, the signature values were higher in tumors scored with 1+ as compared to 0. The HER2-low signature association with IHC categories and its bell-shaped distribution was confirmed in the independent INT3, TCGA and GSE81538 datasets. In the combined INT1 and INT3 datasets, the HER2-low signature achieved an AUC value of 0.74 (95% confidence interval, CI 0.67–0.81) in distinguishing HER2-low vs. the other categories, outperforming the individual ERBB2 mRNA AUC value of 0.52 (95% CI 0.43–0.60). These results represent a proof-of-concept for an observer-independent gene-expression-based classifier of HER2-low status. The herein identified 20-gene signature shows promise in distinguishing between HER2 0 and HER2-low expressing tumors, including those scored as 1+ at IHC, and in developing a selection approach for ADCs candidates.

Profiling the heterogeneity of colorectal cancer consensus molecular subtypes using spatial transcriptomics

The consensus molecular subtypes (CMS) of colorectal cancer (CRC) is the most widely-used gene expression-based classification and has contributed to a better understanding of disease heterogeneity and prognosis. Nevertheless, CMS intratumoral heterogeneity restricts its clinical application, stressing the necessity of further characterizing the composition and architecture of CRC. Here, we used Spatial Transcriptomics (ST) in combination with single-cell RNA sequencing (scRNA-seq) to decipher the spatially resolved cellular and molecular composition of CRC. In addition to mapping the intratumoral heterogeneity of CMS and their microenvironment, we identified cell communication events in the tumor-stroma interface of CMS2 carcinomas. This includes tumor growth-inhibiting as well as -activating signals, such as the potential regulation of the ETV4 transcriptional activity by DCN or the PLAU-PLAUR ligand-receptor interaction. Our study illustrates the potential of ST to resolve CRC molecular heterogeneity and thereby help advance personalized therapy.

Diagnostic Utility of Comprehensive RNA-Seq Analysis in Adult B-ALL

Introduction Adult B-cell acute lymphoblastic leukemia (B-ALL) comprises 75% of adult ALL cases. The classification and risk stratification of patients have conventionally relied on cytogenetics analysis. However, cytogenetics has limitations due to its reliance on metaphase cells, which can mitigate classification accuracy. In recent years, transcriptome sequencing (RNA-seq) of large pediatrics ALL cohorts has expanded the number of recurrent gene fusions recognized as disease defining events and informed risk stratification [Brady et al. 2022, Mullighan et al. 2017]. This study aims to evaluate the diagnostic utility of RNA-seq for clinical subtype classification in adult B-ALL in a large representative cohort. Methods The study cohort comprised 338 adult patients (25-65 years) with newly diagnosed ALL (UKALL14, ISRCTN66541317, NCT01085617). Paired-end RNA-seq data at 60 Million reads per sample were generated. Analysis focused on the detection of disease-defining gene fusions and gene expression signatures. Three RNA fusion callers (FusionCatcher, FuSeq, and STAR-Fusion) and three expression-based classifiers (ALLSorts, ALLspice, and ALLCatchR) were used for consensus fusion calling and expression-based classification, respectively [Nicorici et al. 2014, Vu et al. 2018, Haas et al. 2019, Schmidt et al., Mäkinen et al., Bader et al. 2022]. Integration of classification results from expression-based classifiers and the consensus gene fusions provides 4 levels of RNA-derived subtype information, denoted as “RNA evidence” in this study. These RNA-seq based classification results were compared with clinical cytogenetics findings. Results Out of 338 samples in this cohort, cytogenetics unequivocally classified 73% (246/338) of the samples. By integrating RNA expression-based classification with fusion subtypes, we confidently classified 87% (295/338) of the cohort with at least 2 levels of RNA evidence. In total, consensus gene fusions were identified in 74% (249/338), while expression-based classifiers assigned a confident subtype to 82% (319/338). Subsequently, we compared cytogenetics with RNA-seq derived classification and found a high concordance of 98% (240/246) between the two approaches. Only 6 (&amp;lt;2%) samples with cytogenetics subtypes were not classified by RNA-seq, attributed to specific subtype characteristics (near-haploid) or lower RNA purity/quality. 27% (92/338) of subjects were not classified by cytogenetics (31 missing/failed and 61 indeterminate cytogenetics). Of the 31 samples with failed/missing cytogenetics, RNA-seq confidently assigned a class in 74% (23/31). For the remaining 61 cases with indeterminate cytogenetic (complex/other), RNA-seq resolved classification in 74% (45/61). Taken together, this improved the classification for 20% (68/338) of the entire cohort that was previously unclassified through cytogenetics. Notably, RNA-seq-exclusive findings were enriched in subtypes like DUX4, PAX5alt, and ZNF384, which are challenging for cytogenetics to discern. Importantly, integration of RNA-seq with cytogenetics led to a confident subtype classification for 93% of the cohort (314/338), a significant improvement over the initial 73% based on cytogenetics alone. Conclusions In conclusion, our study highlights the relevance of integrated RNA-seq analysis for reliable and comprehensive subtype classification in adult B-ALL. The high concordance between cytogenetics and RNA-derived subtypes underscores their complementary roles, providing valuable insights into disease mechanisms and guiding personalized treatment strategies.

Gene Expression Machine Learning Models Classify Pediatric AML Subtypes with High Performance

Pediatric acute myeloid leukemia (pAML) encompasses over 20 molecular subtypes driven by unique genetic alterations, including hallmark chromosomal rearrangements and less frequently, point mutations or tandem duplications. Collectively, many of these pAML molecular categories are enriched in pediatric populations and are not represented by current classification systems, including the recently updated WHO or ICC. While fusion detection from RNA-Seq-based approaches is robust, many fusion negative subtypes would need to be defined by expression-based approaches as mutation calling from RNA-Seq data is less developed. Nevertheless, this could be challenging for subtypes with similar transcriptional profiles, such as those with shared HOX expression patterns, including NPM1, NUP98r, UBTF, DEK::NUP214, and KMT2A-PTD. To aid in the appropriate molecular classification of pAML, which is crucial for prognosis, we developed and compared three gene expression-based classifiers. A total of 1707 pAML gene expression profiles were mapped and analyzed from three distinct sources (St. Jude = 659; TARGET = 168; AAML1031 = 880). Raw read count data was normalized and scaled to obtain a relative expression value in transcripts per million (TPM), which served as input for feature selection, model training, and testing. Ground truth labels for all 1707 samples were obtained through multi-omics analysis, including whole genome sequencing, to identify fusions and mutations. For validation purposes, the data was stratified by subtypes and split 70/30 into training (n=1187) and testing (n=520). Three machine learning models were selected: random forest, XGboost, and linear support vector machine (SVM). Each sample had gene expression TPM data for 60,754 transcripts, out of which 20,004 transcripts related to protein-coding genes were incorporated for feature selection. Feature selection was performed using a median absolute deviation (MAD) algorithm to select the 3000 transcripts with the highest variability. Top variable 3000 genes were selected to allow for adequate tuning of the number of predictors. Each model was independently trained using stratified cross-validation and Monte-Carlo search for hyperparameter tuning. The best model was selected based on the Matthews Correlation Coefficient (MCC). Each model was tested on the hold-out TPM set with z-score normalization. On the hold-out testing set (n=520), the linear SVM model outperformed the random forest and XGboost models on five performance metrics across all subtypes (sensitivity=0.9577; precision=0.9577; specificity=0.9978; F1=0.96; accuracy=0.9958). The random forest (sensitivity=0.9154; precision=0.9154; specificity=0.9955; F1=0.92; accuracy=0.9915) and XGboost models (sensitivity=0.9231; precision=0.9231; specificity=0.9960; F1=0.92; accuracy=0.9923) also performed well across all subtypes. Although feature selection was shared across all three models, performance within each subtype varied between models. The linear SVM model demonstrated strong performance overall, driven by high specificity in classifying the KMT2Ar subgroup (n=127) and equal sensitivity across the GLIS-rearranged (n=14), GATA1 (n=10), BCL11B (n=7), CBFB::MYH11 (n=60), CEBPA (n=33), and RUNX1::RUNX1T1 (n=78) subtypes. The primary difference in performance between models is the high false positive rate for KMT2Ar and NPM1 (n=55) in the random forest and XGboost models. A preliminary hypothesis for this might be due to the large representation of KMT2Ar and NPM1 in the training data (24.85% and 10.78%, respectively). Synthetic upsampling (SMOTE) for the training dataset (n=1369) counteracts bias towards the majority classes and increases performance for the random forest (sensitivity=0.9327; precision=0.9327; specificity=0.9965; F1=0.93; accuracy=0.9933), XGboost (sensitivity=0.9404; precision=0.9404; specificity=0.9969; F1=0.94; accuracy=0.9940) and linear SVM (sensitivity=0.9615; precision=0.9615; specificity=0.9980; F1=0.96; accuracy=0.9962) models. Conjointly, these models demonstrate the utility and effectiveness of a machine learning approach for classifying pAML samples from transcriptome sequencing data, which may have broad clinical and research utility, especially for fusion negative subtypes.

Molecular portraits of colorectal cancer morphological regions.

Heterogeneity of colorectal carcinoma (CRC) represents a major hurdle towards personalized medicine. Efforts based on whole tumor profiling demonstrated that the CRC molecular subtypes were associated with specific tumor morphological patterns representing tumor subregions. We hypothesize that whole-tumor molecular descriptors depend on the morphological heterogeneity with significant impact on current molecular predictors. We investigated intra-tumor heterogeneity by morphology-guided transcriptomics to better understand the links between gene expression and tumor morphology represented by six morphological patterns (morphotypes): complex tubular, desmoplastic, mucinous, papillary, serrated, and solid/trabecular. Whole-transcriptome profiling by microarrays of 202 tumor regions (morphotypes, tumor-adjacent normal tissue, supportive stroma, and matched whole tumors) from 111 stage II-IV CRCs identified morphotype-specific gene expression profiles and molecular programs and differences in their cellular buildup. The proportion of cell types (fibroblasts, epithelial and immune cells) and differentiation of epithelial cells were the main drivers of the observed disparities with activation of EMT and TNF-α signaling in contrast to MYC and E2F targets signaling, defining major gradients of changes at molecular level. Several gene expression-based (including single-cell) classifiers, prognostic and predictive signatures were examined to study their behavior across morphotypes. Most exhibited important morphotype-dependent variability within same tumor sections, with regional predictions often contradicting the whole-tumor classification. The results show that morphotype-based tumor sampling allows the detection of molecular features that would otherwise be distilled in whole tumor profile, while maintaining histopathology context for their interpretation. This represents a practical approach at improving the reproducibility of expression profiling and, by consequence, of gene-based classifiers.

Molecular portraits of colorectal cancer morphological regions

Heterogeneity of colorectal carcinoma (CRC) represents a major hurdle towards personalized medicine. Efforts based on whole tumor profiling demonstrated that the CRC molecular subtypes were associated with specific tumor morphological patterns representing tumor subregions. We hypothesize that whole-tumor molecular descriptors depend on the morphological heterogeneity with significant impact on current molecular predictors. We investigated intra-tumor heterogeneity by morphology-guided transcriptomics to better understand the links between gene expression and tumor morphology represented by six morphological patterns (morphotypes): complex tubular, desmoplastic, mucinous, papillary, serrated, and solid/trabecular. Whole-transcriptome profiling by microarrays of 202 tumor regions (morphotypes, tumor-adjacent normal tissue, supportive stroma, and matched whole tumors) from 111 stage II-IV CRCs identified morphotype-specific gene expression profiles and molecular programs and differences in their cellular buildup. The proportion of cell types (fibroblasts, epithelial and immune cells) and differentiation of epithelial cells were the main drivers of the observed disparities with activation of EMT and TNF-α signaling in contrast to MYC and E2F targets signaling, defining major gradients of changes at molecular level. Several gene expression-based (including single-cell) classifiers, prognostic and predictive signatures were examined to study their behavior across morphotypes. Most exhibited important morphotype-dependent variability within same tumor sections, with regional predictions often contradicting the whole-tumor classification. The results show that morphotype-based tumor sampling allows the detection of molecular features that would otherwise be distilled in whole tumor profile, while maintaining histopathology context for their interpretation. This represents a practical approach at improving the reproducibility of expression profiling and, by consequence, of gene-based classifiers.

GeneNet: Transfer learning-based hybrid African buffalo optimization with genetic algorithm for gene expression based cancer classification

Gene therapy is one of the advanced medical diagnosis environments, which is used to identify the solutions to various cancers by identifying the perfect gene expressions. The computer aided detection of gene expressions is a research problem, where cancer type is identified by artificial intelligence methods. However, the conventional machine learning models were failed to provide the accurate cancer classification and resulted in poor quantitative performance. Therefore, this work focuses on the implementation of gene expression-based cancer classification (GECC) network, here after referred as GeneNet. Initially, the proposed GeneNet performs the data preprocessing to normalize the gene-expression data by replacing the unknown characters and missing symbols. Then, the visual geometry group developed VGG16 model for extraction of correlated, and structural features from pre-processed dataset. In addition, hybrid African buffalo optimization with genetic algorithm (HABO-GA) is used to select the optimal features through natural inspired hunting (searching) properties. Finally, low complex convolutional neural network (LC-CNN) model is trained with HABO-GA features, which also performs the multi-class cancer classification from test data. The simulations are conducted on publicly available GECC datasets such as lymphography, ovarian, leukemia, lung, and prostate cancers discloses the superiority of proposed GeneNetas compared to state-of-the-art approaches with 86.6 %, 100 %, 76.92 %, 100 %, and 100 % of accuracy, respectively.

DeepGene Transformer: Transformer for the gene expression-based classification of cancer subtypes

Cancer and its subtypes constitute approximately 30% of all causes of death globally and display a wide range of heterogeneity in terms of clinical and molecular responses to therapy. Molecular subtyping has enabled the use of precision medicine to overcome these challenges and provide significant biological insights to predict prognosis and improve clinical decision-making. Over the past decade, conventional machine learning (ML) and deep learning (DL) algorithms have been widely espoused for the classification of cancer subtypes from gene expression datasets. However, these methods are potentially biased toward the identification of cancer biomarkers. Hence, an end-to-end deep learning approach, DeepGene Transformer, is proposed which addresses the complexity of high-dimensional gene expression with a multi-head self-attention module by identifying relevant biomarkers across multiple cancer subtypes without requiring feature selection as a pre-requisite for the current classification algorithms. Comparative analysis reveals that the proposed DeepGene Transformer outperformed the commonly used traditional and state-of-the-art classification algorithms and can be considered an efficient approach for classifying cancer and its subtypes, indicating that any improvement in deep learning models in computational biologists can be reflected well in this domain as well.

Evaluation and Exploration of Machine Learning and Convolutional Neural Network Classifiers in Detection of Lung Cancer from Microarray Gene-A Paradigm Shift.

Microarray gene expression-based detection and classification of medical conditions have been prominent in research studies over the past few decades. However, extracting relevant data from the high-volume microarray gene expression with inherent nonlinearity and inseparable noise components raises significant challenges during data classification and disease detection. The dataset used for the research is the Lung Harvard 2 Dataset (LH2) which consists of 150 Adenocarcinoma subjects and 31 Mesothelioma subjects. The paper proposes a two-level strategy involving feature extraction and selection methods before the classification step. The feature extraction step utilizes Short Term Fourier Transform (STFT), and the feature selection step employs Particle Swarm Optimization (PSO) and Harmonic Search (HS) metaheuristic methods. The classifiers employed are Nonlinear Regression, Gaussian Mixture Model, Softmax Discriminant, Naive Bayes, SVM (Linear), SVM (Polynomial), and SVM (RBF). The two-level extracted relevant features are compared with raw data classification results, including Convolutional Neural Network (CNN) methodology. Among the methods, STFT with PSO feature selection and SVM (RBF) classifier produced the highest accuracy of 94.47%.

Targeting P21-activated kinase suppresses proliferation and enhances chemosensitivity in T-cell lymphoblastic lymphoma.

T-cell lymphoblastic lymphoma (T-LBL) is a highly aggressive non-Hodgkin lymphoma with a poor prognosis. P21-activated kinase (PAK) is a component of the gene expression-based classifier that can predict the prognosis of T-LBL. However, the role of PAK in T-LBL progression and survival remains poorly understood. Herein, we found that the expression of PAK1 was significantly higher in T-LBL cell lines (Jurkat, SUP-T1, and CCRF-CEM) compared to the human T-lymphoid cell line. Moreover, PAK2 mRNA level of 32 relapsed T-LBL patients was significantly higher than that of 37 cases without relapse (P = .012). T-LBL patients with high PAK1 and PAK2 expression had significantly shorter median RFS than those with low PAK1 and PAK2 expression (PAK1, P = .028; PAK2, P = .027; PAK1/2, P = .032). PAK inhibitors, PF3758309 (PF) and FRAX597, could suppress the proliferation of T-LBL cells by blocking the G1/S cell cycle phase transition. Besides, PF could enhance the chemosensitivity to doxorubicin in vitro and in vivo. Mechanistically, through western blotting and RNA sequencing, we identified that PF could inhibit the phosphorylation of PAK1/2 and downregulate the expression of cyclin D1, NF-κB and cell adhesion signaling pathways in T-LBL cell lines. These findings suggest that PAK might be associated with T-LBL recurrence and further found that PAK inhibitors could suppress proliferation and enhance chemosensitivity of T-LBL cells treated with doxorubicin. Collectively, our present study underscores the potential therapeutic effect of inhibiting PAK in T-LBL therapy.

394 Gene expression based molecular test proves clinical validity as diagnostic aid for the differential diagnosis of psoriasis and eczema in formalin fixed and paraffin embedded tissue

Abstract Highly specific and efficient drugs have been developed to treat noncommunicable chronic inflammatory skin diseases (ncISD). Due to their specificity, these drugs require precise diagnostics to attribute the most efficient treatment to each patient. Diagnosis is complicated by the complex pathogenesis of ncISD and their clinical and histological overlap. Especially, precise diagnosis of psoriasis and atopic eczema is difficult in special cases, and molecular tools need to be developed to support gold standard diagnosis. Psoriasis and eczema are among the most prevalent skin diseases worldwide with a combined incidence rate of at least 2%. Although more than a dozen of immune-modulating therapies, particularly monoclonal antibodies targeting specific cytokines or their receptors, have become available over the last decade allowing highly effective treatment, the ultimate breakthrough would include accurate diagnostic tools. Diagnosis of both diseases largely depends on a subjective visual exam and histopathological examination. Due to phenotypical overlaps of both diseases, up to 50% of cases with a palmoplantar localization are misdiagnosed. To close this diagnostic gap, a gene expression-based classifier using NOS2 and CCL27 has been proposed and its clinical validity examined and proven in various patient cohorts. This study aims to develop a real-time based molecular classifier (MC) to distinguish psoriasis from atopic eczema in formalin-fixed, paraffin-embedded (FFPE)-fixed skin samples for diagnostic measures and to evaluate the potential of minimally invasive microbiopsies and noninvasive tape strips or sampling disks for molecular diagnosis. The FFPE, micro-(Ø1mm) and macrobiopsies (Ø4–6 mm), and tape strips were collected from psoriasis and eczema lesions and analysed by real-time polymerase chain reaction (PCR) for the expression of NOS2 and CCL27. A molecular classifier (MC) was established using a linear regression model. We transferred the RNAprotect-based molecular classifier to FFPE samples and were able to produce comparable results. Whereas the RNAprotect based MC distinguished psoriasis from eczema with a sensitivity and specificity for psoriasis of 97.7% and 100%, respectively, and an AUC of 0.99, the FFPE-based classifier determined probabilities for psoriasis with a sensitivity and specificity and of 92% and 100%, respectively, and an AUC of 0.97. To test if microbiopsies are also adequate tissue samples for the MC, we analysed gene expression in 83 pairs of macro- and microbiopsies by real-time quantitative-PCR. Delta-CT values showed no significant difference between macro- and microbiopsies. The MC led to reliable results in the vast majority of samples (91.57%) independent of biopsy size, with a small increase of misclassification (4.98%). The FFPE-based molecular classifier (MC) precisely separates eczema from psoriasis and efficiently identifies subtypes of both diseases. The reliability of the two markers NOS2 and CCL27 has been shown in FFPE tissue facilitating a possible implementation of this molecular diagnostic aid in routine clinical pathological practice. We further show that the gene expression profile of RNA later fixed microbiopsies is comparable to standard 4–6 mm biopsies and that microbiopsies are equally suited for the MC. We demonstrate the potential of tape strips of inflamed epidermis for molecular diagnostics of ncISD. Due to the minimally invasive sampling procedure, tape stripping may be particularly useful for the testing of visible and sensitive skin areas and in children as well as for repetitive sampling procedures necessary to monitor treatment responses over time. In summary, the MC discriminates psoriasis from eczema producing meaningful results in a broad range of skin tissue derived from invasive to minimally invasive sampling techniques. Further efforts are required to simplify and shorten the labor- and cost-intensive procedures of RNA isolation and real-time PCR. One possible solution is a fully automated and closed system, which would allow the integration of molecular diagnostics into the routine patient management at the point of care.

Xerna™ TME Panel is a machine learning-based transcriptomic biomarker designed to predict therapeutic response in multiple cancers.

Most predictive biomarkers approved for clinical use measure single analytes such as genetic alteration or protein overexpression. We developed and validated a novel biomarker with the aim of achieving broad clinical utility. The Xerna™ TME Panel is a pan-tumor, RNA expression-based classifier, designed to predict response to multiple tumor microenvironment (TME)-targeted therapies, including immunotherapies and anti-angiogenic agents. The Panel algorithm is an artificial neural network (ANN) trained with an input signature of 124 genes that was optimized across various solid tumors. From the 298-patient training data, the model learned to discriminate four TME subtypes: Angiogenic (A), Immune Active (IA), Immune Desert (ID), and Immune Suppressed (IS). The final classifier was evaluated in four independent clinical cohorts to test whether TME subtype could predict response to anti-angiogenic agents and immunotherapies across gastric, ovarian, and melanoma datasets. The TME subtypes represent stromal phenotypes defined by angiogenesis and immune biological axes. The model yields clear boundaries between biomarker-positive and -negative and showed 1.6-to-7-fold enrichment of clinical benefit for multiple therapeutic hypotheses. The Panel performed better across all criteria compared to a null model for gastric and ovarian anti-angiogenic datasets. It also outperformed PD-L1 combined positive score (>1) in accuracy, specificity, and positive predictive value (PPV), and microsatellite-instability high (MSI-H) in sensitivity and negative predictive value (NPV) for the gastric immunotherapy cohort. The TME Panel's strong performance on diverse datasets suggests it may be amenable for use as a clinical diagnostic for varied cancer types and therapeutic modalities.

Comprehensive transcriptomic analysis to identify biological and clinical differences in cholangiocarcinoma.

Cholangiocarcinoma (CC) is a rare and aggressive disease with limited therapeutic options and a poor prognosis. All available public records of cohorts reporting transcriptomic data on intrahepatic cholangiocarcinoma (ICC) and extrahepatic cholangiocarcinoma (ECC) were collected with the aim to provide a comprehensive gene expression-based classification with clinical relevance. A total of 543 patients with primary tumor tissues profiled by RNAseq and microarray platforms from seven public datasets were used as a discovery set to identify distinct biological subgroups. Group predictors developed on the discovery sets were applied to a single cohort of 131 patients profiled with RNAseq for validation and assessment of clinical relevance leveraging machine learning techniques. By unsupervised clustering analysis of gene expression data we identified both in the ICC and ECC discovery datasets four subgroups characterized by a distinct type of immune infiltrate and signaling pathways. We next developed class predictors using short gene list signatures and identified in an independent dataset subgroups of ICC tumors at different prognosis. The developed class-predictor allows identification of CC subgroups with specific biological features and clinical behavior at single-sample level. Such results represent the starting point for a complete molecular characterization of CC, including integration of genomics data to develop in clinical practice.

Abstract P4-09-05: Characterization and validation of biologically-driven HER2-positive breast cancer subgroups in the ALTTO and NeoALTTO clinical trials

Abstract Background: Tumor and microenvironment features, including luminal phenotype as well as metabolic, immune and stroma activation, impact prognosis and treatment response in HER2-positive breast cancer. Here, we aimed to identify subgroups depicting biological processes associated with prognosis in patients receiving adjuvant trastuzumab in the phase III ALTTO trial, and to validate their biological and prognostic characteristics in the phase III neoadjuvant NeoALTTO trial. Methods: By applying a case-control approach (1:2), we selected from the ALTTO trastuzumab arm 134 and 268 patients with and without a distant relapse respectively, matched for clinicopathological characteristics. In ALTTO, RNA was obtained from FFPE tumor cores from surgical samples. RNA sequencing (RNAseq) data available from 254 frozen pre-treatment samples were used for validation in patients receiving trastuzumab and/or lapatinib in NeoALTTO. The Absolute Intrinsic Molecular Subtyping (AIMS) was used to compute PAM50 subtypes. Prognostic genes were identified using a multivariable Cox proportional hazard model (controlling for clinicopathological characteristics and PAM50 HER2-enriched [HER2-E] vs others) for distant relapse-free survival (DRFS). Clusters were identified using non-negative matrix factorization (NMF) and k-means clustering, and characterized with gene expression profiling and Gene Set Variation Analysis (GSVA, hallmark gene sets from MSigDB). Aiming at identifying the subgroups in external cohorts, we identified relevant genes to develop group-specific signatures with LASSO regression, and used the derived scores to build a multinomial classifier. Results: The case-control cohort includes a high-risk population, with higher proportions of &amp;gt;2cm, node positive and G3 tumors compared to the whole ALTTO trastuzumab arm. RNAseq data were generated for 386/402 patients. NMF and k-means clustering performed on genes associated with DRFS (false discovery rate &amp;lt; 0.05) identified 4 groups with distinct biological characteristics and prognosis: immune-enriched (IM, N = 69), proliferative/metabolic (P/M, N = 87, characterized by glycolysis, cholesterol homeostasis and proliferation pathways), mesenchymal/stroma-enriched (M/S, N = 76, characterized by epithelial-mesenchymal transition, angiogenesis and TGF-beta signaling), and hormone receptor positive-enriched (N = 154), further divided into PAM50 HER2-E (N = 91) and non-HER2-E (N = 63, LUM, mainly luminal A/B tumors). The IM and LUM groups presented better DRFS (91% and 87% 5-year DRFS, respectively) compared to the others (5-year DRFS of 72%, 58% and 51% for HER2-E, M/S and P/M, respectively). The gene expression-based classifier identified the same 5 groups in NeoALTTO. GSVA analysis and comparisons among the 5 groups showed similar results in the two studies. Of interest, we observed significant differences in event-free survival in NeoALTTO (P = 0.035 and P = 0.041 in the whole population and in the subgroup with residual disease, respectively), with IM and LUM presenting better outcomes compared to the other groups. Sensitivity to neoadjuvant treatment as described by pathological complete response (pCR, ypT0/is ypN0) differed across the groups, with LUM presenting lower pCR rates (8%) compared to HER2-E (40%), IM (35%), P/M (32%), and M/S (32%). Conclusions: Five biologically-driven HER2-positive breast cancer subgroups were identified in ALTTO, highlighting the heterogeneity of this disease. Of note, their biological features and clinical behavior were validated in the NeoALTTO population, suggesting the robustness of our findings. IM and LUM tumors could be considered for treatment de-escalation approaches. Additional validation in cohorts receiving standard (neo)adjuvant therapies is warranted. Citation Format: Mattia Rediti, David Venet, Andrea Joaquin Garcia, Dominique Agbor-Tarh, Marion Maetens, Delphine Vincent, Samira Majjaj, Sarra El-Abed, Takayuki Ueno, Serena Di Cosimo, Martine Piccart, Lajos Pusztai, Sherene Loi, Roberto Salgado, Giuseppe Viale, Françoise Rothé, Christos Sotiriou. Characterization and validation of biologically-driven HER2-positive breast cancer subgroups in the ALTTO and NeoALTTO clinical trials [abstract]. In: Proceedings of the 2022 San Antonio Breast Cancer Symposium; 2022 Dec 6-10; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2023;83(5 Suppl):Abstract nr P4-09-05.

Genome-Wide Methylation and Haplotype-Resolved Aberrant Somatic Hypermutation Patterns in B-Cell Lymphomas

Introduction Diffuse large B-cell lymphoma (DLBCL) is a genetically, phenotypically, and clinically heterogenous disease. Multiple molecular classification systems using either genomic aberrations or gene expression profiles are used to delineate subgroups with shared biology. The cell of origin (COO) gene expression-based classification system stratifies DLBCLs into germinal center B-cell-like (GCB) and activated B-cell-like subgroups of DLBCL. More recently, the double-hit signature (DHITsig) has been used to further stratify GCB DLBCLs into DHITsig+ and DHITsig-, where the former is characterized by a gene expression profile reminiscent of the GC dark zone and inferior outcomes. Currently, patterns of somatic driver mutations can only partially explain the expression of DHITsig in some tumors. Genetic features such as DNA methylation have been shown to be associated with transcriptional repression and altered chromatin structure. We hypothesized that resolving DNA methylation patterns genome-wide using long read whole genome sequencing (LR-WGS) may reveal epigenetic differences associated with differential gene expression profiles within the DHITsig+ and DHITsig- subgroups and in the context of histologic transformation from follicular lymphoma. Methods Whole genome sequencing data of 10 patient samples and 4 cell line GCB DLBCL samples were generated using both short read (SR-WGS; Illumina) and LR-WGS (Oxford Nanopore Technologies, PromethION). The complementary approach combines somatic variant calling from highly accurate SR data with DNA methylation calling from LR data. GCB samples were selected based on the NanoString DLBCL90 assay and further subdivided into DHITsig+ (N=4) and DHITsig- (N=3). RNA sequencing data of 9 patient samples and 3 GCB DLBCL cell lines were generated and analyzed using Salmon and DESeq2. Simple somatic mutations from SR-WGS data were called using an ensemble of somatic variant callers, which confirmed all cases to belong to the EZB genetic subgroup. Methylated bases were called with Nanopolish followed by a differential methylation analysis using the DSS package in R to identify differences in methylation between both subgroups. Differentially methylated regions (DMRs) were annotated for GC B-cell specific superenhancers, CpG islands, and genes. Using WhatsHap and Nanomethphase, heterozygous single nucleotide polymorphisms (SNPs) from SR and LR data were used to phase LRs into haplotypes. Haplotype analyses were performed to determine the phase of aberrant somatic hypermutation (aSHM) and methylation patterns in the context of proximal translocation breakpoints. Results Most DMRs were found to be associated with regulatory regions such as promoters, superenhancers, and CpG islands. We found methylation differences between DHITsig+ and DHITsig- subgroups in 1430 superenhancers known to be active in GC B-cells. Within these superenhancer regions, DHITsig+ tumors were typically hypomethylated relative to DHITsig-. Several DHITsig classifier genes, including WNK2, TNFSF8, and IL21R, had methylation patterns inverted relative to the typical expression pattern (i.e. low methylation with high expression). Similarly, among 633 differentially methylated promoters, several including TCF3 and WNK2 demonstrated the same inverted methylation pattern. CpG islands showed distinct patterns of methylation in both subgroups that were inversely associated with expression, this also included genes differentially expressed between DHITsig+ and DHITsig- tumors. The DHITsig- group had increased levels of methylation of DHITsig genes in both promoters and gene bodies suggesting a functional role for gene body methylation. Finally, we observed that a high density of mutations near the BCL2 transcription start site consistently phased into the same haplotype as the BCL2-IGH translocation in five tumours, demonstrating the allele specificity of aSHM associated with these events (Panel B). Conclusions These results demonstrate differences in methylation between the subgroups. The association of hypomethylated superenhancers, promoters, and CpG islands with genes over-expressed in DHITsig+ suggests these methylation patterns may contribute to the DHITsig phenotype. These results may aid in improving GCB DLBCL patient classification and reveal factors contributing to its heterogeneity. Figure 1View largeDownload PPTFigure 1View largeDownload PPT Close modal