Exploring combinations of dimensionality reduction, transfer learning, and regularization methods for predicting binary phenotypes with transcriptomic data.

BACKGROUND: Numerous transcriptomic-based models have been developed to predict or understand the fundamental mechanisms driving biological phenotypes. However, few models have successfully transitioned into clinical practice due to challenges associated with generalizability and interpretability. To address these issues, researchers have turned to dimensionality reduction methods and have begun implementing transfer learning approaches. METHODS: In this study, we aimed to determine the optimal combination of dimensionality reduction and regularization methods for predictive modeling. We applied seven dimensionality reduction methods to various datasets, including two supervised methods (linear optimal low-rank projection and low-rank canonical correlation analysis), two unsupervised methods [principal component analysis and consensus independent component analysis (c-ICA)], and three methods [autoencoder (AE), adversarial variational autoencoder, and c-ICA] within a transfer learning framework, trained on > 140,000 transcriptomic profiles. To assess the performance of the different combinations, we used a cross-validation setup encapsulated within a permutation testing framework, analyzing 30 different transcriptomic datasets with binary phenotypes. Furthermore, we included datasets with small sample sizes and phenotypes of varying degrees of predictability, and we employed independent datasets for validation. RESULTS: Our findings revealed that regularized models without dimensionality reduction achieved the highest predictive performance, challenging the necessity of dimensionality reduction when the primary goal is to achieve optimal predictive performance. However, models using AE and c-ICA with transfer learning for dimensionality reduction showed comparable performance, with enhanced interpretability and robustness of predictors, compared to models using non-dimensionality-reduced data. CONCLUSION: These findings offer valuable insights into the optimal combination of strategies for enhancing the predictive performance, interpretability, and generalizability of transcriptomic-based models.

Robust metabolic transcriptional components in 34,494 patient-derived cancer-related samples and cell lines.

BACKGROUND: Patient-derived bulk expression profiles of cancers can provide insight into the transcriptional changes that underlie reprogrammed metabolism in cancer. These profiles represent the average expression pattern of all heterogeneous tumor and non-tumor cells present in biopsies of tumor lesions. Hence, subtle transcriptional footprints of metabolic processes can be concealed by other biological processes and experimental artifacts. However, consensus independent component analyses (c-ICA) can capture statistically independent transcriptional footprints of both subtle and more pronounced metabolic processes. METHODS: We performed c-ICA with 34,494 bulk expression profiles of patient-derived tumor biopsies, non-cancer tissues, and cell lines. Gene set enrichment analysis with 608 gene sets that describe metabolic processes was performed to identify the transcriptional components enriched for metabolic processes (mTCs). The activity of these mTCs was determined in all samples to create a metabolic transcriptional landscape. RESULTS: A set of 555 mTCs was identified of which many were robust across different datasets, platforms, and patient-derived tissues and cell lines. We demonstrate how the metabolic transcriptional landscape defined by the activity of these mTCs in samples can be used to explore the associations between the metabolic transcriptome and drug sensitivities, patient outcomes, and the composition of the immune tumor microenvironment. CONCLUSIONS: To facilitate the use of our transcriptional metabolic landscape, we have provided access to all data via a web portal ( www.themetaboliclandscapeofcancer.com ). We believe this resource will contribute to the formulation of new hypotheses on how to metabolically engage the tumor or its (immune) microenvironment.

Colorectal anastomotic leak: transcriptomic profile analysis.

BACKGROUND: Anastomotic leakage in patients undergoing colorectal surgery is associated with morbidity and mortality. Although multiple risk factors have been identified, the underlying mechanisms are mainly unknown. The aim of this study was to perform a transcriptome analysis of genes underlying the development of anastomotic leakage. METHODS: A set of human samples from the anastomotic site collected during stapled colorectal anastomosis were used in the study. Transcriptomic profiles were generated for patients who developing anastomotic leakage and case-matched controls with normal anastomotic healing to identify genes and biological processes associated with the development of anastomotic leakage. RESULTS: The analysis included 22 patients with and 69 without anastomotic leakage. Differential expression analysis showed that 44 genes had adjusted P < 0.050, consisting of two upregulated and 42 downregulated genes. Co-functionality analysis of the 150 most upregulated and 150 most downregulated genes using the GenetICA framework showed formation of clusters of genes with different enrichment for biological pathways. The enriched pathways for the downregulated genes are involved in immune response, angiogenesis, protein metabolism, and collagen cross-linking. The enriched pathways for upregulated genes are involved in cell division. CONCLUSION: These data indicate that patients who develop anastomotic leakage start the healing process with an error at the level of gene regulation at the time of surgery. Despite normal macroscopic appearance during surgery, the transcriptome data identified several differences in gene expression between patients who developed anastomotic leakage and those who did not. The expressed genes and enriched processes are involved in the different stages of wound healing. These provide therapeutic and diagnostic targets for patients at risk of anastomotic leakage.

Driving innovation for rare skin cancers: utilizing common tumours and machine learning to predict immune checkpoint inhibitor response.

Metastatic Merkel cell carcinoma (MCC) and cutaneous squamous cell carcinoma (cSCC) are rare and both show impressive responses to immune checkpoint inhibitor treatment. However, at least 40% of patients do not respond to these expensive and potentially toxic drugs. Development of predictive biomarkers of response and rational, effective combination treatment strategies in these rare, often frail patient populations is challenging. This review discusses the pathophysiology and treatment of MCC and cSCC, with a particular focus on potential biomarkers of response to immunotherapy, and discusses how transfer learning using big data collected from patients with common tumours can be used in combination with deep phenotyping of rare tumours to develop predictive biomarkers and elucidate novel treatment targets.

The antibody-drug conjugate target landscape across a broad range of tumour types.

BACKGROUND: Antibody-drug conjugates (ADCs), consisting of an antibody designed against a specific target at the cell membrane linked with a cytotoxic agent, are an emerging class of therapeutics. Because ADC tumour cell targets do not have to be drivers of tumour growth, ADCs are potentially relevant for a wide range of tumours currently lacking clear oncogenic drivers. Therefore, we aimed to define the landscape of ADC targets in a broad range of tumours. MATERIALS AND METHODS: PubMed and ClinicalTrials.gov were searched for ADCs that are or were evaluated in clinical trials. Gene expression profiles of 18 055 patient-derived tumour samples representing 60 tumour (sub)types and 3520 healthy tissue samples were collected from the public domain. Next, we applied Functional Genomic mRNA-profiling to predict per tumour type the overexpression rate at the protein level of ADC targets with healthy tissue samples as a reference. RESULTS: We identified 87 ADCs directed against 59 unique targets. A predicted overexpression rate of ≥ 10% of samples for multiple ADC targets was observed for high-incidence tumour types like breast cancer (n = 31 with n = 23 in triple negative breast cancer), colorectal cancer (n = 18), lung adenocarcinoma (n = 18), squamous cell lung cancer (n = 16) and prostate cancer (n = 5). In rare tumour types we observed, amongst others, a predicted overexpression rate of 55% of samples for CD22 and 55% for ENPP3 in adrenocortical carcinomas, 81% for CD74 and 81% for FGFR3 in osteosarcomas, and 95% for c-MET in uveal melanomas. CONCLUSION: This study provides a data-driven prioritization of clinically available ADCs directed against 59 unique targets across 60 tumour (sub)types. This comprehensive ADC target landscape can guide clinicians and drug developers which ADC is of potential interest for further evaluation in which tumour (sub)type.

ATR inhibition preferentially targets homologous recombination-deficient tumor cells.

Homologous recombination (HR) is required for faithful repair of double-strand DNA breaks. Defects in HR repair cause severe genomic instability and challenge cellular viability. Paradoxically, various cancers are HR defective and have apparently acquired characteristics to survive genomic instability. We aimed to identify these characteristics to uncover therapeutic targets for HR-deficient cancers. Cytogenetic analysis of 1143 ovarian cancers showed that the degree of genomic instability was correlated to amplification of replication checkpoint genes ataxia telangiectasia and Rad3-related kinase (ATR) and CHEK1. To test whether genomic instability leads to increased reliance on replication checkpoint signaling, we inactivated Rad51 to model HR-related genomic instability. Rad51 inactivation caused defective HR repair and induced aberrant replication dynamics. Notably, inhibition of Rad51 led to increased ATR/checkpoint kinase-1 (Chk1)-mediated replication stress signaling. Importantly, inhibition of ATR or Chk1 preferentially killed HR-deficient cancer cells. Combined, our data show that defective HR caused by Rad51 inhibition results in differential sensitivity for ATR and Chk1 inhibitors, implicating replication checkpoint kinases as potential drug targets for HR-defective cancers.

A bioinformatical and functional approach to identify novel strategies for chemoprevention of colorectal cancer.

Comparing normal colorectal mucosa and adenomas focusing on deregulated pathways obtains insight into the biological processes of early colorectal carcinogenesis. Publicly available microarray expression data from 26 normal mucosa and 47 adenoma samples were analyzed. Biological pathways enriched in adenomas were identified with Gene Set Enrichment Analysis (GSEA). The analysis revealed 10, 11 and 16 gene sets distinguishing adenomas from normal mucosa according to Kyoto Encyclopedia of Genes and Genomes (KEGG), Gene Map Annotator and Pathway Profiler (GenMAPP) and Biocarta databases, respectively. Biological pathways known to be involved in colon carcinogenesis such as cell cycle (P=0.002) and Wnt signaling (P=0.007) were enriched in adenomas. In addition, we found enrichment of novel pathways such as retinoblastoma (Rb) pathway (P=0.002), Src pathway (P=0.004), folate biosynthesis (P=0.019) and Hedgehog signaling (P=0.037) in adenomas. Microarray results for Rb and Src pathway genes were validated by quantitative reverse transcriptase-polymerase chain reaction (qRT-PCR) on mRNA isolated from an independent set of adenoma and normal colon samples. A high correlation between microarray data and qRT-PCR expression data was found. The relevance of targeting of identified pathways was shown using the Rb pathway inhibitors roscovitine and PD-0332991 and the Src pathway inhibitor dasatinib. All inhibitors used induced cell growth reduction in adenoma cells. This study shows a bioinformatical and functional approach leading to potentially new options for chemoprevention of colorectal cancer.

Identification of genes and pathways associated with cytotoxic T lymphocyte infiltration of serous ovarian cancer.

BACKGROUND: Tumour-infiltrating lymphocytes (TILs) are predictors of disease-specific survival (DSS) in ovarian cancer. It is largely unknown what factors contribute to lymphocyte recruitment. Our aim was to evaluate genes and pathways contributing to infiltration of cytotoxic T lymphocytes (CTLs) in advanced-stage serous ovarian cancer. METHODS: For this study global gene expression was compared between low TIL (n=25) and high TIL tumours (n=24). The differences in gene expression were evaluated using parametric T-testing. Selectively enriched biological pathways were identified with gene set enrichment analysis. Prognostic influence was validated in 157 late-stage serous ovarian cancer patients. Using immunohistochemistry, association of selected genes from identified pathways with CTL was validated. RESULTS: The presence of CTL was associated with 320 genes and 23 pathways (P<0.05). In addition, 54 genes and 8 pathways were also associated with DSS in our validation cohort. Immunohistochemical evaluation showed strong correlations between MHC class I and II membrane expression, parts of the antigen processing and presentation pathway, and CTL recruitment. CONCLUSION: Gene expression profiling and pathway analyses are valuable tools to obtain more understanding of tumour characteristics influencing lymphocyte recruitment in advanced-stage serous ovarian cancer. Identified genes and pathways need to be further investigated for suitability as therapeutic targets.