Ensemble learning models that predict surface protein abundance from single-cell multimodal omics data.

Single-cell protein abundance is a fundamental type of information to characterize cell states. Due to high cost and technical barriers, however, direct quantification of proteins is difficult. Single-cell RNA sequencing (scRNA-seq) data, serving as a cost-effective substitute of single-cell proteomics, may not accurately reflect protein expression levels due to measurement error, noise, post-transcriptional and translational regulation, etc. The recently emerging single-cell multimodal omics data, e.g. CITE-seq and REAP-seq, can simultaneously profile RNA and protein abundances in single cells, providing labeled data for predictive modeling in a supervised learning framework. Deep neural network-based transfer learning method has been applied to imputation of surface protein abundances from single-cell transcriptomic data. However, it is unclear if the artificial neural network is the best model, and it is desirable to improve the prediction performance (e.g. accuracy, interpretability) of machine learning models. In this paper, we compared several tree-based ensemble learning methods with neural network models, and found that ensemble learning often performed better than neural network, and Random Forest (RF) performed the best overall. Moreover, we used the feature importance scores from RF to interpret biological mechanisms underlying the prediction. Our study demonstrates the effectiveness of ensemble learning for reliable protein abundances prediction using single-cell multimodal omics data, and paves the way for knowledge discovery by mining single-cell multi-omics data in large scale.

PIKE-R2P: Protein-protein interaction network-based knowledge embedding with graph neural network for single-cell RNA to protein prediction.

BACKGROUND: Recent advances in simultaneous measurement of RNA and protein abundances at single-cell level provide a unique opportunity to predict protein abundance from scRNA-seq data using machine learning models. However, existing machine learning methods have not considered relationship among the proteins sufficiently. RESULTS: We formulate this task in a multi-label prediction framework where multiple proteins are linked to each other at the single-cell level. Then, we propose a novel method for single-cell RNA to protein prediction named PIKE-R2P, which incorporates protein-protein interactions (PPI) and prior knowledge embedding into a graph neural network. Compared with existing methods, PIKE-R2P could significantly improve prediction performance in terms of smaller errors and higher correlations with the gold standard measurements. CONCLUSION: The superior performance of PIKE-R2P indicates that adding the prior knowledge of PPI to graph neural networks can be a powerful strategy for cross-modality prediction of protein abundances at the single-cell level.

Sunglasses to hide behind may also prevent melanoma of the eyes.

In 1967, Sandy Posey pronounced that sunglasses are essential beachwear ( https://www.youtube.com/watch?v=4HVBEb-GA1Y ). Now, whole-genome sequencing reveals that ultraviolet radiation (UVR) can contribute to melanomas in the iris and conjunctiva, data that provide a molecular explanation for why it is important to protect our eyes from exposure to UVR.

Weight Loss and Exercise Alter the High-Density Lipoprotein Lipidome and Improve High-Density Lipoprotein Functionality in Metabolic Syndrome.

OBJECTIVE: High-density lipoprotein (HDL) lipid composition and function may better reflect cardiovascular risk than HDL cholesterol concentration. This study characterized the relationships between HDL composition, metabolism, and function in metabolic syndrome (MetS) patients and how changes in composition after weight loss (WL) and exercise treatments are related to function. APPROACH AND RESULTS: Plasma samples from MetS patients (n=95) and healthy individuals (n=40) were used in this study. Subsets of the MetS group underwent 12 weeks of no treatment (n=17), WL (n=19), or WL plus exercise (WLEX; n=17). HDL was isolated using density-gradient ultracentrifugation. The HDL lipidome was analyzed by mass spectrometry, and particle size determined by nuclear magnetic resonance. Cholesteryl ester transfer protein activity and ex vivo HDL cholesterol efflux capacity (CEC) were assessed. The HDL lipidome in the MetS patients was substantially different from that in healthy individuals, mean particle size was smaller, and CEC was lower. Several HDL phospholipid and sphingolipid species were associated with HDL diameter and CEC. The HDL lipidome and particle size were modified toward the healthy individuals after WL and WLEX treatments, with greater effects observed in the latter group. Cholesteryl ester transfer protein activity was reduced after WL and WLEX, and CEC was improved after WLEX. CONCLUSIONS: WLEX treatment in MetS patients normalizes the HDL lipidome and particle size profile and enhances CEC. HDL lipids associated with diminished CEC may represent novel biomarkers for early prediction of HDL dysfunction and disease risk and may represent potential therapeutic targets for future HDL therapies. CLINICAL TRIAL REGISTRATION: URL: http://www.clinicaltrials.gov. Unique identifier: NCT00163943.

Plasma Lipidomic Profiles Improve on Traditional Risk Factors for the Prediction of Cardiovascular Events in Type 2 Diabetes Mellitus.

BACKGROUND: Clinical lipid measurements do not show the full complexity of the altered lipid metabolism associated with diabetes mellitus or cardiovascular disease. Lipidomics enables the assessment of hundreds of lipid species as potential markers for disease risk. METHODS: Plasma lipid species (310) were measured by a targeted lipidomic analysis with liquid chromatography electrospray ionization-tandem mass spectrometry on a case-cohort (n=3779) subset from the ADVANCE trial (Action in Diabetes and Vascular Disease: Preterax and Diamicron-MR Controlled Evaluation). The case-cohort was 61% male with a mean age of 67 years. All participants had type 2 diabetes mellitus with ≥1 additional cardiovascular risk factors, and 35% had a history of macrovascular disease. Weighted Cox regression was used to identify lipid species associated with future cardiovascular events (nonfatal myocardial infarction, nonfatal stroke, and cardiovascular death) and cardiovascular death during a 5-year follow-up period. Multivariable models combining traditional risk factors with lipid species were optimized with the Akaike information criteria. C statistics and NRIs were calculated within a 5-fold cross-validation framework. RESULTS: Sphingolipids, phospholipids (including lyso- and ether- species), cholesteryl esters, and glycerolipids were associated with future cardiovascular events and cardiovascular death. The addition of 7 lipid species to a base model (14 traditional risk factors and medications) to predict cardiovascular events increased the C statistic from 0.680 (95% confidence interval [CI], 0.678-0.682) to 0.700 (95% CI, 0.698-0.702; P<0.0001) with a corresponding continuous NRI of 0.227 (95% CI, 0.219-0.235). The prediction of cardiovascular death was improved with the incorporation of 4 lipid species into the base model, showing an increase in the C statistic from 0.740 (95% CI, 0.738-0.742) to 0.760 (95% CI, 0.757-0.762; P<0.0001) and a continuous net reclassification index of 0.328 (95% CI, 0.317-0.339). The results were validated in a subcohort with type 2 diabetes mellitus (n=511) from the LIPID trial (Long-Term Intervention With Pravastatin in Ischemic Disease). CONCLUSIONS: The improvement in the prediction of cardiovascular events, above traditional risk factors, demonstrates the potential of plasma lipid species as biomarkers for cardiovascular risk stratification in diabetes mellitus. CLINICAL TRIAL REGISTRATION: URL: https://clinicaltrials.gov. Unique identifier: NCT00145925.

A distinct plasma lipid signature associated with poor prognosis in castration-resistant prostate cancer.

Lipids are known to influence tumour growth, inflammation and chemoresistance. However, the association of circulating lipids with the clinical outcome of metastatic castration-resistant prostate cancer (CRPC) is unknown. We investigated associations between the plasma lipidome and clinical outcome in CRPC. Lipidomic profiling by liquid chromatography-tandem mass spectrometry was performed on plasma samples from a Phase 1 discovery cohort of 96 CRPC patients. Results were validated in an independent Phase 2 cohort of 63 CRPC patients. Unsupervised analysis of lipidomic profiles (323 lipid species) classified the Phase 1 cohort into two patient subgroups with significant survival differences (HR 2.31, 95% CI 1.44-3.68, p = 0.0005). The levels of 46 lipids were individually prognostic and were predominantly sphingolipids with higher levels associated with poor prognosis. A prognostic three-lipid signature was derived (ceramide d18:1/24:1, sphingomyelin d18:2/16:0, phosphatidylcholine 16:0/16:0) and was also associated with shorter survival in the Phase 2 cohort (HR 4.8, 95% CI 2.06-11.1, p = 0.0003). The signature was an independent prognostic factor when modelled with clinicopathological factors or metabolic characteristics. The association of plasma lipids with CRPC prognosis suggests a possible role of these lipids in disease progression. Further research is required to determine if therapeutic modulation of the levels of these lipids by targeting their metabolic pathways may improve patient outcome.

Gene and sample selection using T-score with sample selection.

Gene selection from high-dimensional microarray gene-expression data is statistically a challenging problem. Filter approaches to gene selection have been popular because of their simplicity, efficiency, and accuracy. Due to small sample size, all samples are generally used to compute relevant ranking statistics and selection of samples in filter-based gene selection methods has not been addressed. In this paper, we extend previously-proposed simultaneous sample and gene selection approach. In a backward elimination method, a modified logistic regression loss function is used to select relevant samples at each iteration, and these samples are used to compute the T-score to rank genes. This method provides a compromise solution between T-score and other support vector machine (SVM) based algorithms. The performance is demonstrated on both simulated and real datasets with criteria such as classification performance, stability and redundancy. Results indicate that computational complexity and stability of the method are improved compared to SVM based methods without compromising the classification performance.

The prediction of type 2 diabetes in women with previous gestational diabetes mellitus using lipidomics.

AIMS/HYPOTHESIS: The risk of developing diabetes is greater for women with previous gestational diabetes mellitus (GDM). In the general population, plasma lipidomic analysis can identify individuals at risk of developing type 2 diabetes. The aim of this study was to determine whether circulating lipid levels 12 weeks following a GDM pregnancy were associated with an increased risk of developing type 2 diabetes. METHODS: Plasma lipid profiles containing >300 lipids were measured in 104 normal glucose-tolerant women 12 weeks following an index GDM pregnancy using electrospray-ionisation tandem mass spectrometry. Women were assessed for 10 years for development of overt type 2 diabetes. RESULTS: Among the 104 women with previous GDM, 21 (20%) developed diabetes during the median follow-up period of 8.5 years. Three lipid species, the cholesteryl ester species CE 20:4, the alkenylphosphatidylethanolamine species PE(P-36:2) and the phosphatidylserine species PS 38:4, were independently and positively associated with the development of type 2 diabetes. In a clinical model of prediction of type 2 diabetes that included age, BMI, and levels of pregnancy fasting glucose, postnatal fasting glucose, triacylglycerol and total cholesterol, the addition of these three lipid species resulted in an improvement in the net reclassification index of 22.3%. CONCLUSIONS/INTERPRETATION: The lipid species CE 20:4, PE(P-36:2) and PS 38:4 were significant risk factors for the development of type 2 diabetes in women with a previous history of GDM. This report is the first to use plasma lipidomic analysis to identify individual lipids as potential biomarkers for the prediction of type 2 diabetes in women with a history of GDM.

Postprandial Plasma Phospholipids in Men Are Influenced by the Source of Dietary Fat.

BACKGROUND: Postprandial lipemia represents a risk factor for chronic diseases, including type 2 diabetes. Little is known about the effect of dietary fat on the plasma lipidome in the postprandial period. OBJECTIVE: The objective of this study was to assess the effect of dairy fat and soy oil on circulating postprandial lipids in men. METHODS: Men (40-60 y old, nonsmokers; n = 16) were randomly assigned in a crossover design to consume 2 breakfast meals of dairy-based or soy oil-based foods. The changes in the plasma lipidome during the 4-h postprandial period were analyzed with electrospray ionization tandem mass spectrometry and included 316 lipid species in 23 classes and subclasses, representing sphingolipids, phospholipids, glycerolipids, and sterols. RESULTS: Nonparametric Friedman tests showed significant changes in multiple plasma lipid classes, subclasses, and species in the postprandial period after both dairy and soy meals. No difference was found in triglyceridemia after each meal. However, 6 endogenous lipid classes increased after dairy but decreased after soy (P < 0.05), including ether-linked phospholipids and plasmalogens and sphingomyelin (not present in soy), dihexosylceramide, and GM3 ganglioside. Phosphatidylcholine and phosphatidylinositol were not affected by the soy meal but were significantly elevated after the dairy meal (8.3% and 16%, respectively; P < 0.05). CONCLUSIONS: The changes in postprandial plasma phospholipids in men relate to the diet composition and the relative size of the endogenous phospholipid pools. Despite similar lipemic responses as measured by changes in triglyceride concentrations, the differential responses to dairy and soy meals derived through lipidomic analysis of phospholipids suggest differences in the metabolism of soybean oil and dairy fat. The increased concentrations of plasmalogens, with potential antioxidant capacity, in the postprandial period after dairy but not soy meals may represent a further important difference in the response to these sources of fat. The trial was registered at www.anzctr.org.au as ACTRN12610000562077.

Frequency of naevus cells in lymph nodes of melanoma and breast cancer patients.

INTRODUCTION: We aimed to study the frequency (prevalence) and histology of benign melanocytic naevus cells in regional lymph nodes in relation to age and sex and nodal location. MATERIAL AND METHODS: Histopathology reports of sentinel lymph node (SLN) biopsies from melanoma patients, 2002 - 2014, and from breast cancer patients, 2010- 2019, were obtained from records of a single hospital in England. All sections were similarly processed and examined. For standardisation, presence of naevus cells was assessed in a single node per patient: the first SLN biopsied (melanoma) or the node nearest the first SLN (breast cancer). RESULTS: Associations were tested using Fisher's exact test. Naevus cells were found in 10% (60/585) of melanoma patients' index SLNs. Frequency varied significantly by anatomic region: 13% in axillary to 0% cervical SLNs (p = 0.03), but not by sex or age. Within nodes, naevus cells were present in capsular or pericapsular tissue (93%), or trabeculae (7%). In breast cancer patients' index axillary nodes, 6% (11/196) contained naevus cells, all intracapsular. In the predominant 40-69 years age-group, prevalence was similar in breast cancer (7%) and female melanoma (9%) patients, but in those aged 70-100, prevalence was lower in breast cancer (2%) than in female melanoma (15%) patients (p = 0.05). CONCLUSIONS: Standard methods of assessment yielded no clear pattern of naevus cell frequency in lymph nodes by age or sex, but confirmed naevus cell location as mostly intracapsular.

Functional classification of DDOST variants of uncertain clinical significance in congenital disorders of glycosylation.

Congenital disorders of glycosylation (CDG) are rare genetic disorders with a spectrum of clinical manifestations caused by abnormal N-glycosylation of secreted and cell surface proteins. Over 130 genes are implicated and next generation sequencing further identifies potential disease drivers in affected individuals. However, functional testing of these variants is challenging, making it difficult to distinguish pathogenic from non-pathogenic events. Using proximity labelling, we identified OST48 as a protein that transiently interacts with lysyl oxidase (LOX), a secreted enzyme that cross-links the fibrous extracellular matrix. OST48 is a non-catalytic component of the oligosaccharyltransferase (OST) complex, which transfers glycans to substrate proteins. OST48 is encoded by DDOST, and 43 variants of DDOST are described in CDG patients, of which 34 are classified as variants of uncertain clinical significance (VUS). We developed an assay based on LOX N-glycosylation that confirmed two previously characterised DDOST variants as pathogenic. Notably, 39 of the 41 remaining variants did not have impaired activity, but we demonstrated that p.S243F and p.E286del were functionally impaired, consistent with a role in driving CDG in those patients. Thus, we describe a rapid assay for functional testing of clinically relevant CDG variants to complement genome sequencing and support clinical diagnosis of affected individuals.

Heritable defects in telomere and mitotic function selectively predispose to sarcomas.

Cancer genetics has to date focused on epithelial malignancies, identifying multiple histotype-specific pathways underlying cancer susceptibility. Sarcomas are rare malignancies predominantly derived from embryonic mesoderm. To identify pathways specific to mesenchymal cancers, we performed whole-genome germline sequencing on 1644 sporadic cases and 3205 matched healthy elderly controls. Using an extreme phenotype design, a combined rare-variant burden and ontologic analysis identified two sarcoma-specific pathways involved in mitotic and telomere functions. Variants in centrosome genes are linked to malignant peripheral nerve sheath and gastrointestinal stromal tumors, whereas heritable defects in the shelterin complex link susceptibility to sarcoma, melanoma, and thyroid cancers. These studies indicate a specific role for heritable defects in mitotic and telomere biology in risk of sarcomas.

Tumour infiltrating B cells discriminate checkpoint blockade-induced responses.

BACKGROUND: Immune cell-driven anti-cancer activity is paramount for effective responses to checkpoint inhibitors (ICB). However, the contribution of the different immune cell subsets in the circulation and within the tumour is poorly understood. MATERIALS AND METHODS: To elucidate the role of the different cell subsets in anti-tumour responses elicited by ICB, we performed single-cell analysis of the transcriptome and surface proteome of paired pre- and early on-treatment metastatic melanoma tumour biopsies and matched peripheral blood mononuclear cell samples. We next compared the survival of metastatic melanoma patients treated with ICB according to the abundance of pre-treatment tumour-infiltrating B cell clonotypes. RESULTS: We identified cell clusters associated with disease control or progression, defined differential expression of biological pathways likely involved in the immune awakening against the tumour and examined how cell-cell communication patterns between the tumour cell subsets change during treatment. Furthermore, we discovered that B cells (immunoglobulin expression and abundance of B cell clonotypes) discriminate the clinical response after ICB and propose that B cells likely contribute to anti-tumour immunity by antigen presentation through major histocompatibility complex molecules. Finally, we demonstrated that the abundance of tumour-infiltrating B cell clonotypes at baseline identifies two distinct risk groups, a finding that we confirmed in an independent cohort. CONCLUSIONS: Our exploratory translational study provides new insights on the mechanistic role of B cells in anti-melanoma immunity during treatment with ICB. Additionally, we support pre-treatment B cell tumour infiltration as a promising prognostic biomarker to be further validated as a tool for clinical risk stratification.

Stability of building gene regulatory networks with sparse autoregressive models.

BACKGROUND: Biological networks are constantly subjected to random perturbations, and efficient feedback and compensatory mechanisms exist to maintain their stability. There is an increased interest in building gene regulatory networks (GRNs) from temporal gene expression data because of their numerous applications in life sciences. However, because of the limited number of time points at which gene expressions can be gathered in practice, computational techniques of building GRN often lead to inaccuracies and instabilities. This paper investigates the stability of sparse auto-regressive models of building GRN from gene expression data. RESULTS: Criteria for evaluating the stability of estimating GRN structure are proposed. Thereby, stability of multivariate vector autoregressive (MVAR) methods - ridge, lasso, and elastic-net - of building GRN were studied by simulating temporal gene expression datasets on scale-free topologies as well as on real data gathered over Hela cell-cycle. Effects of the number of time points on the stability of constructing GRN are investigated. When the number of time points are relatively low compared to the size of network, both accuracy and stability are adversely affected. At least, the number of time points equal to the number of genes in the network are needed to achieve decent accuracy and stability of the networks. Our results on synthetic data indicate that the stability of lasso and elastic-net MVAR methods are comparable, and their accuracies are much higher than the ridge MVAR. As the size of the network grows, the number of time points required to achieve acceptable accuracy and stability are much less relative to the number of genes in the network. The effects of false negatives are easier to improve by increasing the number time points than those due to false positives. Application to HeLa cell-cycle gene expression dataset shows that biologically stable GRN can be obtained by introducing perturbations to the data. CONCLUSIONS: Accuracy and stability of building GRN are crucial for investigation of gene regulations. Sparse MVAR techniques such as lasso and elastic-net provide accurate and stable methods for building even GRN of small size. The effect of false negatives is corrected much easier with the increased number of time points than those due to false positives. With real data, we demonstrate how stable networks can be derived by introducing random perturbation to data.

The T cell receptor repertoire of tumor infiltrating T cells is predictive and prognostic for cancer survival.

Tumor infiltration by T cells is paramount for effective anti-cancer immune responses. We hypothesized that the T cell receptor (TCR) repertoire of tumor infiltrating T lymphocytes could therefore be indicative of the functional state of these cells and determine disease course at different stages in cancer progression. Here we show that the diversity of the TCR of tumor infiltrating T cell at baseline is prognostic in various cancers, whereas the TCR clonality of T cell infiltrating metastatic melanoma pre-treatment is predictive for activity and efficacy of PD1 blockade immunotherapy.

Ultraviolet radiation drives mutations in a subset of mucosal melanomas.

Although identified as the key environmental driver of common cutaneous melanoma, the role of ultraviolet radiation (UVR)-induced DNA damage in mucosal melanoma is poorly defined. We analyze 10 mucosal melanomas of conjunctival origin by whole genome sequencing and our data shows a predominance of UVR-associated single base substitution signature 7 (SBS7) in the majority of the samples. Our data shows mucosal melanomas with SBS7 dominance have similar genomic patterns to cutaneous melanomas and therefore this subset should not be excluded from treatments currently used for common cutaneous melanoma.

Bayesian Data Fusion of Gene Expression and Histone Modification Profiles for Inference of Gene Regulatory Network.

Accurately reconstructing gene regulatory networks (GRNs) from high-throughput gene expression data has been a major challenge in systems biology for decades. Many approaches have been proposed to solve this problem. However, there is still much room for the improvement of GRN inference. Integrating data from different sources is a promising strategy. Epigenetic modifications have a close relationship with gene regulation. Hence, epigenetic data such as histone modification profiles can provide useful information for uncovering regulatory interactions between genes. In this paper, we propose a method to integrate epigenetic data into the inference of GRNs. In particular, a dynamic Bayesian network (DBN) is employed to infer gene regulations from time-series gene expression data. Epigenetic data (histone modification profiles here) are integrated into the prior probability distribution of the Bayesian model. Our method has been validated on both synthetic and real datasets. Experimental results show that the integration of epigenetic data can significantly improve the performance of GRN inference. As more epigenetic datasets become available, our method would be useful for elucidating the gene regulatory mechanisms driving various cellular activities. The source code and testing datasets are available at https://github.com/Zheng-Lab/MetaGRN/tree/master/histonePrior.

Immune-awakening revealed by peripheral T cell dynamics after one cycle of immunotherapy.

Our understanding of how checkpoint inhibitors (CPI) affect T cell evolution is incomplete, limiting our ability to achieve full clinical benefit from these drugs. Here we analyzed peripheral T cell populations after one cycle of CPI and identified a dynamic awakening of the immune system revealed by T cell evolution in response to treatment. We sequenced T cell receptors (TCR) in plasma cell-free DNA (cfDNA) and peripheral blood mononuclear cells (PBMC) and performed phenotypic analysis of peripheral T cell subsets from metastatic melanoma patients treated with CPI. We found that early peripheral T cell turnover and TCR repertoire dynamics identified which patients would respond to treatment. Additionally, the expansion of a subset of immune-effector peripheral T cells we call TIE cells correlated with response. These events are prognostic and occur within 3 weeks of starting immunotherapy, raising the potential for monitoring patients responses using minimally invasive liquid biopsies."

Stroma remodeling and reduced cell division define durable response to PD-1 blockade in melanoma.

Although immune checkpoint inhibitors (ICIs) have achieved unprecedented results in melanoma, the biological features of the durable responses initiated by these drugs remain unknown. Here we show the genetic and phenotypic changes induced by treatment with programmed cell death-1 (PD-1) blockade in a genetically engineered mouse model of melanoma driven by oncogenic BRAF. In this controlled system anti-PD-1 treatment yields responses in ~35% of the tumors, and prolongs survival in ~27% of the animals. We identify increased stroma remodeling and reduced expression of proliferation markers as features associated with prolonged response. These traits are corroborated in two independent early on-treatment anti-PD-1 melanoma patient cohorts. These insights into the biological responses of tumors to ICI provide a strategy for identification of durable response early during the course of treatment and could improve patient stratification for checkpoint inhibitory drugs.

Ultraviolet radiation-induced DNA damage is prognostic for outcome in melanoma.

The melanoma genome is dominated by ultraviolet radiation (UVR)-induced mutations. Their relevance in disease progression is unknown. Here we classify melanomas by mutation signatures and identify ten recurrently mutated UVR signature genes that predict patient survival. We validate these findings in primary human melanomas; in mice we show that this signature is imprinted by short-wavelength UVR and that four exposures to UVR are sufficient to accelerate melanomagenesis.