Originator: Computational Framework Separating Single-Cell RNA-Seq by Genetic and Contextual Information.

Single-cell RNA sequencing data from complex human tissues are prone to blood contamination in sample preparation, and some comprise cells of different genetic makeups, necessitating rigorous preprocessing and cell filtering prior to the downstream functional analysis. Our proposed new computational framework, Originator, deciphers single cells by genetic origin and separates blood cells from tissue-resident cells. It improves the quality of data analysis, exemplified by pancreatic ductal adenocarcinoma and placenta tissues.

Challenges and perspectives in computational deconvolution of genomics data.

Deciphering cell-type heterogeneity is crucial for systematically understanding tissue homeostasis and its dysregulation in diseases. Computational deconvolution is an efficient approach for estimating cell-type abundances from a variety of omics data. Despite substantial methodological progress in computational deconvolution in recent years, challenges are still outstanding. Here we enlist four important challenges related to computational deconvolution: the quality of the reference data, generation of ground truth data, limitations of computational methodologies, and benchmarking design and implementation. Finally, we make recommendations on reference data generation, new directions of computational methodologies, and strategies to promote rigorous benchmarking.

Deep learning and transfer learning identify breast cancer survival subtypes from single-cell imaging data.

BACKGROUND: Single-cell multiplex imaging data have provided new insights into disease subtypes and prognoses recently. However, quantitative models that explicitly capture single-cell resolution cell-cell interaction features to predict patient survival at a population scale are currently missing. METHODS: We quantified hundreds of single-cell resolution cell-cell interaction features through neighborhood calculation, in addition to cellular phenotypes. We applied these features to a neural-network-based Cox-nnet survival model to identify survival-associated features. We used non-negative matrix factorization (NMF) to identify patient survival subtypes. We identified atypical subpopulations of triple-negative breast cancer (TNBC) patients with moderate prognosis and Luminal A patients with poor prognosis and validated these subpopulations by label transferring using the UNION-COM method. RESULTS: The neural-network-based Cox-nnet survival model using all cellular phenotype and cell-cell interaction features is highly predictive of patient survival in the test data (Concordance Index > 0.8). We identify seven survival subtypes using the top survival features, presenting distinct profiles of epithelial, immune, and fibroblast cells and their interactions. We reveal atypical subpopulations of TNBC patients with moderate prognosis (marked by GATA3 over-expression) and Luminal A patients with poor prognosis (marked by KRT6 and ACTA2 over-expression and CDH1 under-expression). These atypical subpopulations are validated in TCGA-BRCA and METABRIC datasets. CONCLUSIONS: This work provides an approach to bridge single-cell level information toward population-level survival prediction.

It may be possible to separate patients with cancer into different groups or subtypes based on the features of their tumor, such as the interactions between different types of cells in the tumor. In this study, we develop a computer-based model to calculate the interactions between cells in breast cancer images. We use these interactions to identify seven subtypes of patients with breast cancer with differences in their survival. We identify some subpopulations of patients with atypical survival outcomes. This work may ultimately help clinicians and researchers to identify patients with breast cancer at increased risk of poorer outcomes and to tailor their treatments accordingly.

Prediction of spontaneous preterm birth using supervised machine learning on metabolomic data: A case-cohort study.

OBJECTIVES: To identify and internally validate metabolites predictive of spontaneous preterm birth (sPTB) using multiple machine learning methods and sequential maternal serum samples, and to predict spontaneous early term birth (sETB) using these metabolites. DESIGN: Case-cohort design within a prospective cohort study. SETTING: Cambridge, UK. POPULATION OR SAMPLE: A total of 399 Pregnancy Outcome Prediction study participants, including 98 cases of sPTB. METHODS: An untargeted metabolomic analysis of maternal serum samples at 12, 20, 28 and 36 weeks of gestation was performed. We applied six supervised machine learning methods and a weighted Cox model to measurements at 28 weeks of gestation and sPTB, followed by feature selection. We used logistic regression with elastic net penalty, followed by best subset selection, to reduce the number of predictive metabolites further. We applied coefficients from the chosen models to measurements from different gestational ages to predict sPTB and sETB. MAIN OUTCOME MEASURES: sPTB and sETB. RESULTS: We identified 47 metabolites, mostly lipids, as important predictors of sPTB by two or more methods and 22 were identified by three or more methods. The best 4-predictor model had an optimism-corrected area under the receiver operating characteristics curve (AUC) of 0.703 at 28 weeks of gestation. The model also predicted sPTB in 12-week samples (0.606, 95% CI 0.544-0.667) and 20-week samples (0.657, 95% CI 0.597-0.717) and it predicted sETB in 36-week samples (0.727, 95% CI 0.606-0.849). A lysolipid, 1-palmitoleoyl-GPE (16:1)*, was the strongest predictor of sPTB at 12 weeks of gestation (0.609, 95% CI 0.548-0.670), 20 weeks (0.630, 95% CI 0.569-0.690) and 28 weeks (0.660, 95% CI 0.599-0.722), and of sETB at 36 weeks (0.739, 95% CI 0.618-0.860). CONCLUSIONS: We identified and internally validated maternal serum metabolites predictive of sPTB. A lysolipid, 1-palmitoleoyl-GPE (16:1)*, is a novel predictor of sPTB and sETB. Further validation in external populations is required.

Deep-learning and transfer learning identify new breast cancer survival subtypes from single-cell imaging data.

Quantitative models that explicitly capture single-cell resolution cell-cell interaction features to predict patient survival at population scale are currently missing. Here, we computationally extracted hundreds of features describing single-cell based cell-cell interactions and cellular phenotypes from a large, published cohort of cyto-images of breast cancer patients. We applied these features to a neural-network based Cox-nnet survival model and obtained high accuracy in predicting patient survival in test data (Concordance Index > 0.8). We identified seven survival subtypes using the top survival features, which present distinct profiles of epithelial, immune, fibroblast cells, and their interactions. We identified atypical subpopulations of TNBC patients with moderate prognosis (marked by GATA3 over-expression) and Luminal A patients with poor prognosis (marked by KRT6 and ACTA2 over-expression and CDH1 under-expression). These atypical subpopulations are validated in TCGA-BRCA and METABRIC datasets. This work provides important guidelines on bridging single-cell level information towards population-level survival prediction. STATEMENT OF TRANSLATIONAL RELEVANCE: Our findings from a breast cancer population cohort demonstrate the clinical utility of using the single-cell level imaging mass cytometry (IMC) data as a new type of patient prognosis prediction marker. Not only did the prognosis prediction achieve high accuracy with a Concordance index score greater than 0.8, it also enabled the discovery of seven survival subtypes that are more distinguishable than the molecular subtypes. These new subtypes present distinct profiles of epithelial, immune, fibroblast cells, and their interactions. Most importantly, this study identified and validated atypical subpopulations of TNBC patients with moderate prognosis (GATA3 over-expression) and Luminal A patients with poor prognosis (KRT6 and ACTA2 over-expression and CDH1 under-expression), using multiple large breast cancer cohorts.

Severe preeclampsia is not associated with significant DNA methylation changes but cell proportion changes in the cord blood - caution on the importance of confounding adjustment.

Epigenome-wide DNA methylation analysis (EWAS) is an important approach to identify biomarkers for early disease detection and prognosis prediction, yet its results could be confounded by other factors such as cell-type heterogeneity and patient characteristics. In this study, we address the importance of confounding adjustment by examining DNA methylation patterns in cord blood exposed to severe preeclampsia (PE), a prevalent and potentially fatal pregnancy complication. Without such adjustment, a misleading global hypomethylation pattern is obtained. However, after adjusting cell type proportions and patient clinical characteristics, most of the so-called significant CpG methylation changes associated with severe PE disappear. Rather, the major effect of PE on cord blood is through the proportion changes in different cell types. These results are validated using a previously published cord blood DNA methylation dataset, where global hypomethylation pattern was also wrongfully obtained without confounding adjustment. Additionally, several cell types significantly change as gestation progress (eg. granulocyte, nRBC, CD4T, and B cells), further confirming the importance of cell type adjustment in EWAS study of cord blood tissues. Our study urges the community to perform confounding adjustments in EWAS studies, based on cell type heterogeneity and other patient characteristics.

ASGARD is A Single-cell Guided Pipeline to Aid Repurposing of Drugs.

Single-cell RNA sequencing technology has enabled in-depth analysis of intercellular heterogeneity in various diseases. However, its full potential for precision medicine has yet to be reached. Towards this, we propose A Single-cell Guided Pipeline to Aid Repurposing of Drugs (ASGARD) that defines a drug score to recommend drugs by considering all cell clusters to address the intercellular heterogeneity within each patient. ASGARD shows significantly better average accuracy on single-drug therapy compared to two bulk-cell-based drug repurposing methods. We also demonstrated that it performs considerably better than other cell cluster-level predicting methods. In addition, we validate ASGARD using the drug response prediction method TRANSACT with Triple-Negative-Breast-Cancer patient samples. We find that many top-ranked drugs are either approved by the Food and Drug Administration or in clinical trials treating corresponding diseases. In conclusion, ASGARD is a promising drug repurposing recommendation tool guided by single-cell RNA-seq for personalized medicine. ASGARD is free for educational use at https://github.com/lanagarmire/ASGARD .

Computational Methods for Single-cell Multi-omics Integration and Alignment.

Recently developed technologies to generate single-cell genomic data have made a revolutionary impact in the field of biology. Multi-omics assays offer even greater opportunities to understand cellular states and biological processes. The problem of integrating different omics data with very different dimensionality and statistical properties remains, however, quite challenging. A growing body of computational tools is being developed for this task, leveraging ideas ranging from machine translation to the theory of networks, and represents another frontier on the interface of biology and data science. Our goal in this review is to provide a comprehensive, up-to-date survey of computational techniques for the integration of single-cell multi-omics data, while making the concepts behind each algorithm approachable to a non-expert audience.

Placental telomere length shortening is not associated with severe preeclampsia but the gestational age.

Variations in telomere length (TL) have been associated with aging, stress, and many diseases. Placenta TL is an essential molecular component influencing the outcome of birth. Previous investigations into the relationship between placenta TL and preeclampsia (PE) have produced conflicting findings. We conducted a retrospective case-control analysis in this study to address the disparity. We used placenta samples from 224 births received from Hawaii Biorepository (HiBR) between 2006 and 2013, comprising 129 healthy full-term controls and 95 severe PE samples. The average absolute placental TL was calculated using the quantitative polymerase chain reaction (qPCR) technique. We utilized multiple linear regressions to associate placental TL with severe PE and other demographic, clinical and physiological data. The outcome demonstrates that the placental TL of severe PE cases did not significantly differ from that of healthy controls. Instead, there is a strong correlation between gestational age and placenta TL shortening. Placental TL also exhibits racial differences: (1) Latino moms' TL is significantly longer than non-Latino mothers' (p=0.009). (2) Caucasian patients with severe PE have shorter TL than non-Caucasian patients (p=0.0037). This work puts the long-standing question of whether severe PE influences placental TL to rest. Placental TL is not related to severe PE but is negatively associated with gestational age and is also affected by race.

Heterogeneity of neutrophils and inflammatory responses in patients with COVID-19 and healthy controls.

Severe respiratory viral infections, including SARS-CoV-2, have resulted in high mortality rates despite corticosteroids and other immunomodulatory therapies. Despite recognition of the pathogenic role of neutrophils, in-depth analyses of this cell population have been limited, due to technical challenges of working with neutrophils. We undertook an unbiased, detailed analysis of neutrophil responses in adult patients with COVID-19 and healthy controls, to determine whether distinct neutrophil phenotypes could be identified during infections compared to the healthy state. Single-cell RNA sequencing analysis of peripheral blood neutrophils from hospitalized patients with mild or severe COVID-19 disease and healthy controls revealed distinct mature neutrophil subpopulations, with relative proportions linked to disease severity. Disruption of predicted cell-cell interactions, activated oxidative phosphorylation genes, and downregulated antiviral and host defense pathway genes were observed in neutrophils obtained during severe compared to mild infections. Our findings suggest that during severe infections, there is a loss of normal regulatory neutrophil phenotypes seen in healthy subjects, coupled with the dropout of appropriate cellular interactions. Given that neutrophils are the most abundant circulating leukocytes with highly pathogenic potential, current immunotherapies for severe infections may be optimized by determining whether they aid in restoring an appropriate balance of neutrophil subpopulations.

Blood-derived lncRNAs as biomarkers for cancer diagnosis: the Good, the Bad and the Beauty.

Cancer ranks as one of the deadliest diseases worldwide. The high mortality rate associated with cancer is partially due to the lack of reliable early detection methods and/or inaccurate diagnostic tools such as certain protein biomarkers. Cell-free nucleic acids (cfNA) such as circulating long noncoding RNAs (lncRNAs) have been proposed as a new class of potential biomarkers for cancer diagnosis. The reported correlation between the presence of tumors and abnormal levels of lncRNAs in the blood of cancer patients has notably triggered a worldwide interest among clinicians and oncologists who have been actively investigating their potentials as reliable cancer biomarkers. In this report, we review the progress achieved ("the Good") and challenges encountered ("the Bad") in the development of circulating lncRNAs as potential biomarkers for early cancer diagnosis. We report and discuss the diagnostic performance of more than 50 different circulating lncRNAs and emphasize their numerous potential clinical applications ("the Beauty") including therapeutic targets and agents, on top of diagnostic and prognostic capabilities. This review also summarizes the best methods of investigation and provides useful guidelines for clinicians and scientists who desire conducting their own clinical studies on circulating lncRNAs in cancer patients via RT-qPCR or Next Generation Sequencing (NGS).

Emerging artificial intelligence applications in Spatial Transcriptomics analysis.

Spatial transcriptomics (ST) has advanced significantly in the last few years. Such advancement comes with the urgent need for novel computational methods to handle the unique challenges of ST data analysis. Many artificial intelligence (AI) methods have been developed to utilize various machine learning and deep learning techniques for computational ST analysis. This review provides a comprehensive and up-to-date survey of current AI methods for ST analysis.

Maternal plasma lipids are involved in the pathogenesis of preterm birth.

BACKGROUND: Preterm birth is defined by the onset of labor at a gestational age shorter than 37 weeks, and it can lead to premature birth and impose a threat to newborns' health. The Puerto Rico PROTECT cohort is a well-characterized prospective birth cohort that was designed to investigate environmental and social contributors to preterm birth in Puerto Rico, where preterm birth rates have been elevated in recent decades. To elucidate possible relationships between metabolites and preterm birth in this cohort, we conducted a nested case-control study to conduct untargeted metabolomic characterization of maternal plasma of 31 women who experienced preterm birth and 69 controls who underwent full-term labor at 24-28 gestational weeks. RESULTS: A total of 333 metabolites were identified and annotated with liquid chromatography/mass spectrometry. Subsequent weighted gene correlation network analysis shows that the fatty acid and carene-enriched module has a significant positive association (P = 8e-04, FDR = 0.006) with preterm birth. After controlling for potential clinical confounders, a total of 38 metabolites demonstrated significant changes uniquely associated with preterm birth, where 17 of them were preterm biomarkers. Among 7 machine-learning classifiers, the application of random forest achieved a highly accurate and specific prediction (AUC = 0.92) for preterm birth in testing data, demonstrating their strong potential as biomarkers for preterm births. The 17 preterm biomarkers are involved in cell signaling, lipid metabolism, and lipid peroxidation functions. Additional modeling using only the 19 spontaneous preterm births (sPTB) and controls identifies 16 sPTB markers, with an AUC of 0.89 in testing data. Half of the sPTB overlap with those markers for preterm births. Further causality analysis infers that suberic acid upregulates several fatty acids to promote preterm birth. CONCLUSIONS: Altogether, this study demonstrates the involvement of lipids, particularly fatty acids, in the pathogenesis of preterm birth.

A Dual-Filtration System for Single-Cell Sequencing of Circulating Tumor Cells and Clusters in HCC.

Hepatocellular carcinoma (HCC) is a leading cause of cancer death worldwide. Identification and sequencing of circulating tumor (CT) cells and clusters may allow for noninvasive molecular characterization of HCC, which is an unmet need, as many patients with HCC do not undergo biopsy. We evaluated CT cells and clusters, collected using a dual-filtration system in patients with HCC. We collected and filtered whole blood from patients with HCC and selected individual CT cells and clusters with a micropipette. Reverse transcription, polymerase chain reaction, and library preparation were performed using a SmartSeq2 protocol, followed by single-cell RNA sequencing (scRNAseq) on an Illumina MiSeq V3 platform. Of the 8 patients recruited, 6 had identifiable CT cells or clusters. Median age was 64 years old; 7 of 8 were male; and 7 of 8 had and Barcelona Clinic Liver Cancer stage C. We performed scRNAseq of 38 CT cells and 33 clusters from these patients. These CT cells and clusters formed two distinct groups. Group 1 had significantly higher expression than group 2 of markers associated with epithelial phenotypes (CDH1 [Cadherin 1], EPCAM [epithelial cell adhesion molecule], ASGR2 [asialoglycoprotein receptor 2], and KRT8 [Keratin 8]), epithelial-mesenchymal transition (VIM [Vimentin]), and stemness (PROM1 [CD133], POU5F1 [POU domain, class 5, transcription factor 1], NOTCH1, STAT3 [signal transducer and activator of transcription 3]) (P < 0.05 for all). Patients with identifiable group 1 cells or clusters had poorer prognosis than those without them (median overall survival 39 vs. 384 days; P = 0.048 by log-rank test). Conclusion: A simple dual-filtration system allows for isolation and sequencing of CT cells and clusters in HCC and may identify cells expressing candidate genes known to be involved in cancer biology. Presence of CT cells/clusters expressing candidate genes is associated with poorer prognosis in advanced-stage HCC.

BML: a versatile web server for bipartite motif discovery.

Motif discovery and characterization are important for gene regulation analysis. The lack of intuitive and integrative web servers impedes the effective use of motifs. Most motif discovery web tools are either not designed for non-expert users or lacking optimization steps when using default settings. Here we describe bipartite motifs learning (BML), a parameter-free web server that provides a user-friendly portal for online discovery and analysis of sequence motifs, using high-throughput sequencing data as the input. BML utilizes both position weight matrix and dinucleotide weight matrix, the latter of which enables the expression of the interdependencies of neighboring bases. With input parameters concerning the motifs are given, the BML achieves significantly higher accuracy than other available tools for motif finding. When no parameters are given by non-expert users, unlike other tools, BML employs a learning method to identify motifs automatically and achieve accuracy comparable to the scenario where the parameters are set. The BML web server is freely available at http://motif.t-ridership.com/ (https://github.com/Mohammad-Vahed/BML).

ASGARD: A Single-cell Guided pipeline to Aid Repurposing of Drugs.

Intercellular heterogeneity is a major obstacle to successful precision medicine. Single-cell RNA sequencing (scRNA-seq) technology has enabled in-depth analysis of intercellular heterogeneity in various diseases. However, its full potential for precision medicine has yet to be reached. Towards this, we propose a new drug recommendation system called: A Single-cell Guided Pipeline to Aid Repurposing of Drugs (ASGARD). ASGARD defines a novel drug score predicting drugs by considering all cell clusters to address the intercellular heterogeneity within each patient. We tested ASGARD on multiple diseases, including breast cancer, acute lymphoblastic leukemia, and coronavirus disease 2019 (COVID-19). On single-drug therapy, ASGARD shows significantly better average accuracy (AUC of 0.92) compared to two other bulk-cell-based drug repurposing methods (AUC of 0.80 and 0.76). It is also considerably better (AUC of 0.82) than other cell cluster level predicting methods (AUC of 0.67 and 0.55). In addition, ASGARD is also validated by the drug response prediction method TRANSACT with Triple-Negative-Breast-Cancer patient samples. Many top-ranked drugs are either approved by FDA or in clinical trials treating corresponding diseases. In silico cell-type specific drop-out experiments using triple-negative breast cancers show the importance of T cells in the tumor microenvironment in affecting drug predictions. In conclusion, ASGARD is a promising drug repurposing recommendation tool guided by single-cell RNA-seq for personalized medicine. ASGARD is free for educational use at https://github.com/lanagarmire/ASGARD.

The maternal blood lipidome is indicative of the pathogenesis of severe preeclampsia.

Preeclampsia is a pregnancy-specific syndrome characterized by hypertension and proteinuria after 20 weeks of gestation. However, it is not well understood what lipids are involved in the development of this condition, and even less is known how these lipids mediate its formation. To reveal the relationship between lipids and preeclampsia, we conducted lipidomic profiling of maternal sera of 44 severe preeclamptic and 20 healthy pregnant women from a multiethnic cohort in Hawaii. Correlation network analysis showed that oxidized phospholipids have increased intercorrelations and connections in preeclampsia, whereas other lipids, including triacylglycerols, have reduced network correlations and connections. A total of 10 lipid species demonstrate significant changes uniquely associated with preeclampsia but not any other clinical confounders. These species are from the lipid classes of lysophosphatidylcholines, phosphatidylcholines (PCs), cholesteryl esters, phosphatidylethanolamines, lysophosphatidylethanolamines, and ceramides. A random forest classifier built on these lipids shows highly accurate and specific prediction (F1 statistic = 0.94; balanced accuracy = 0.88) of severe preeclampsia, demonstrating their potential as biomarkers for this condition. These lipid species are enriched in dysregulated biological pathways, including insulin signaling, immune response, and phospholipid metabolism. Moreover, causality inference shows that various PCs and lysophosphatidylcholines mediate severe preeclampsia through PC 35:1e. Our results suggest that the lipidome may play a role in the pathogenesis and serve as biomarkers of severe preeclampsia.

Computational reconstruction of the signalling networks surrounding implanted biomaterials from single-cell transcriptomics.

The understanding of the foreign-body responses to implanted biomaterials would benefit from the reconstruction of intracellular and intercellular signalling networks in the microenvironment surrounding the implant. Here, by leveraging single-cell RNA-sequencing data from 42,156 cells collected from the site of implantation of either polycaprolactone or an extracellular-matrix-derived scaffold in a mouse model of volumetric muscle loss, we report a computational analysis of intercellular signalling networks reconstructed from predictions of transcription-factor activation. We found that intercellular signalling networks can be clustered into modules associated with specific cell subsets, and that biomaterial-specific responses can be characterized by interactions between signalling modules for immune, fibroblast and tissue-specific cells. In a Il17ra-/- mouse model, we validated that predicted interleukin-17-linked transcriptional targets led to concomitant changes in gene expression. Moreover, we identified cell subsets that had not been implicated in the responses to implanted biomaterials. Single-cell atlases of the cellular responses to implanted biomaterials will facilitate the design of implantable biomaterials and the understanding of the ensuing cellular responses.

DeepProg: an ensemble of deep-learning and machine-learning models for prognosis prediction using multi-omics data.

Multi-omics data are good resources for prognosis and survival prediction; however, these are difficult to integrate computationally. We introduce DeepProg, a novel ensemble framework of deep-learning and machine-learning approaches that robustly predicts patient survival subtypes using multi-omics data. It identifies two optimal survival subtypes in most cancers and yields significantly better risk-stratification than other multi-omics integration methods. DeepProg is highly predictive, exemplified by two liver cancer (C-index 0.73-0.80) and five breast cancer datasets (C-index 0.68-0.73). Pan-cancer analysis associates common genomic signatures in poor survival subtypes with extracellular matrix modeling, immune deregulation, and mitosis processes. DeepProg is freely available at https://github.com/lanagarmire/DeepProg.

Two-stage Cox-nnet: biologically interpretable neural-network model for prognosis prediction and its application in liver cancer survival using histopathology and transcriptomic data.

Pathological images are easily accessible data with the potential of prognostic biomarkers. Moreover, integration of heterogeneous data types from multi-modality, such as pathological image and gene expression data, is invaluable to help predicting cancer patient survival. However, the analytical challenges are significant. Here, we take the hepatocellular carcinoma (HCC) pathological image features extracted by CellProfiler, and apply them as the input for Cox-nnet, a neural network-based prognosis prediction model. We compare this model with the conventional Cox proportional hazards (Cox-PH) model, CoxBoost, Random Survival Forests and DeepSurv, using C-index and log-rank P-values. The results show that Cox-nnet is significantly more accurate than Cox-PH and Random Survival Forests models and comparable with CoxBoost and DeepSurv models, on pathological image features. Further, to integrate pathological image and gene expression data of the same patients, we innovatively construct a two-stage Cox-nnet model, and compare it with another complex neural-network model called PAGE-Net. The two-stage Cox-nnet complex model combining histopathology image and transcriptomic RNA-seq data achieves much better prognosis prediction, with a median C-index of 0.75 and log-rank P-value of 6e-7 in the testing datasets, compared to PAGE-Net (median C-index of 0.68 and log-rank P-value of 0.03). Imaging features present additional predictive information to gene expression features, as the combined model is more accurate than the model with gene expression alone (median C-index 0.70). Pathological image features are correlated with gene expression, as genes correlated to top imaging features present known associations with HCC patient survival and morphogenesis of liver tissue. This work proposes two-stage Cox-nnet, a new class of biologically relevant and interpretable models, to integrate multiple types of heterogenous data for survival prediction.