Posters

Poster presentations at ISMB 2020 will be presented virtually. Authors will pre-record their poster talk (5-7 minutes) and will upload it to the virtual conference platform site along with a PDF of their poster. All registered conference participants will have access to the poster and presentation through the conference and content until October 31, 2020. There are Q&A opportunities through a chat function to allow interaction between presenters and participants.

Preliminary information on preparing your poster and poster talk are available at: https://www.iscb.org/ismb2020-general/presenterinfo#posters

Ideally authors should be available for interactive chat during the times noted below:

Search:

View Posters By Category

Poster Session A: July 13 & July 14 7:45 am - 9:15 am Eastern Daylight Time	Session B: July 15 and July 16 between 7:45 am - 9:15 am Eastern Daylight Time
Bioinfo-core CAMDA COSI COVID-19 Education COSI EvoCompGen COSI Function / CAFA 4 MLCSB COSI SCANGEN (Special Session) SysMod COSI Systems Immunology (Special Session) Text Mining July 14 between 10:40 am - 2:00 pm EDT iRNA COSI	3DSIG COSI Bio-Ontologies COSI BioVis COSI CompMS COSI COVID-19 EvoCompGen COSI HitSeq COSI MICROBIOME COSI NetBio COSI RegSys COSI TransMed COSI VarI COSI General Comp Bio

Short Abstract: The Cancer Dependency Map (DepMap) projects recently employed genome-scale CRISPR-Cas9 loss-of-function screens to identify genes essential for cancer cell proliferation and survival across cancer cell lines. However, it remains very challenging to translate these in vitro results to impracticable-to-screen tumors. To address the challenge, we devised a deep learning model with a unique transfer-learning framework to predict gene dependencies of tumors. The model has a 3-stage design that enables a representation learning of unlabeled tumor genomic data, the prediction of gene dependencies in labeled cell-line screening data, and the application to predict tumor dependencies. The prediction performance was verified using cell-line data. Applying our model to ~8,000 tumors of The Cancer Genome Atlas, we constructed a pan-cancer dependency map of tumors. The results were confirmed by several biomarkers and the response to targeted therapies of the TCGA clinical records. Further investigations revealed gene dependencies associated with specific genomic patterns, such as higher tumor mutation burdens and unique expression/methylation signatures. We also identified highly selective gene dependencies of which inhibitor drugs have been approved to treat cancers. We expect the model to evolve with rapidly developing in vitro CRISPR-Cas9 viability screens and facilitate the translation to identifying therapeutic targets of tumors.

Short Abstract: Pre-clinical models have extensively been used to understand the molecular underpinnings of cancer. Cell lines and Patient Derived Xenografts (PDX) are amenable to screening for a wide range of anti-cancer therapeutics. These screens offer a direct measure of drug response for many drugs – data that cannot be collected for human tumors. Pre-clinical models do, however, show behavioral discrepancies with respect to human tumors that impedes the transfer of biomarkers of drug response from pre-clinical models to patients. We present a novel framework for integrating omics data derived from pre-clinical models and tumors. Our approach employs non-linear dimensionality reduction to capture complex genetic interaction patterns that are common to pre-clinical models and humans. These patterns are then used to train a drug response predictor on human tumor data. This work extends PRECISE to allow incorporation of non-linear similarity measures between samples while retaining equivalence to the linear setting.

Short Abstract: Cancer heterogeneity introduces significant challenges in the design of effective treatment strategies. Since tumours can exhibit different sensitivities to cytotoxic drugs among their distinct clonal populations, targeted treatments and chemotherapies could play a role in the appearance of drug resistance mechanisms. Hence, there is a need for studying clone-specific vulnerabilities in order to properly characterize drug responses in patient-specific data.
In this work, we present beyondcell, a novel R package for the analysis of single-cell RNA-seq data and in silico prescription of anticancer drugs. We have examined a public single-cell RNA-seq dataset of 451Lu naïve and BRAF inhibitor (BRAFi) resistant cells [1]. Moreover, we have obtained Drug-induced Expression (DE) signatures from the LINCS L1000 database [2]. Using these signatures, beyondcell computes a score summarizing the cell-based drug sensitivity. As a result, beyondcell was able to recapitulate the resistance to BRAFi in 451Lu cells, as well as to identify clusters of cells with distinct drug vulnerabilities.
We believe beyondcell can be a valuable resource in personalized medicine and lead to a better understanding of the biological and therapeutic impact of tumour heterogeneity.

Short Abstract: Alzheimer’s disease AD is neurodegenerative disease which is progressive in nature with alterations in function, cognition and behavior. Exact pathways or mechanisms involved in the progression are difficult to understand due to complexity. Huntington’s disease HD is another complex neurodegenerative disease often characterized by neuronal dysfunction in brain regions mainly striatum and cortex. In this study, we have compared expression changes in human prefrontal brain samples in Alzheimer’s disease, Huntington’s disease patients with healthy control samples. Microarray dataset GSE33000 was downloaded from GEO database. The study was aimed at identifying disease specific hub genes and pathways in AD and HD. In case of AD Vs Controls GNG2, NMU, MCHR2, NMS, GNB5, GNG3, ANXA1, GNG12,HTR5A,GABBR2 were identified as top 10 hub genes. HD Vs healthy controls revealed identification of C3AR1, C5AR1, FPR1, CXCL1, CXCL2, GNG2, NMU, ADORA3, MCHR2, NMS as hub genes. Most enriched pathways in AD were Chemokine signaling pathway, Neuroactive ligand-receptor interaction, while in HD G-protein-coupled receptor signaling pathway neuropeptide signaling pathway were identified. In order to validate the results, we have further investigated the variation in each group using expression plots and hierarchical clustering. These identified hub genes could be considered as potential biomarker for further disease analysis.

Short Abstract: Chemical compounds are tested in various assays from which Efficacy Metrics (EM) can be estimated. Compounds are selected with the aim of identifying at least one sufficiently potent and efficient to go into preclinical testing. Selection is based on EM meeting a specific threshold or by comparison to other compounds.
Current analysis methods only suggest estimates of EM and hardly consider the inevitable experimental noise, thus failing to quantify the uncertainty on EM on which conclusions are based. We propose to extend our previously introduced rigorous statistical methods (EM inference) to a panel of compounds. Given an efficacy criteria, we aim at identifying the compounds with the highest probability of meeting that criteria.
We use a hierarchical Bayesian model to infer EM from dose-response assays. Given the empirical values distributions for an EM of interest, our novel ranking method returns the probability that each compounds within a set is able to achieve a given rank. We are able to identify all compounds of a experimental dose-response set with at least 1% chance of being amongst the best for a given EM. To further analysis, we generate DAGs where path between two compounds identifies which is statistically better.

Short Abstract: Cancer cell lines have been used extensively as a proxy to primary tumors to characterize cancer biology and drug responses. Understanding how well cancer cell lines could represent the actual tumor is an important question. Several studies have explored similarity between cell lines and primary tissues relying on a genome-wide correlation-based approach between bulk tumor tissue without considering the heterogeneous cell population in bulk tumors. Here, we propose CTDPathSim, a pathway activity-based approach to compute similarity between primary tumors and cell lines. To address bulk tumor heterogeneity, we utilized a deconvolution method to get cell type-specific expression and methylation profiles. We tested CTDPathSim on breast cancer data in The Cancer Genome Atlas (TCGA) and cancer cell lines data in the Cancer Cell Line Encyclopedia (CCLE) database. Our results showed that highly similar patient-cell line pairs were significantly associated with the responses of several cancer drugs, such as, Paclitaxel, Vinorelbine and Doxorubicin. In highly similar patient-cell line pairs, the presence of breast cancer cell lines was significantly higher than in lowly similar pairs. Also, the highly similar cell lines were found to be clinical biomarkers for patients. CTDPathSim also outperformed a singular value decomposition-based method based on several evaluation criteria.

Short Abstract: Pancreatic ductal adenocarcinoma (PDAC) shows an increasing number of cases every year and its mortality rate remains high. A convenient method for the analysis of the tumor molecular landscape is the analysis of the transcriptome. In our work, we analyzed three transcriptomic datasets related to PDAC. These were TCGA data (183 samples), other published results from a PDAC cohort (96 samples), and a PDAC dataset obtained at DKFZ and Heidelberg University Clinics, Germany (457 samples). The data from bulk-samples were deconvolved using consensus independent component analysis (ICA), which allowed a reproducible separation of transcriptional signals from different cell types and technical factors. We identified similar components in all three datasets. In addition, our results on bulk-sample data were complemented by a similar analysis of single-cell data, improving interpretability of the determined components. We identified biological functions and cell types, which can be positively or negatively linked to survival. We also showed that signal weights obtained in the deconvolution could be used as predictive markers of survival across independent patient cohorts.

Short Abstract: Model organism studies have demonstrated the possibility of lifespan extension up to 10-fold through genetic interventions. Although the effect size is relatively smaller, several drugs have also been shown to modulate lifespan and health during ageing in model organisms. Translation of this information to humans, however, is challenging and requires further investigation. In this study, we perform a comparative and integrative analysis of different drug repurposing studies for human ageing together with the known lifespan modulators in model organisms. We use two different approaches we developed, i) targeting genes which change expression during ageing in humans, and ii) targeting genes associated with an increased risk of multiple late-onset diseases. The first set included a significant number of known lifespan modulators, which also improves health in model organisms. However, drugs targeting multiple diseases did not overlap with the known pro-longevity drugs. This offers new avenues to explore experimentally. Through a systems-level analysis of the targeted pathways and their regulators, we aim to elucidate the mechanisms of lifespan modulation that can also improve health in the elderly.

Short Abstract: The development of sequencing and targeted therapies has led to the emergence of precision medicine (PM): molecular biomarkers guide the choice of treatment, by a physician or based on an algorithm. The growing number of such algorithms raises the question of how to quantify their clinical impact, simulating clinically relevant clinical trials comparing PM strategies to various controls.

We specify a framework to derive causal effects of PM strategies from observational (non-randomized) data. It takes into account the different targeted drugs involved in the treated arm to better analyse the heterogeneity of responses and correct for confounding factors. Then, we can compare the treatment assignment algorithm to control arms such as a single standard treatment or random assignment of the different targeted treatments.

Causal estimates of the PM effects are first evaluated on simulated data and they demonstrate a lower bias compared with naive estimation of the clinical trial outcome based on observed data. Many different cases can be investigated thanks to an interactive application..

The method is then applied to public data from patient-derived xenografts: the availability of different treatment responses for each model makes it possible to verify in a pre-clinical context, the validity of the proposed method.

Short Abstract: Background
Immune thrombotic thrombocytopenic purpura (iTTP) is a potentially fatal syndrome, resulting primarily from autoantibodies against ADAMTS13. However, the mechanism underlying autoantibody formation is unknown. Also unknown is the contribution of any genetic alterations that may contribute to the pathogenesis of iTTP or the acquired deficiency of plasma ADAMT13 activity.

Results
Here we performed whole-exome sequencing (WES) of DNA samples obtained from 40 patients with iTTP and 15 age- and ethnicity-matched control subjects. WES revealed glycosylation, particularly O-linked glycosylation, to be a major pathway affected in patients with iTTP. A mass spectrometry analysis of glycoproteins from normal and iTTP patient samples confirmed that the samples do differ drastically in their glycoproteins profiles, and the difference is largely affected proteins involved in lipid metabolism. We propose that the altered glycosylation may predispose the patients towards the formation of autoantibodies against ADAMTS13 and might also affect antibody binding properties. Moreover, we also identified variants in the ANKRD family, particularly ANKRD36C and its paralogs, associated with iTTP. The function of the ANKRD36 family of proteins remains unknown.

Conclusion
Together, our findings provide novel insight into plausible mechanisms underlying autoantibody production and the potential contribution of genetic abnormalities in the pathogenesis of iTTP.

Short Abstract: In clinical practice, normal values or reference intervals are the main point of reference for interpreting a wide array of measurements, including biochemical laboratory tests, anthropometrical measurements, physiological or physical ability tests. They are historically defined to separate a healthy population from unhealthy and therefore serve a diagnostic purpose. Numerous cross-sectional studies use various classical parametric and nonparametric approaches to calculate reference intervals. Based on a large cross-sectional study (N = 60,799), we compute reference intervals for subpopulations (e.g. males and females) which illustrate that subpopulations may have their own specific and more narrow reference intervals. We further argue that each healthy subject may actually have its own reference interval (subject- specific reference intervals or SSRIs). However, for estimating such SSRIs longitudinal data are required, for which the traditional reference interval estimating methods cannot be used. In this study, a linear quantile mixed model (LQMM) is proposed for estimating SSRIs from longitudinal data. The SSRIs can help clinicians to give a more accurate diagnosis as they provide an interval for each individual patient. We conclude that it is worthwhile to develop a dedicated methodology to bring the idea of subject-specific reference intervals to the preventive healthcare landscape.

Short Abstract: Glioma refers to tumour cells with histomorphological glial cell resemblance. Initially, its classification relied on histological features but following high-throughput molecular profiling emergence, biomarkers became an important part of the glioma grading system. Nowadays, the isocitrate dehydrogenase (IDH) gene is mainly used for this classification in three categories: IDH-mutated astrocytoma bearing a 1p/19q chromosomal codeletion, IDH-mutated astrocytoma without the 1p/19q codeletion and finally IDH-WT astrocytoma, showing the worst prognosis. The IDH enzyme participates in the TCA cycle, metabolizing the isocitrate substrate to alpha-ketoglutarate, which is one of the major substrates for GABA and glutamate neurotransmitter metabolism. In a previous study, we identified members of these pathways associated with a glioma subgroup with better prognosis than usual, supporting the importance to better characterize their involvement in tumor biology when deregulated.

We hypothesize that GABA and glutamate metabolic pathways may play a crucial role in glioma aggressivity behaviour.

In this study, we profiled the GABA and glutamate pathways in IDH-mutated and IDH-wildtype glioma using the TCGA and CGGA databases. Unsupervised clustering identified genes with expression levels correlating with the IDH mutation presence. We also showed that these genes were strongly associated with survival, demonstrating their potential use as prognosis biomarkers.

Short Abstract: Colorectal cancer is a virulent disease with diverse outcomes in cancer progression and survival for patients that receive the standard of care. Recent studies have demonstrated that African Americans are susceptible to an aggressive, early-onset subtype of colorectal cancer with higher mortality rates. At the molecular level, this subtype is characterized by the absence of mutations in the most common driver of colorectal cancer, the gene Adenomatous Polyposis Coli (APC). An understanding of the mechanisms driving APC negative tumors would help treat the most challenging colorectal cancer patients as well as reduce health disparities. To better characterize APC negative colorectal tumors, we identified 90 APC positive primary tumors and 90 APC negative primary tumors matched for age and sex from The Cancer Genome Atlas (TCGA). Gene Set Enrichment Analysis (GSEA) showed that APC negative tumors have higher immune signaling and a greater dependency on the electron transport chain. We then corrected for immune infiltration and performed a comparative analysis on patient-specific transcriptional regulatory networks. We validated changes in network structure with independent data from DNA methylation analysis. These findings highlight the crucial role of network analysis in disease stratification and identify APC mutation negative regulatory mechanisms to inform therapeutic options.

Short Abstract: The distribution and abundance of immune cells, particularly T-cell subsets,
play pivotal roles in cancer immunology and therapy. T cells have many
subsets with specific function and current methods are limited in estimating
them, thus, a method for predicting comprehensive T-cell subsets is
urgently needed in cancer immunology research. Here, Immune Cell
Abundance Identifier (ImmuCellAI), a gene set signature-based method, is
introduced for precisely estimating the abundance of 24 immune cell
types including 18 T-cell subsets, from gene expression data. Performance
evaluation on both the sequencing data with flow cytometry results
and public expression data indicate that ImmuCellAI can estimate the
abundance of immune cells with superior accuracy to other methods
especially on many T-cell subsets. Application of ImmuCellAI to
immunotherapy datasets reveals that the abundance of dendritic cells,
cytotoxic T, and gamma delta T cells is significantly higher both in
comparisons of on-treatment versus pre-treatment and responders versus
non-responders. Meanwhile, an ImmuCellAI result-based model is built for
predicting the immunotherapy response with high accuracy (area under
curve 0.80–0.91). These results demonstrate the powerful and unique
function of ImmuCellAI in tumor immune infiltration estimation and
immunotherapy response prediction.

Short Abstract: The goal of tumor genomic profiling is to identify somatic mutations. However, most clinical-grade implementations lack patient-matched germline DNA, and additional analyses are needed to infer variants’ mutational status (somatic or germline) based on accurate estimates of specimens’ tumor content. We present an integrated bioinformatics tool to estimate specimen purity based on allele frequencies and focal copy-numbers of detected variants in individual tumor samples. Implemented information-theoretic algorithms permit selecting the most consistent somatic-versus-germline model and infer loss of heterozygosity and cancer cell fraction while accounting for biases inherent to clinical sequencing and sample purity estimation. Using simulations and two large independent clinical sequencing datasets comprising 3,492 solid tumors, we demonstrate the accuracy and precision of our pipeline, especially to identify likely germline mutations. We discuss cases for which these analyses provide a model for tumor heterogeneity and its evolution and present meaningful information regarding possible treatment options. Our user-friendly, interactive presentation of inference results has the potential to simplify the access and ultimately increase the adoption of computational methods for interpreting clinical sequencing data, improving the speed with which patients can be assessed for potential therapies and contributing to improved outcomes.

Short Abstract: The influence of seasons on biological processes, particularly at a molecular level, is poorly understood. Moreover, seasons are arbitrarily defined based on four equal segments in the calendar year. In order to identify biological seasonal patterns in humans based on diverse molecular data, rather than calendar dates, we leveraged the power of longitudinal multi-omics data from deep profiling cohort of 105 individuals. These individuals underwent intensive clinical measures and emerging omics profiling technologies including transcriptome, proteome, metabolome, cytokinome as well as gut and nasal microbiome monitoring for up to four years. We identified more than 1000 seasonal variations in omics analytes and clinical measures, including molecular and microbial markers with known seasonality changes, as well as new molecular and microbial markers with seasonality fluctuations. The different molecules grouped into two major seasonal patterns which correlate with peaks in late spring and late fall/early winter in the San Francisco Bay Area. Lastly, we used our recently developed omcis longitudinal differential analysis method, OmicsLonDA, to identify molecules and microbes that demonstrated different seasonal patterns in insulin-sensitive and insulin-resistant individuals. These gained insights have important implications for human health and our methodology framework can be applied to any geographical location.

Short Abstract: The T cell-inflamed tumor microenvironment, characterized by CD8 T cells and type I/II interferon transcripts, is an important cancer immunotherapy biomarker. Tumor mutational profile may also dictate response with oncogenes known to mediate immune exclusion. We performed an integrated multi-omic analysis of human cancer in multiple databases and clinical trial data, correlating the T cell-inflamed gene expression signature with somatic mutations, transcriptional programs and proteome for immune exclusion phenotype. Strong correlations were noted between mutations in oncogenes and tumor suppressor genes and non-T cell-inflamed tumors with examples including IDH1 as well as less well-known genes including KDM6A. Analyzing gene expression patterns, we identify oncogenic mediators of immune exclusion broadly active across cancer types including HIF1A and MYC with validation in independent external cohorts. Novel examples from specific tumors include sonic hedgehog signaling and hormone signaling. An increasing number of co-activated pathways is associated with lower T cell-inflamed gene expression. Validating these analyses, we observe highly consistent inverse relationships between pathway protein levels and the T cell-inflamed gene expression across cancers. These results nominate molecular targets potentially available for immediate translation into clinical trials for patients with cancer. Rational targeted-immunotherapy combinations may be prioritized to overcome the non-T cell-inflamed tumor microenvironment.

Short Abstract: Pancreatic ductal adenocarcinoma (PDAC) is categorized as the seventh leading cause of cancer mortality in the world. It has a 5-year survival rate of less than 5%. It remains a highly challenging task to identify those factors that provide more insight into long-term (LT) survival. Understanding survival heterogeneity requires integrative analyses that considers multi-faceted data sources, such as mutation information from WES (whole exome sequencing), transcriptome data from RNA-Seq experiments as well as clinical-pathological information. Using a discovery cohort of 19 PDAC patients from CHU-Liège (Belgium), we first identified differentially expressed genes (DEGs) between LT/ST survivors. Then, we identified somatic variants, consistent with both VarScan2 and MuTect2, and performed a detailed mutational signature analysis via trinucleotide frequencies in LT and short-term (ST) survivors. We also looked at chromosomal rearrangement LT/ST differences (insertions/deletions) in 30 cancer exomes across PDAC cancer types from TCGA. We finally integrated individual-level profiles from WES to group level profiles from RNA-Seq by using graph diffusion on a protein-interaction network. Based on our results multiple genes of potential interest to LT/ST survival could be prioritized, including CFAP74, MUC6 and MUC3A.

Short Abstract: Identification of novel genetic causes for rare or genetically heterogeneous diseases relies on fast and straightforward accessibility of related OMICs studies. Although there is already a multitude of disease-specific genomic web databases, many diseases are still not or not sufficiently represented. In order to support researchers in the domains of medicine and biology to implement disease-specific web applications in their fields, the open source template myGeneAtlas is currently under development. It is based on a recently published web application for the field of male infertility, the Male Fertility Gene Atlas [1]. The template, provides all main functionalities of gene atlas web applications: Registered users can upload data of relevant OMICs studies, i.e. meta information, annotations regarding phenotypes, processed cells and tissues as well as images and tables. Data, most importantly images and tables, can then easily be viewed on the website and automatically be searched for mentioning of gene names and other annotations. MyGeneAtlas is java-based and employs the frameworks Hibernate and Spring Boot on the server side and Bootstrap on the client side for a modern and responsive web design. Wherever possible, appropriate ontologies are integrated.
[1] Krenz et al. MedRxiv (2020). doi.org/10.1101/2020.02.10.20021790.

Short Abstract: We have recently presented network meta-analysis as an alternative to the analysis of merged transcriptome profiles from multiple independent experiments (Winter et al., BMC Bioinf 2019, 20(1): 144). While data merging in transcriptomics becomes difficult when the data has been taken by different technologies (e.g. microarrays and RNA-seq) or when batch effects cannot be properly removed, network meta-analysis uses the results from each individual study. We demonstrate also the usability of network meta-analysis for indirect group comparisons that have not been performed in the individual studies.
We employ the functionality implemented in the R-package `netmeta’ (Rücker, Res Synth Meth 2012, 3(4): 312-224) and apply it to data retrieved from Gene Expression Omnibus and ArrayExpress. In addition, we study the correlation between the analysis results from merged gene expression data and the results of network meta-analysis within a simulation study.
The simulation study as well as the real world data example from infection research show that the results of network meta-analysis are highly correlated to the results of merged data sets. Furthermore, network meta-analysis seems to be superior to data merging when the number of studies becomes large.

Short Abstract: Screening drugs against gene mutants can identify mutations that sensitize/suppress a drug’s effect. These chemical-genetic interaction (CGI) screens can be performed in human cell lines using a pooled lentiviral CRISPR-Cas9 approach to assess drug sensitivity/resistance of single-gene knockouts across the human genome. While these screens can inform candidates for drug development, many labs lack the resources to scale these screens across many drugs. Our pilot screens with Bortezomib (proteasome inhibitor) show that a compressed CRISPR library can 1) recover biological information at higher signal:noise ratio compared to genome-wide screens, and 2) reduce costs to allow for higher-throughput drug screening.

Phenotypic readout from these screens can be translated to CGI profiles. These profiles are analogous to genetic interaction (GI) profiles, which represent sensitivity/resistance of genetic perturbations to a second genetic perturbation rather than a drug. To predict a drug’s genetic target(s) and modes-of-action, we are leveraging the property that a drug’s CGI profile will be similar to its target’s GI profile. We are developing novel computational approaches to integrate existing CGI and GI data to support this drug target prediction approach. The combination of this targeted screening approach and novel computational framework will provide a more scalable platform for drug discovery.

Short Abstract: In this work, we applied a data-driven deconvolution method called consensus independent component analysis (ICA) to resolve the complexity of bulk and single cell data. The method was able to separate technical biases from signals of biological interest, isolate signals from different biological processes and cell subpopulations in different components, and integrate transcriptomics data with patient clinical data. The method was tested on TCGA and several independent datasets. Two single-cell RNA-seq datasets composed of 2544 normal pancreas cells and 3304 cancer cells from two patient-derived glioblastoma cell-lines were considered. Matching deconvolution results obtained on bulk cancer samples and single-cell data, we found two the most correlated components linked to the cell cycle and two others represented overall gene expression. ICA allows isolating signal from specific cell type, as acinar, mesenchymal, pancreatic ductal and pancreatic A cells, in components and match it with components of bulk sample data. By mapping of the components between bulk and single cell datasets, we improved the annotation of the components and therefore established a link between observation at cellular level and clinical data. Furthermore, new annotation can be used in interpretation of predicting models for patient classification.

Short Abstract: Precision oncology necessitates detailed characterization of aberrant pathways of individual tumors. This process requires an accurate cancer diagnosis in order to align the observed changes against the appropriate background and comparators. Cancer classifiers leveraging transcriptome and genomic data can optimize the task of distinguishing various cancers, but understanding the biological underpinning of the diagnosed cancer still remains a complex and laborious manual task. Pathway Impact Evaluation (PIE) uses classifiers trained with high-dimensional gene expression data from 40 cancer types to provide single-sample biological pathway importance scores. The resultant pathway profiles recapitulate cancer-specific biology of tumors in The Cancer Genome Atlas, enable single-sample analysis of treatment-resistant cancers to help explain diagnosis and subtyping, and identify biochemical pathways associated with drug response and disease progression. PIE is available with a pan-cancer classifier as a python package (‘cancerscope’). Pre-calculated sample-level PIE

Short Abstract: Data-driven, machine learning (ML) decision support systems have the potential to revolutionize cancer care. Patient clinical data, which are routinely collected and stored in electronic health record (EHR) systems, can provide clinical evidence to drive ML algorithms to learn the nonlinear mappings between biomarkers, disease features, treatment options, clinical outcomes and toxicity. In this respect, a framework for ML-based learning decision support system (LDSS) for cancer care is explored. The framework enables adaptive treatment decision-making based on estimated patient health state, treatment outcomes and toxicity. Finite state machines (FSMs), with states tied to the RECIST guideline, are used to estimate patient health state. DNNs (Deep Neural Networks) trained on EHR and tumor sequencing data provide predictions of outcomes and toxicity. Sequential decision-making algorithms are used to synthesize treatment recommendations, guided by a reward function that integrates predictions about patient health state and treatment outcomes and toxicity. The proposed LDSS is a step towards realizing the vision of rapid learning systems for cancer care. However, such vision will require interoperability standards for EHR systems and data governance frameworks that address patient consent, regulatory procedures and ethical considerations, to leverage the value of data being amassed in hospitals towards cancer care.

Short Abstract: Chronic Obstructive Pulmonary Disease (COPD) is the fourth leading cause of death worldwide and has been broadly studied given its medical relevance. However, consensus regarding the pathology and etiology of COPD is still lacking. Using PulmonDB, a gene expression compendium of processed transcriptomic data along with metadata from publicly available lung disease datasets, we aimed to identify expression profiles of COPD. Taking all available COPD samples from lung tissue, we looked for genes commonly deregulated across study, disease's status, age, gender and smoking habits, to identify shared pathways across heterogeneous COPD samples.
Differentially expressed genes were identified independently within studies at an adjusted p-value threshold of 0.1. Few genes were identified in at least 3 of the 8 studies included in the analysis, with some even showing discordant directions across studies.

Short Abstract: Precision-based therapeutics for cancer patients have resulted in reducing tumor progression while alleviating harmful side effects. Unfortunately, tumors with the same oncogenic mutation, but derived within different tissues, can have highly variable responses to the same precision-based therapeutic. For example, melanoma patients that contain a V600E BRAF mutation often exhibit sensitivity towards BRAF inhibitors, while colorectal tumors with the same BRAF mutation are less likely to respond to a BRAF inhibitor. Machine learning models that are generated using genomic information have had the greatest success in determining if a specific tumor will be sensitive or resistant to an anti-cancer drug. Despite the predictive capacity of machine learning models, they provide little insight into the cancer specific mechanisms that drive drug response. To identify how a machine learning model utilizes information across pan-cancer data to predict the drug response of individual cancer types, we created the method BISON (Bayesian Inferred Shared Oncogenic Networks). By applying BISON to the Genomic of Drug Sensitivity of Cancer (GDSC) database, we were able to increase the prediction of 27 percent of cancer specific models, relative to lasso and TANDEM regression, and improve the identification of cancer specific biomarkers for individual anti-cancer drugs.