Drug Sensitivity Prediction Research Articles

High-Content Analysis-Based Sensitivity Prediction and Novel Therapeutics Screening for c-Met-Addicted Glioblastoma.

Simple SummaryReal-time ex vivo drug testing tailors individual therapeutics based on predicted drug responses. Most technologies to date rely on conventional drug screening that provides low confidence data. Here, we present high-content analysis-based drug testing of glioblastoma patients to identify the right glioblastoma patients for a given drug. This generates multi-parameter biomarker and phenotype readouts providing a better reliability of the assay. Additionally, we showed a high-content drug repurposing screen and defined a new c-Met-inhibiting function of the CDK4/6 inhibitor Abemaciclib. Large-scale high throughput screening results demonstrate that Abemaciclib sensitivity in glioblastoma patients is highly correlated with the c-Met inhibitors sensitivity, further supporting the accuracy of the platform and important new clinical implications regarding multiple functions of Abemaciclib.(1) Background: Recent advances in precision oncology research rely on indicating specific genetic alterations associated with treatment sensitivity. Developing ex vivo systems to identify cancer patients who will respond to a specific drug remains important. (2) Methods: cells from 12 patients with glioblastoma were isolated, cultured, and subjected to high-content screening. Multi-parameter analyses assessed the c-Met level, cell viability, apoptosis, cell motility, and migration. A drug repurposing screen and large-scale drug sensitivity screening data across 59 cancer cell lines and patient-derived cells were obtained from 125 glioblastoma samples. (3) Results: High-content analysis of patient-derived cells provided robust and accurate drug responses to c-Met-targeted agents. Only the cells of one glioblastoma patient (PDC6) showed elevated c-Met level and high susceptibility to the c-Met inhibitors. Multi-parameter image analysis also reflected a decreased c-Met expression and reduced cell growth and motility by a c-Met-targeting antibody. In addition, a drug repurposing screen identified Abemaciclib as a distinct CDK4/6 inhibitor with a potent c-Met-inhibitory function. Consistent with this, we present large-scale drug sensitivity screening data showing that the Abemaciclib response correlates with the response to c-Met inhibitors. (4) Conclusions: Our study provides a new insight into high-content screening platforms supporting drug sensitivity prediction and novel therapeutics screening.

Predicting chemosensitivity using drug perturbed gene dynamics

BackgroundOne of the current directions of precision medicine is the use of computational methods to aid in the diagnosis, prognosis, and treatment of disease based on data driven approaches. For instance, in oncology, there has been a particular focus on development of algorithms and biomarkers that can be used for pre-clinical and clinical applications. In particular large-scale omics-based models to predict drug sensitivity in in vitro cancer cell line panels have been used to explore the utility and aid in the development of these models as clinical tools. Additionally, a number of web-based interfaces have been constructed for researchers to explore the potential of drug perturbed gene expression as biomarkers including the NCI Transcriptional Pharmacodynamic Workbench. In this paper we explore the influence of drug perturbed gene dynamics of the NCI Transcriptional Pharmacodynamics Workbench in computational models to predict in vitro drug sensitivity for 15 drugs on the NCI60 cell line panel.ResultsThis work presents three main findings. First, our models show that gene expression profiles that capture changes in gene expression after 24 h of exposure to a high concentration of drug generates the most accurate predictive models compared to the expression profiles under different dosing conditions. Second, signatures of 100 genes are developed for different gene expression profiles; furthermore, when the gene signatures are applied across gene expression profiles model performance is substantially decreased when gene signatures developed using changes in gene expression are applied to non-drugged gene expression. Lastly, we show that the gene interaction networks developed on these signatures show different network topologies and can be used to inform selection of cancer relevant genes.ConclusionOur models suggest that perturbed gene signatures are predictive of drug response, but cannot be applied to predict drug response using unperturbed gene expression. Furthermore, additional drug perturbed gene expression measurements in in vitro cell lines could generate more predictive models; but, more importantly be used in conjunction with computational methods to discover important drug disease relationships.

Multiplex Gene Quantification as Digital Cancer Markers for Extremely Rapid Evaluation of Chemodrug Sensitivity

In clinical cancer medicine, many patients require immediate chemotherapy after hospitalization. Current administrations for precision drug uses are limited in evaluation speed, including genomic sequencing and tumor organoid evaluation. An extremely rapid evaluation protocol is in high demand to realize drug recommendation within a few hours after tumor sampling. Herein, we propose multiplex gene-based digital cancer markers for personalized drug sensitivity prediction. We first screen cancer transcriptional profiles by applying 2 - 3 gene filters, scoring the genes by their impact on drug sensitivity, and finalise the gene lists by K-nearest-neighbours (KNN) enumeration. The digital markers are cancer type-dependent and composed by assembling predictive genes for individual drugs in clinical uses. The digital markers, comprised of the digital readings of 10s–100s of gene expressions, are rapidly quantified by reverse transcription–quantitative real-time PCR (RT-qPCR) within 1 - 3 hours after cancer sampling. The area under the receiver operating characteristic (ROC) curve (AUC) reached 0.88 when testing the performance of digital markers in patient tumors. An algorithm and its graphic interface were developed to demonstrate the promise of digital cancer markers for extremely rapid drug recommendation.

Network-based drug sensitivity prediction

BackgroundDrug sensitivity prediction and drug responsive biomarker selection on high-throughput genomic data is a critical step in drug discovery. Many computational methods have been developed to serve this purpose including several deep neural network models. However, the modular relations among genomic features have been largely ignored in these methods. To overcome this limitation, the role of the gene co-expression network on drug sensitivity prediction is investigated in this study.MethodsIn this paper, we first introduce a network-based method to identify representative features for drug response prediction by using the gene co-expression network. Then, two graph-based neural network models are proposed and both models integrate gene network information directly into neural network for outcome prediction. Next, we present a large-scale comparative study among the proposed network-based methods, canonical prediction algorithms (i.e., Elastic Net, Random Forest, Partial Least Squares Regression, and Support Vector Regression), and deep neural network models for drug sensitivity prediction. All the source code and processed datasets in this study are available at https://github.com/compbiolabucf/drug-sensitivity-prediction.ResultsIn the comparison of different feature selection methods and prediction methods on a non-small cell lung cancer (NSCLC) cell line RNA-seq gene expression dataset with 50 different drug treatments, we found that (1) the network-based feature selection method improves the prediction performance compared to Pearson correlation coefficients; (2) Random Forest outperforms all the other canonical prediction algorithms and deep neural network models; (3) the proposed graph-based neural network models show better prediction performance compared to deep neural network model; (4) the prediction performance is drug dependent and it may relate to the drug’s mechanism of action.ConclusionsNetwork-based feature selection method and prediction models improve the performance of the drug response prediction. The relations between the genomic features are more robust and stable compared to the correlation between each individual genomic feature and the drug response in high dimension and low sample size genomic datasets.

Two-step multi-omics modelling of drug sensitivity in cancer cell lines to identify driving mechanisms.

Drug sensitivity prediction models for human cancer cell lines constitute important tools in identifying potential computational biomarkers for responsiveness in a pre-clinical setting. Integrating information derived from a range of heterogeneous data is crucial, but remains non-trivial, as differences in data structures may hinder fitting algorithms from assigning adequate weights to complementary information that is contained in distinct omics data. In order to counteract this effect that tends to lead to just one data type dominating supposedly multi-omics models, we developed a novel tool that enables users to train single-omics models separately in a first step and to integrate them into a multi-omics model in a second step. Extensive ablation studies are performed in order to facilitate an in-depth evaluation of the respective contributions of singular data types and of combinations thereof, effectively identifying redundancies and interdependencies between them. Moreover, the integration of the single-omics models is realized by a range of distinct classification algorithms, thus allowing for a performance comparison. Sets of molecular events and tissue types found to be related to significant shifts in drug sensitivity are returned to facilitate a comprehensive and straightforward analysis of potential computational biomarkers for drug responsiveness. Our two-step approach yields sets of actual multi-omics pan-cancer classification models that are highly predictive for a majority of drugs in the GDSC data base. In the context of targeted drugs with particular modes of action, its predictive performances compare favourably to those of classification models that incorporate multi-omics data in a simple one-step approach. Additionally, case studies demonstrate that it succeeds both in correctly identifying known key biomarkers for sensitivity towards specific drug compounds as well as in providing sets of potential candidates for additional computational biomarkers.

Unbiased in vivo preclinical evaluation of anticancer drugs identifies effective therapy for the treatment of pancreatic adenocarcinoma

Pancreatic ductal adenocarcinoma (PDAC) is typically diagnosed at an advanced stage, which limits surgical options and portends a dismal prognosis. Current oncologic PDAC therapies confer marginal benefit and, thus, a significant unmet clinical need exists for new therapeutic strategies. To identify effective PDAC therapies, we leveraged a syngeneic orthotopic PDAC transplant mouse model to perform a large-scale, in vivo screen of 16 single-agent and 41 two-drug targeted therapy combinations in mice. Among 57 drug conditions screened, combined inhibition of heat shock protein (Hsp)-90 and MEK was found to produce robust suppression of tumor growth, leading to an 80% increase in the survival of PDAC-bearing mice with no significant toxicity. Mechanistically, we observed that single-agent MEK inhibition led to compensatory activation of resistance pathways, including components of the PI3K/AKT/mTOR signaling axis, which was overcome with the addition of HSP90 inhibition. The combination of HSP90(i) + MEK(i) was also active in vitro in established human PDAC cell lines and in vivo in patient-derived organoid PDAC transplant models. These findings encourage the clinical development of HSP90(i) + MEK(i) combination therapy and highlight the power of clinically relevant in vivo model systems for identifying cancer therapies.

TMOD-07. HUMAN DIFFUSE MIDLINE GLIOMA AVATARS AS A PLATFORM TO SEARCH FOR NOVEL THERAPEUTIC TARGETS

Abstract Diffuse midline glioma is the leading cause of brain tumor death among the pediatric population. Drugs that show notable promise in preclinical models inevitably fail to demonstrate efficacy in clinical trials, likely due to the inadequacy of preclinical models. We have recently proposed glioblastoma models derived from human induced pluripotent stem cells (hiPSCs) genetically engineered with different combinations of glioblastoma-associated genetic alterations as a platform to search for therapeutic targets. These glioblastoma avatars authentically recapitulated the different pathobiology of glioblastoma subtypes, depending on what genetic alterations to be introduced. To investigate the biology and to develop novel therapeutics for diffuse midline glioma with H3K27M mutation, we have established a novel model by introducing H3.3 K27M mutation together with one of the most common concurrent genetic alterations, TP53 R248Q mutation, into hiPSCs through CRISPR/Cas9 genome engineering. Orthotopic engraftment of the neural progenitor cells derived from these edited hiPSCs formed diffusely invasive brainstem tumors with histological features of the diffuse midline glioma. These tumor avatars presented a global reduction in H3K27me3 accompanied by the expression of H3K27M. Transcriptome analyses of these models revealed that these avatars with H3K27M cluster apart from the pediatric glioma samples without this particular mutation, and that they present signatures of oligodendroglial progenitor differentiation as discovered in patient samples with this mutation. Using these models faithfully recapitulating histology and pathobiology of the patient tumors, we have performed drug screening and confirmed that their sensitivity to known drugs, including an EZH2 inhibitor and histone deacetylase inhibitors. On these faithful human avatars of diffuse midline glioma with H3K27M, we have applied bioinformatics algorithms of drug sensitivity prediction aiming at developing novel therapeutics for this devastating pediatric glioma.

Co-Occurring Mutation Clusters Predict Drug Sensitivity in Acute Myeloid Leukemia

Background The molecular origin of cancer drug resistance is not apparent for most cases of acute myeloid leukemia (AML). Clonal evolution appears to be associated with increasing drug resistance. We sought to determine whether the mutation patterns are associated with drug susceptibility in AML. Methods Seventy-two patient blood or marrow samples were enriched for CD34+ blasts by immunomagnetic bead selection. High throughput drug sensitivity screens were performed for 223 drugs after a 72-hour exposure to 8-12 customized drug concentrations (within the range of 5pM to 100µM) of each drug spanning 4-5 logs. Post exposure viability was determined using CellTiter Glo luminescent reagent. XLFit was used to analyze the data and generate dose response curves based on a standard 4-parameter logistic fit. Mutation analysis was performed by MyAML™ utilizing next generation sequencing (NGS) to analyze the 3' and 5' UTRs and exonic regions of 194 AML-associated genes and genomic breakpoints. Co-mutation and mutual exclusivity scores of gene aberrations were computed using three different statistical methods on three large publicly available datasets: 1) TCGA-AML, 2) BeatAML, and 3) data from 1540 patients (Moritz Gerstung, et al., Nat Genet. 2017). The scores for each co-mutation or mutually exclusive gene pair were then logarithmically transformed from the aggregated p-value using a robust rank aggregation method and used to construct a graph from which communities of co-mutated genes could be detected, resulting in higher mutual exclusivity between modules and higher co-occurrence within modules. Results We identified five main groups of co-mutations, including the following: 1) RUNX1 group, 2) CEBPA group, 3) NPM1 group, 4) TP53 group, and 5) RAS group (see Fig 1A). The co-mutation community in the RUNX1 group is featured with transcriptional dysregulation (RUNX1, ASXL1 and EZH2), and dysregulation in splicing (U2AF1, SRSF2 or SF3B1). The TP53 and CEBPA groups exhibit transcriptional dysregulation, and transcription factor alterations. The NPM1 and RAS groups exhibit signaling alteration, particularly the Ras/MAPK signaling pathway. Among each co-mutation community, the variant allele frequencies (VAFs) exhibited differences for specific mutations. For example, in the NPM1 group, FLT3 exhibits the lowest VAF, followed by NPM1 while DNMT3A shows the highest VAF. The five co-mutation groups are associated with different overall survival, including better survival for the CEBPA group and poorer survival for the TP53 group. Within each co-mutation group, specific associations with overall survival were detected, including better survival for NPM1-RAD21 co-mutated vs. wild type (WT), better survival for WT vs. ASXL1-RUNX1 co-mutated, worse survival for DNMT3A-FLT3 co-mutated vs. WT, and others. We also found that the co-mutation clusters were different between the de novo and relapse groups. Most importantly, we found significant correlations with drug sensitivity for the different co-occurring mutation groups. Cells from patients with mutations in the NPM1 co-mutation group show higher sensitivity to most of the drugs, including the tyrosine kinases inhibitors and PI3K-AKT-MTOR inhibitors. Cells from patients with mutations in the RAS cluster exhibited sensitivity to MEK inhibitors (Fig 1B), while cells from patients with mutations in the TP53 cluster show resistance to many drugs, including the MDM2 inhibitor AMG232 (Fig 1C). By mapping the mutated genes and drug targets into signaling pathways, we found that mutations in downstream signaling of the drug target exhibited resistance, while mutations in upstream signaling conferred higher sensitivity. We also constructed drug sensitivity prediction models based on the co-mutation groups. Conclusion We detected co-mutation groups through integrative analysis of large publicly available AML data sets. Patients with mutations in different groups of genes show different overall survival, while cells from patients with mutations in different co-mutation groups show different drug sensitivity. The co-mutation groups may not only reflect the clonal evolution history of AML, but also serve as important features for drug sensitivity prediction. Disclosures Carson: Invivoscribe, Inc: Current Employment. Patay:Invivoscribe, Inc: Current Employment. Becker:Cardiff Oncology: Research Funding; Novartis: Research Funding; SecuraBio: Research Funding; JW Pharmaceutical: Research Funding; Glycomimetics: Research Funding; Abbvie: Research Funding; Bristol Myers Squibb: Research Funding; Pfizer: Research Funding; Invivoscribe: Research Funding; Accordant Health Services/Caremark: Membership on an entity's Board of Directors or advisory committees. OffLabel Disclosure: There are ~200 drugs on the screening panel, and many would be off label for acute myeloid leukemia.

Machine Leaning Algorithm on Chemotherapeutic Drug Resistance Related Gene Classifier in Acute Myeloid Leukemia

Background: Acute myeloid leukemia (AML) is a heterogeneous disease in which 20-50% patients are resistant to chemotherapy. Relapsed/refractory (R/R) AML has a poor long-term prognosis resulting from chemotherapeutic drug resistance. However, the specific gene-drug pairs have been rarely reported. The purpose of our study was to explore an accurate algorithm for screening gene profiles classifier associated with chemotherapy drug sensitivity. Method: 43 AML patients from the Institute of Hematology and Blood Diseases Hospital were enrolled, including newly diagnosis (n=23) and R/R (n=20) AML between March 2019 to January 2020. We separated bone marrow cells (BMCs) of the 43 patients. Additionally, 24 therapeutic drugs panel (Figure 1) were selected and 7-concentration gradient were used to test drug sensitivity in vitro by HDS (High-throuphput Drug Sensitivity Analysis Strategy). Paired molecular data of RNA-seq and DNA methylation 850K CHIP implemented simultaneously. K-Means, an unsupervised algorithm, was applied to cluster all samples into two groups (Resistance/Sensitive group) according to IC50 and 100% PPC inhibition rate of each drug. We used a combination of different algorithms to build a proper gene classifier. The whole process was shown as Figure 2. The major objectives of the evaluating algorithm included F1 score, precision rate, recall rate and AUC. "pRRophetic", a R package, was modified to specify target type as "AML" in aim to estimate IC50 of samples based on GDSC database. Furthermore, we cultured Ara-C resistant cell line of HL-60 in purpose of comparing relative gene expression with wide type (wt) by quantitative Real-time PCR (qPCR) to prove the selected gene signature. Result: The HDS data suggested that sensitivities of newly diagnosis AML to chemotherapeutic-drugs were highly variable. However, Newly Diagnosis Group was more sensitive to majority chemotherapeutic drugs in contrast to R/R AML (P<0.05). As heterogeneous drug response in vitro, we found K-means (k=2) was more reasonable algorithm, as compared to K-means (k=3) for grouping all samples. SVM of transcription data for each drug showed a significantly advantage over other algorithms (P<0.001) with the median F1 score was 0.805295 (0.710648-0.934641), median AUC was 0.818 (0.545-1), median precision rate was 0.75 (0.55-1), and median recall rate was 0.76(0.66-1). The screening feature genes from SVM also performed well on different models (SVM, RF, logistics regression, KNN, Decision Tree) of test set. The SVM mean F1 scores for each algotithm were as follows:0.7382, 0.6128, 0.6530, 0.6168,and 0.5956. It is worth mentioning that SVM equally applies to DMP analysis (P<0.001).We confirmed our algorithm by using Ara-C selected genes grounded by SVM and RF. The difference of estimated IC 50 between resistant group and sensitive group based on SVM (P=0.058) was better than RF (p=0.087) (Figure 3). Compared to newly diagnosis AML, R/R Group possessed 2134 upregulated genes and 1210 down-regulated genes in the aspect of RNA-seq data. Crosslinking analysis of RNA-seq and methylation data of the two group, 1 gene were described as up-regulated expression with hypo-methylation, and 39 genes was down-regulated expression with hyper-methylation. 8/39 genes are corresponded with SVM algorithm. Moreover, 4-gene-drug pairs, including FOXC1 for anthracyclines drug, IGFBP5 for Ara-C, VTRNA1-1 and TKTL1 for pan-drug, were investigated by overlapping TOP 5 genes for each drug of SVM algorithm and DEGs between R/R and newly diagnosis AML. High expression of four genes was identified as a risk factor for AML prognosis. The result of qPCR revealed that IGFBP5 is overexpression in Ara-C-resistant HL-60 than wt. Conclusion: We construct a model of K-Means and RFE-DEG/DMP-SVM, a validated and precise computational approach, for predicting drug sensitivity related genes in AML patients. Up-regulated expression with hypo-methylation genes may be signature genes for drug resistance. 4-gene-based classifier may make contributions to chemotherapeutic-drug resistance prediction and AML treatment decision-making. Disclosures No relevant conflicts of interest to declare.

Clonal Architecture and Variant Allele Frequency Predict Clinical Outcomes and Drug Response in Acute Myeloid Leukemia

Introduction: AML is an aggressive cancer that develops from the sequential accumulation and clonal expansion of somatic mutations in hematopoietic stem and progenitor cells. Recent next-generation sequencing (NGS) studies of AML have correlated mutations with clinical outcomes and response to targeted therapies. Additionally, emerging reports have suggested that increased clonal heterogeneity and mutation burden tend to correlate with worse survival outcomes. However, due to previous cohort sizes, the architecture of clonal evolution and variant allele frequency (VAF) of recurrent mutations have yet to be robustly correlated with response to therapy or with more granular risk stratification. To address previous limitations, we combined available datasets of sequenced AML to model features of clonality and determine their correlations with clinical outcomes and drug sensitivity. Methods: A systematic literature review was performed to identify cohorts of clinically annotated and genetically profiled adult AML. Studies were included if: (i) their sequencing panel targeted at least 30 of the most commonly mutated genes, (ii) censored overall survival data was reported, and (iii) data were publicly available. An additional cohort of patients profiled at Stanford was also included. Leveraging statistical learning methods and robust clonal modeling algorithms (PyClone and ClonEvol), we performed a meta-analysis of the clonal architecture of mutations, their temporal relationships, sensitivity to drugs, and correlation with outcomes in AML. Results: A total of 12 studies were aggregated into a uniformly annotated database comprising 2,987 AML patient samples profiled with an array of DNA sequencing modalities (2,884 with VAFs) and ex vivo drug screening results (nsamples = 562; ndrugs = 122); survival outcomes were available for 2,606 patients. To investigate broad features of leukemia evolution, we used clonal modeling algorithms to infer clonal architecture. Interestingly, patients exhibiting linear evolution (sequential mutations in the same clone) displayed worse outcomes compared to those with branched architecture (distinct subclonal populations). Additionally, mutational burden and clonal heterogeneity only stratified patients with branched structure. These results motivated us to understand how the temporal acquisition of mutations might further stratify outcomes. Using dynamic VAF thresholds, we identified novel high-risk patient populations for 15 recurrently mutated genes. Greater VAF was associated with statistically significant improved survival in genotypes such as GATA2mut and WT1mut and with worse outcomes for patients with NRAS and NF1 mutations. Next, we leveraged VAFs to infer the temporal ordering of individual mutations and functional mutation categories. Patients where NRAS mutations occurred before GATA2 mutations showed a significant correlation with worse outcomes. We also observed that patients in which (i) DNA methylation mutations occurred before those in tumor suppressors and (ii) splicing factor mutations occurred before RTK/RAS signaling components showed significantly shortened overall survival. These results indicate that patients with the same genotype can be stratified by the timing of mutations in the clonal evolution of their leukemia. Finally, we used linear regression between drug sensitivity and VAF to identify several mutations which predict drug sensitivity exclusively in a VAF-dependent manner. Increased WT1 VAF correlated with sensitivity to ABT-737 and elevated FLT3-TKD VAF predicted sensitivity to cabozantinib, among other clinically notable drug-gene relationships. These results suggest potential biomarkers for clinical response to emerging targeted agents. Conclusions: We show that VAF can identify novel high-risk patient populations at the individual mutation level (e.g. BCOR and NF1) and can also be leveraged to stratify outcomes based on inferring the temporal ordering of mutations (e.g. NRAS and GATA2). Our observation that patients with leukemias exhibiting branched evolution showed improved survival compared to linear evolution was also striking and warrants further experimental and clinical validation. Incorporating these results with our findings of drug sensitivity validate the clinical utility of integrating clonal analysis into the molecular evaluation and treatment of AML. Disclosures Majeti: Zenshine Pharmaceuticals: Membership on an entity's Board of Directors or advisory committees; Kodikaz Therapeutic Solutions Inc.: Membership on an entity's Board of Directors or advisory committees; Stanford University: Patents & Royalties: pending patent application on CD93 CAR ; Coherus BioSciences: Membership on an entity's Board of Directors or advisory committees; BeyondSpring Pharmaceuticals: Membership on an entity's Board of Directors or advisory committees; Forty-Seven Inc.: Divested equity in a private or publicly-traded company in the past 24 months; Gilead Sciences, Inc.: Patents & Royalties: inventor on patents related to CD47 cancer immunotherapy; CircBio Inc.: Research Funding.

Q-Rank: Reinforcement Learning for Recommending Algorithms to Predict Drug Sensitivity to Cancer Therapy.

In personalized medicine, a challenging task is to identify the most effective treatment for a patient. In oncology, several computational models have been developed to predict the response of drugs to therapy. However, the performance of these models depends on multiple factors. This paper presents a new approach, called Q-Rank, to predict the sensitivity of cell lines to anti-cancer drugs. Q-Rank integrates different prediction algorithms and identifies a suitable algorithm for a given application. Q-Rank is based on reinforcement learning methods to rank prediction algorithms on the basis of relevant features (e.g., omics characterization). The best-ranked algorithm is recommended and used to predict the response of drugs to therapy. Our experimental results indicate that Q-Rank outperforms the integrated models in predicting the sensitivity of cell lines to different drugs.

Recent Advances in Drug Discovery and Cancer Diagnoses

This editorial provides a brief overview of the thematic issue and the papers in it. The thematic issue is proposed to help chemists and biologists track the most recent advances in drug discovery and cancer diagnoses. The process of drug discovery involves the identification and validation of biological targets, the identification and optimization of lead compounds, preclinical development, and clinical trials. Cancer is a major public health problem in the world. The results of tissue diagnosis, blood tests, computed tomography scans, and cytogenetic analyses can provide informative clues about molecular changes and indicate proper prognoses. Timely detection of cancer significantly improves cancer outcomes by providing care at the earliest possible stage thus contributing greatly to the prevention and exacerbation and has become an important public health strategy in all settings. The collection of this thematic issue includes five articles. The first one reviews the current advances and limitations of deep learning in anticancer drug sensitivity prediction. The next review summarizes the most recent and high-quality research related to anticancer activities of Vitamin C. The third one reports the efficacy of two different sets of natural products (terpenoids and flavonoids) towards caspase-3 activity. The fourth one proposes a novel in silico method for predicting cancer biomarkers in human body fluids. The fifth article performs an in silico and in vitro investigation on isothymusin, which serves as a potential inhibitor of cancer cell proliferation.

Current Advances and Limitations of Deep Learning in Anticancer Drug Sensitivity Prediction.

Anticancer drug screening can accelerate drug discovery to save the lives of cancer patients, but cancer heterogeneity makes this screening challenging. The prediction of anticancer drug sensitivity is useful for anticancer drug development and the identification of biomarkers of drug sensitivity. Deep learning, as a branch of machine learning, is an important aspect of in silico research. Its outstanding computational performance means that it has been used for many biomedical purposes, such as medical image interpretation, biological sequence analysis, and drug discovery. Several studies have predicted anticancer drug sensitivity based on deep learning algorithms. The field of deep learning has made progress regarding model performance and multi-omics data integration. However, deep learning is limited by the number of studies performed and data sources available, so it is not perfect as a pre-clinical approach for use in the anticancer drug screening process. Improving the performance of deep learning models is a pressing issue for researchers. In this review, we introduce the research of anticancer drug sensitivity prediction and the use of deep learning in this research area. To provide a reference for future research, we also review some common data sources and machine learning methods. Lastly, we discuss the advantages and disadvantages of deep learning, as well as the limitations and future perspectives regarding this approach.

Representation of features as images with neighborhood dependencies for compatibility with convolutional neural networks

Deep learning with Convolutional Neural Networks has shown great promise in image-based classification and enhancement but is often unsuitable for predictive modeling using features without spatial correlations. We present a feature representation approach termed REFINED (REpresentation of Features as Images with NEighborhood Dependencies) to arrange high-dimensional vectors in a compact image form conducible for CNN-based deep learning. We consider the similarities between features to generate a concise feature map in the form of a two-dimensional image by minimizing the pairwise distance values following a Bayesian Metric Multidimensional Scaling Approach. We hypothesize that this approach enables embedded feature extraction and, integrated with CNN-based deep learning, can boost the predictive accuracy. We illustrate the superior predictive capabilities of the proposed framework as compared to state-of-the-art methodologies in drug sensitivity prediction scenarios using synthetic datasets, drug chemical descriptors as predictors from NCI60, and both transcriptomic information and drug descriptors as predictors from GDSC.

Identification of Therapeutic and Prognostic Biomarkers of Lamin C (LAMC) Family Members in Head and Neck Squamous Cell Carcinoma.

BackgroundHead and neck squamous cell carcinoma (HNSC) is an invasive malignancy with a high worldwide mortality, despite considerable recent advancements in diagnosis and treatment. Increasing evidence indicates that the Lamin C (LAMC) gene family is associated with the progression of diverse cancers, nevertheless, this association is not well understood.Material/MethodsA systematic study addressing the expression and prognostic value of LAMC, and the relationship between LAMC and tumor immune response in HNSC was done. Finally, we performed drug screening to identify specific drugs.ResultsCompared to normal samples, expressions of LAMC1 and LAMC2 were significantly increased in HNSC, and LAMC2 was obviously correlated with an adverse prognosis for patients. LAMC2 expression level was significantly correlated with the infiltrating levels of B cells, CD8+ T cells, CD4+ T cells, and macrophages. Moreover, LAMC2 exhibited strong correlations with diverse immune markers, immune microenvironment, and immune checkpoint molecules. Finally, candidate drugs that targeted LAMC2 were identified.ConclusionsThis study suggests that LAMC2 could serve as a new prognostic biomarker, and it could be used for efficacy of target for immune response and for drug sensitivity prediction in HNSC.

Pathway-Guided Deep Neural Network toward Interpretable and Predictive Modeling of Drug Sensitivity.

To efficiently save cost and reduce risk in drug research and development, there is a pressing demand to develop in silico methods to predict drug sensitivity to cancer cells. With the exponentially increasing number of multi-omics data derived from high-throughput techniques, machine learning-based methods have been applied to the prediction of drug sensitivities. However, these methods have drawbacks either in the interpretability of the mechanism of drug action or limited performance in modeling drug sensitivity. In this paper, we presented a pathway-guided deep neural network (DNN) model to predict the drug sensitivity in cancer cells. Biological pathways describe a group of molecules in a cell that collaborates to control various biological functions like cell proliferation and death, thereby abnormal function of pathways can result in disease. To take advantage of the excellent predictive ability of DNN and the biological knowledge of pathways, we reshaped the canonical DNN structure by incorporating a layer of pathway nodes and their connections to input gene nodes, which makes the DNN model more interpretable and predictive compared to canonical DNN. We have conducted extensive performance evaluations on multiple independent drug sensitivity data sets and demonstrated that our model significantly outperformed the canonical DNN model and eight other classical regression models. Most importantly, we observed a remarkable activity decrease in disease-related pathway nodes during forward propagation upon inputs of drug targets, which implicitly corresponds to the inhibition effect of disease-related pathways induced by drug treatment on cancer cells. Our empirical experiments showed that our method achieves pharmacological interpretability and predictive ability in modeling drug sensitivity in cancer cells. The web server, the processed data sets, and source codes for reproducing our work are available at http://pathdnn.denglab.org.

Measuring the Contractile Force of Multilayered Human Cardiac Cell Sheets.

Three-dimensional (3D) cardiac tissue reconstruction using tissue engineering technology is a rapidly growing area of regenerative medicine and drug screening development. However, there remains an urgent need for the development of a method capable of accurately measuring the contractile force of physiologically relevant 3D myocardial tissues to facilitate the prediction of human heart tissue drug sensitivity. To this end, our laboratory has developed a novel drug screening model that measures the contractile force of cardiac cell sheets prepared using temperature-responsive culture dishes. To circumvent the difficulties that commonly arise during the stacking of cardiomyocyte sheets, we established a stacking method using centrifugal force, making it possible to measure 3D myocardial tissue. Human induced pluripotent stem cell-derived cardiomyocytes were seeded in a temperature-responsive culture dish and processed into a sheet. The cardiac cell sheets were multilayered to construct 3D cardiac tissue. Measurement of the contractile force and cross-sectional area of the multilayered 3D cardiac tissue were then obtained and used to determine the relationship between the cross-sectional area of the cardiac tissue and its contractile force. The contractile force of the 1-, 3-, and 5-layer tissues increased linearly in proportion to the cross-sectional area. A result of 6.4 mN/mm2, accounting for one-seventh of the contractile force found in adult tissue, was obtained. However, with 7-layer tissues, there was a sudden drop in the contractile force, possibly because of limited oxygen and nutrient supply. In conclusion, we established a method wherein the thickness of the cell sheets was controlled through layering, thus enabling accurate evaluation of the cardiac contractile function. This method may enable comparisons with living heart tissue while providing information applicable to regenerative medicine and drug screening models.

Abstract CT119: Organoid sensitivity in pancreatic cancer correlates with clinical response to treatment and reveals utility for reducing toxicity: Preliminary results from the HOPE trial

Abstract The HOPE trial (Harnessing Organoids for PErsonalized Therapy) was a pilot study to test the feasibility of generating patient derived organoids (PDOs) from patients with pancreatic cancer under real world conditions, test drug sensitivity against these PDOs, and correlate these findings with clinical outcomes. Biopsies were obtained primarily during routine clinical care from surgical specimens, ascites, fine needle biopsies (FNB) of primary tumors, and IR guided core biopsies of liver and lymph node metastases. PDOs were grown in WNT free media according to our previously published methods. PDO drug sensitivity testing was performed on a panel of drugs, AUC calculated, and sensitivity ranked. Patients were followed clinically and assessed for disease control. At data cutoff (January 2020), we enrolled a total of 76 subjects representing all stages of disease. Drug testing was performed successfully on PDOs generated from 12 of these subjects (16%). Factors contributing to success obtaining sufficient cells for PDO generation included modality, body part, and tumor cellularity. H&E and IHC corresponded in matched PDOs and donor tumors, as did DNA alterations. Transcriptomes of PDOs were classified as both ‘basal' and ‘classical' subtypes. When AUC values were annotated with clinical data, the Jenks break of 1.69 segregated matched PDO/AUC values into disease control and progressive disease. We estimated that a PDO AUC value <1.66 yields a > 99% probability of disease control from a regimen that contains this drug, whereas if all drugs in a regimen had an AUC > 2.75 there is a > 80% probability of accurately predicting resistance. To illustrate the potential of PDO testing to tailor treatment for an individual patient, we described a case of a subject with stage IV PDAC with a KRAS mutation and ERBB2 amplification. The subject had disease control with FOLFIRINOX, which was held for toxicity. The PDO showed resistance to oxaliplatin and the patient subsequently had an extended period of disease control with regimens which did not include oxaliplatin, highlighting the potential of PDO drug sensitivity testing to exclude ineffective treatments from combination chemotherapy and limit toxicity. In conclusion, we have shown the feasibility of collecting material via real-world clinical practice sufficient to develop PDOs suitable for rapidly screening multiple drugs, and have shown a high degree of correlation between clinical outcomes in patients with PDAC and matched PDO drug sensitivity. We determined preliminary criteria based on the AUC of individual drugs in PDOs to predict drug sensitivity in subjects. These results highlight the potential of PDOs to personalize therapy and allow for the exclusion of ineffective drugs from combination regimens thereby reducing toxicity. We anticipate this approach will be used in future trials to prospectively inform treatment selection for patients with PDAC. Citation Format: Joseph Elan Grossman, Ling Huang, Lakshmi Muthuswamy, Sofia Perea, Dipikaa Akshinthala, Raul Gonzalez, Leo Tsai, Jonah Cohen, Mandeep Sawhney, Douglas Pleskow, Tyler M. Berzin, Bruno Bockorny, Andrea Bullock, Benjamin Schlechter, Mary Linton B. Peters, Catherine Conahan, Supraja Narasimhan, Christine Lim, Roger Davis, Robert Besaw, Martin Smith, Tara Kent, Mark Callery, Senthil K. Muthuswamy, Manuel Hidalgo. Organoid sensitivity in pancreatic cancer correlates with clinical response to treatment and reveals utility for reducing toxicity: Preliminary results from the HOPE trial [abstract]. In: Proceedings of the Annual Meeting of the American Association for Cancer Research 2020; 2020 Apr 27-28 and Jun 22-24. Philadelphia (PA): AACR; Cancer Res 2020;80(16 Suppl):Abstract nr CT119.

Drug sensitivity prediction with normal inverse Gaussian shrinkage informed by external data.

In precision medicine, a common problem is drug sensitivity prediction from cancer tissue cell lines. These types of problems entail modelling multivariate drug responses on high‐dimensional molecular feature sets in typically >1000 cell lines. The dimensions of the problem require specialised models and estimation methods. In addition, external information on both the drugs and the features is often available. We propose to model the drug responses through a linear regression with shrinkage enforced through a normal inverse Gaussian prior. We let the prior depend on the external information, and estimate the model and external information dependence in an empirical‐variational Bayes framework. We demonstrate the usefulness of this model in both a simulated setting and in the publicly available Genomics of Drug Sensitivity in Cancer data.

Proteome activity landscapes of tumor cell lines determine drug responses

Integrated analysis of genomes, transcriptomes, proteomes and drug responses of cancer cell lines (CCLs) is an emerging approach to uncover molecular mechanisms of drug action. We extend this paradigm to measuring proteome activity landscapes by acquiring and integrating quantitative data for 10,000 proteins and 55,000 phosphorylation sites (p-sites) from 125 CCLs. These data are used to contextualize proteins and p-sites and predict drug sensitivity. For example, we find that Progesterone Receptor (PGR) phosphorylation is associated with sensitivity to drugs modulating estrogen signaling such as Raloxifene. We also demonstrate that Adenylate kinase isoenzyme 1 (AK1) inactivates antimetabolites like Cytarabine. Consequently, high AK1 levels correlate with poor survival of Cytarabine-treated acute myeloid leukemia patients, qualifying AK1 as a patient stratification marker and possibly as a drug target. We provide an interactive web application termed ATLANTiC (http://atlantic.proteomics.wzw.tum.de), which enables the community to explore the thousands of novel functional associations generated by this work.