Abstract
Despite the fact that deep learning has achieved remarkable success in various domains over the past decade, its application in molecular informatics and drug discovery is still limited. ...Recent advances in adapting deep architectures to structured data have opened a new paradigm for pharmaceutical research. In this survey, we provide a systematic review on the emerging field of graph convolutional networks and their applications in drug discovery and molecular informatics. Typically we are interested in why and how graph convolution networks can help in drug-related tasks. We elaborate the existing applications through four perspectives: molecular property and activity prediction, interaction prediction, synthesis prediction and de novo drug design. We briefly introduce the theoretical foundations behind graph convolutional networks and illustrate various architectures based on different formulations. Then we summarize the representative applications in drug-related problems. We also discuss the current challenges and future possibilities of applying graph convolutional networks to drug discovery.
Artificial intelligence and machine learning techniques are breaking into biomedical research and health care, which importantly includes cancer research and oncology, where the potential ...applications are vast. These include detection and diagnosis of cancer, subtype classification, optimization of cancer treatment and identification of new therapeutic targets in drug discovery. While big data used to train machine learning models may already exist, leveraging this opportunity to realize the full promise of artificial intelligence in both the cancer research space and the clinical space will first require significant obstacles to be surmounted. In this Viewpoint article, we asked four experts for their opinions on how we can begin to implement artificial intelligence while ensuring standards are maintained so as transform cancer diagnosis and the prognosis and treatment of patients with cancer and to drive biological discovery.
Pathological evaluation of tumor tissue is pivotal for diagnosis in cancer patients and automated image analysis approaches have great potential to increase precision of diagnosis and help reduce ...human error.
In this study, we utilize several computational methods based on convolutional neural networks (CNN) and build a stand-alone pipeline to effectively classify different histopathology images across different types of cancer.
In particular, we demonstrate the utility of our pipeline to discriminate between two subtypes of lung cancer, four biomarkers of bladder cancer, and five biomarkers of breast cancer. In addition, we apply our pipeline to discriminate among four immunohistochemistry (IHC) staining scores of bladder and breast cancers.
Our classification pipeline includes a basic CNN architecture, Google's Inceptions with three training strategies, and an ensemble of two state-of-the-art algorithms, Inception and ResNet. Training strategies include training the last layer of Google's Inceptions, training the network from scratch, and fine-tunning the parameters for our data using two pre-trained version of Google's Inception architectures, Inception-V1 and Inception-V3.
We demonstrate the power of deep learning approaches for identifying cancer subtypes, and the robustness of Google's Inceptions even in presence of extensive tumor heterogeneity. On average, our pipeline achieved accuracies of 100%, 92%, 95%, and 69% for discrimination of various cancer tissues, subtypes, biomarkers, and scores, respectively. Our pipeline and related documentation is freely available at https://github.com/ih-_lab/CNN_Smoothie.
•This work presents the power of deep learning approaches for classifying pathology images in presence of extensive tumor heterogeneity.•A basic CNN architecture and several state-of-the-art image classification architectures are compared in an open source pipeline.•The pipeline is applicable to wide rang of histopathology images.
Both clinical and genetic factors drive the risk of venous thromboembolism. However, whether clinically recorded risk factors and genetic variants can be combined into a clinically applicable ...predictive score remains unknown. Using Cox proportional-hazard models, we analyzed the association of risk factors with the likelihood of venous thromboembolism in U.K. Biobank, a large prospective cohort. We then created a polygenic risk score of 36 single nucleotide polymorphisms and a clinical score determined by age, sex, body mass index, previous cancer diagnosis, smoking status, and fracture in the last 5 years. Participants were at significantly increased risk of venous thromboembolism if they were at high clinical risk (subhazard ratio, 4.37 95% CI, 3.85-4.97) or high genetic risk (subhazard ratio, 3.02 95% CI, 2.63-3.47) relative to participants at low clinical or genetic risk, respectively. The combined model, consisting of clinical and genetic components, was significantly better than either the clinical or the genetic model alone (P < 0.001). Participants at high risk in the combined score had nearly an eightfold increased risk of venous thromboembolism relative to participants at low risk (subhazard ratio, 7.51 95% CI, 6.28-8.98). This risk score can be used to guide decisions regarding venous thromboembolism prophylaxis, although external validation is needed.
The Missing Pieces of Artificial Intelligence in Medicine Gilvary, Coryandar; Madhukar, Neel; Elkhader, Jamal ...
Trends in pharmacological sciences (Regular ed.),
August 2019, 2019-08-00, 20190801, Letnik:
40, Številka:
8
Journal Article
Recenzirano
Stakeholders across the entire healthcare chain are looking to incorporate artificial intelligence (AI) into their decision-making process. From early-stage drug discovery to clinical decision ...support systems, we have seen examples of how AI can improve efficiency and decrease costs. In this Opinion, we discuss some of the key factors that should be prioritized to enable the successful integration of AI across the healthcare value chain. In particular, we believe a focus on model interpretability is crucial to obtain a deeper understanding of the underlying biological mechanisms and guide further investigations. Additionally, we discuss the importance of integrating diverse types of data within any AI framework to limit bias, increase accuracy, and model the interdisciplinary nature of medicine. We believe that widespread adoption of these practices will help accelerate the continued integration of AI into our current healthcare framework.
Deep neural networks have revolutionized the field of computer vision, which has subsequently led to advances in AI-driven dermatology, radiology, pathology, and many other fields of research.A number of different AI methods have shown promise when applied to bottlenecks in early-stage drug development – such as biomarker prediction and drug repurposing.The availability of large-scale data in the form of high-throughput screenings, chemical structures, and electronic health records allow for novel research approaches to be utilized.As AI is becoming more commonly used within precision medicine, it is important to evaluate models not only on traditional performance measures, but model interpretability and data integration as well.
Over the past decade, the rate of drug attrition due to clinical trial failures has risen substantially. Unfortunately it is difficult to identify compounds that have unfavorable toxicity properties ...before conducting clinical trials. Inspired by the effective use of sabermetrics in predicting successful baseball players, we sought to use a similar “moneyball” approach that analyzes overlooked features to predict clinical toxicity. We introduce a new data-driven approach (PrOCTOR) that directly predicts the likelihood of toxicity in clinical trials. PrOCTOR integrates the properties of a compound's targets and its structure to provide a new measure, the PrOCTOR score. Drug target network connectivity and expression levels, along with molecular weight, were identified as important indicators of adverse clinical events. Our method provides a data-driven, broadly applicable strategy to identify drugs likely to possess manageable toxicity in clinical trials and will help drive the design of therapeutic agents with less toxicity.
•Computational approach predicts the likelihood of clinical trial toxicity•Identification of molecule and target properties associated with clinical toxicity•Development of a tool to facilitate interaction and interpretation of the model
Gayvert et al. present a data-driven approach that accurately predicts the likelihood of clinical trial toxicity by integrating the structural and target-based properties of a drug.
Methylation of the N6 position of adenosine (m6A) is a posttranscriptional modification of RNA with poorly understood prevalence and physiological relevance. The recent discovery that FTO, an obesity ...risk gene, encodes an m6A demethylase implicates m6A as an important regulator of physiological processes. Here, we present a method for transcriptome-wide m6A localization, which combines m6A-specific methylated RNA immunoprecipitation with next-generation sequencing (MeRIP-Seq). We use this method to identify mRNAs of 7,676 mammalian genes that contain m6A, indicating that m6A is a common base modification of mRNA. The m6A modification exhibits tissue-specific regulation and is markedly increased throughout brain development. We find that m6A sites are enriched near stop codons and in 3′ UTRs, and we uncover an association between m6A residues and microRNA-binding sites within 3′ UTRs. These findings provide a resource for identifying transcripts that are substrates for adenosine methylation and reveal insights into the epigenetic regulation of the mammalian transcriptome.
Display omitted
► m6A is a widespread RNA modification in many tissues with high levels in the brain ► MeRIP-Seq identifies m6A in 7,913 genes encoding both coding and noncoding RNAs ► m6A is enriched near stop codons and within 3′ UTRs in both mouse and human mRNAs ► The transcriptome-wide landscape of m6A provides important insights into m6A function
Adenosine methylation (m6A) is a widespread RNA modification found in >7,000 mammalian genes, encoding both mRNAs and noncoding RNAs, with an especially high prevalence in the developing brain. In both mouse and human mRNAs, m6A is enriched within 3′ UTRs that also contain miRNA-binding sites.
Muscle regeneration relies on the regulation of muscle stem cells (MuSCs) through paracrine signaling interactions. We analyzed muscle regeneration in mice using single-cell RNA sequencing ...(scRNA-seq) and generated over 34,000 single-cell transcriptomes spanning four time-points. We identified 15 distinct cell types including heterogenous populations of muscle stem and progenitor cells. We resolved a hierarchical map of these myogenic cells by trajectory inference and observed stage-specific regulatory programs within this continuum. Through ligand-receptor interaction analysis, we identified over 100 candidate regeneration-associated paracrine communication pairs between MuSCs and non-myogenic cells. We show that myogenic stem/progenitor cells exhibit heterogeneous expression of multiple Syndecan proteins in cycling myogenic cells, suggesting that Syndecans may coordinate myogenic fate regulation. We performed ligand stimulation in vitro and confirmed that three paracrine factors (FGF2, TGFβ1, and RSPO3) regulate myogenic cell proliferation in a Syndecan-dependent manner. Our study provides a scRNA-seq reference resource to investigate cell communication interactions in muscle regeneration.
Display omitted
•Single-cell RNA-sequencing identifies cell populations involved in muscle regeneration•Muscle stem/progenitor cells form a hierarchy with stage-specific regulatory programs•Bioinformatic analysis identified paracrine factors influencing muscle stem cells•Syndecan-1/2/4 coordinate paracrine ligand-specific muscle progenitor proliferation
De Micheli et al. present an annotated, time-resolved single-cell transcriptomic atlas of muscle regeneration in adult mice. They observe a hierarchy of muscle stem and progenitor cells that exhibit stage-specific expression programs and show that Syndecan proteins regulate muscle progenitor cell fates by interaction with newly discovered paracrine communication factors.
The expression of antigens that are recognized by self-reactive T cells is essential for immune-mediated tumor rejection by immune checkpoint blockade (ICB) therapy. Growing evidence suggests that ...mutation-associated neoantigens drive ICB responses in tumors with high mutational burden. In most patients, only a few of the mutations in the cancer exome that are predicted to be immunogenic are recognized by T cells. One factor that limits this recognition is the level of expression of the mutated gene product in cancer cells. Substantial preclinical data show that radiation can convert the irradiated tumor into a site for priming of tumor-specific T cells, that is, an in situ vaccine, and can induce responses in otherwise ICB-resistant tumors. Critical for radiation-elicited T-cell activation is the induction of viral mimicry, which is mediated by the accumulation of cytosolic DNA in the irradiated cells, with consequent activation of the cyclic GMP-AMP synthase (cGAS)/stimulator of interferon (IFN) genes (STING) pathway and downstream production of type I IFN and other pro-inflammatory cytokines. Recent data suggest that radiation can also enhance cancer cell antigenicity by upregulating the expression of a large number of genes that are involved in the response to DNA damage and cellular stress, thus potentially exposing immunogenic mutations to the immune system. Here, we discuss how the principles of antigen presentation favor the presentation of peptides that are derived from newly synthesized proteins in irradiated cells. These concepts support a model that incorporates the presence of immunogenic mutations in genes that are upregulated by radiation to predict which patients might benefit from treatment with combinations of radiotherapy and ICB.