Deep learning allows computational models that are composed of multiple processing layers to learn representations of data with multiple levels of abstraction. These methods have dramatically improved the state-of-the-art in speech recognition, visual object recognition, object detection and many other domains such as drug discovery and genomics. Deep learning discovers intricate structure in large data sets by using the backpropagation algorithm to indicate how a machine should change its internal parameters that are used to compute the representation in each layer from the representation in the previous layer. Deep convolutional nets have brought about breakthroughs in processing images, video, speech and audio, whereas recurrent nets have shone light on sequential data such as text and speech.

Deep learning

Deep learning is a form of machine learning that enables computers to learn from experience and understand the world in terms of a hierarchy of concepts. Because the computer gathers knowledge from experience, there is no need for a human computer operator to formally specify all the knowledge that the computer needs. The hierarchy of concepts allows the computer to learn complicated concepts by building them out of simpler ones; a graph of these hierarchies would be many layers deep. This book introduces a broad range of topics in deep learning. The text offers mathematical and conceptual background, covering relevant concepts in linear algebra, probability theory and information theory, numerical computation, and machine learning. It describes deep learning techniques used by practitioners in industry, including deep feedforward networks, regularization, optimization algorithms, convolutional networks, sequence modeling, and practical methodology; and it surveys such applications as natural language processing, speech recognition, computer vision, online recommendation systems, bioinformatics, and videogames. Finally, the book offers research perspectives, covering such theoretical topics as linear factor models, autoencoders, representation learning, structured probabilistic models, Monte Carlo methods, the partition function, approximate inference, and deep generative models. Deep Learning can be used by undergraduate or graduate students planning careers in either industry or research, and by software engineers who want to begin using deep learning in their products or platforms. A website offers supplementary material for both readers and instructors.

Deep Learning

The reference human genome sequence set the stage for studies of genetic variation and its association with human disease, but epigenomic studies lack a similar reference. To address this need, the NIH Roadmap Epigenomics Consortium generated the largest collection so far of human epigenomes for primary cells and tissues. Here we describe the integrative analysis of 111 reference human epigenomes generated as part of the programme, profiled for histone modification patterns, DNA accessibility, DNA methylation and RNA expression. We establish global maps of regulatory elements, define regulatory modules of coordinated activity, and their likely activators and repressors. We show that disease- and trait-associated genetic variants are enriched in tissue-specific epigenomic marks, revealing biologically relevant cell types for diverse human traits, and providing a resource for interpreting the molecular basis of human disease. Our results demonstrate the central role of epigenomic information for understanding gene regulation, cellular differentiation and human disease.

Integrative analysis of 111 reference human epigenomes

The use of artificial intelligence, and the deep-learning subtype in particular, has been enabled by the use of labeled big data, along with markedly enhanced computing power and cloud storage, across all sectors. In medicine, this is beginning to have an impact at three levels: for clinicians, predominantly via rapid, accurate image interpretation; for health systems, by improving workflow and the potential for reducing medical errors; and for patients, by enabling them to process their own data to promote health. The current limitations, including bias, privacy and security, and lack of transparency, along with the future directions of these applications will be discussed in this article. Over time, marked improvements in accuracy, productivity, and workflow will likely be actualized, but whether that will be used to improve the patient-doctor relationship or facilitate its erosion remains to be seen.

High-performance medicine: the convergence of human and artificial intelligence

1.1 Photodynamic Therapy and Imaging
The purpose of this review is to present the current state of the role of imaging in photodynamic therapy (PDT). In order for the reader to fully appreciate the context of the discussions embodied in this article we begin with an overview of the PDT process, starting with a brief historical perspective followed by detailed discussions of specific applications of imaging in PDT. Each section starts with an overview of the specific topic and, where appropriate, ends with summary and future directions. The review closes with the authors’ perspective of the areas of future emphasis and promise. The basic premise of this review is that a combination of imaging and PDT will provide improved research and therapeutic strategies.

PDT is a photochemistry-based approach that uses a light-activatable chemical, termed a photosensitizer (PS), and light of an appropriate wavelength, to impart cytotoxicity via the generation of reactive molecular species (Figure 1a). In clinical settings, the PS is typically administered intravenously or topically, followed by illumination using a light delivery system suitable for the anatomical site being treated (Figure 1b). The time delay, often referred to as drug-light interval, between PS administration and the start of illumination with currently used PSs varies from 5 minutes to 24 hours or more depending on the specific PS and the target disease. Strictly speaking, this should be referred to as the PS-light interval, as at the concentrations typically used the PS is not a drug, but the drug-light interval terminology seems to be used fairly frequently. Typically, the useful range of wavelengths for therapeutic activation of the PS is 600 to 800 nm, to avoid interference by endogenous chromophores within the body, and yet maintain the energetics necessary for the generation of cytotoxic species (as discussed below) such as singlet oxygen (1O2). However, it is important to note that photosensitizers can also serve as fluorescence imaging agents for which activation with light in the 400nm range is often used and has been extremely useful in diagnostic imaging applications as described extensively in Section 2 of this review. The obvious limitation of short wavelength excitation is the lack of tissue penetration so that the volumes that are probed under these conditions are relatively shallow.






Open in a separate window


Figure 1


(A) A schematic representation of PDT where PS is a photoactivatable multifunctional agent, which, upon light activation can serve as both an imaging agent and a therapeutic agent. (B) A schematic representation of the sequence of administration, localization and light activation of the PS for PDT or fluorescence imaging. Typically the PS is delivered systemically and allowed to circulate for an appropriate time interval (the “drug-light interval”), during which the PS accumulates preferentially in the target lesion(s) prior to light activation. In the idealized depiction here the PS is accumulation is shown to be entirely in the target tissue, however, even if this is not the case, light delivery confers a second layer of selectivity so that the cytotoxic effect will be generated only in regions where both drug and light are present. Upon localization of the PS, light activation will result in fluorescence emission which can be implemented for imaging applications, as well as generation cytotoxic species for therapy. In the former case light activation is achieved with a low fluence rate to generate fluorescence emission with little or no cytotoxic effect, while in the latter case a high fluence rate is used to generate a sufficient concentration of cytotoxic species to achieve biological effects.

/pdf/imaging-and-photodynamic-therapy-mechanisms-monitoring-and-76rc7acuq5.pdf

Imaging and photodynamic therapy: mechanisms, monitoring, and optimization.

We report on imaging of microcirculation by calculating the speckle variance of optical coherence tomography (OCT) structural images acquired using a Fourier domain mode-locked swept-wavelength laser. The algorithm calculates interframe speckle variance in two-dimensional and three-dimensional OCT data sets and shows little dependence to the Doppler angle ranging from 75 degrees to 90 degrees . We demonstrate in vivo detection of blood flow in vessels as small as 25 microm in diameter in a dorsal skinfold window chamber model with direct comparison with intravital fluorescence confocal microscopy. This technique can visualize vessel-size-dependent vascular shutdown and transient vascular occlusion during Visudyne photodynamic therapy and may provide opportunities for studying therapeutic effects of antivascular treatments without on exogenous contrast agent.

/pdf/speckle-variance-detection-of-microvasculature-using-swept-3heito4d44.pdf

Speckle variance detection of microvasculature using swept-source optical coherence tomography

Motivation: Alternative splicing (AS) is a regulated process that directs the generation of different transcripts from single genes. A computational model that can accurately predict splicing patterns based on genomic features and cellular context is highly desirable, both in understanding this widespread phenomenon, and in exploring the effects of genetic variations on AS. Methods: Using a deep neural network, we developed a model inferred from mouse RNA-Seq data that can predict splicing patterns in individual tissues and differences in splicing patterns across tissues. Our architecture uses hidden variables that jointly represent features in genomic sequences and tissue types when making predictions. A graphics processing unit was used to greatly reduce the training time of our models with millions of parameters. Results: We show that the deep architecture surpasses the performance of the previous Bayesian method for predicting AS patterns. With the proper optimization procedure and selection of hyperparameters, we demonstrate that deep architectures can be beneficial, even with a moderately sparse dataset. An analysis of what the model has learned in terms of the genomic features is presented. Contact: frey@psi.toronto.edu Supplementary information: Supplementary data are available at Bioinformatics online.

Deep learning of the tissue-regulated splicing code.

We optimize speckle variance optical coherence tomography (svOCT) imaging of microvasculature in high and low bulk tissue motion scenarios. To achieve a significant level of image contrast, frame rates must be optimized such that tissue displacement between frames is less than the beam radius. We demonstrate that higher accuracy estimates of speckle variance can enhance the detection of capillaries. These findings are illustrated in vivo by imaging the dorsal window chamber model (low bulk motion). We also show svOCT imaging of the nonstabilized finger (high bulk motion), using optimized imaging parameters, demonstrating better vessel detection than Doppler OCT.

/pdf/optimized-speckle-variance-oct-imaging-of-microvasculature-17tmgg3ess.pdf

Optimized speckle variance OCT imaging of microvasculature

Purpose: The aim of this study is to understand the characteristics of secondary electrons generated from the interaction of gold nanoparticles (GNPs) with x-rays as a function of nanoparticle size and beam energy and thereby further the understanding of GNP-enhanced radiotherapy. Methods: The effective range, deflection angle, dose deposition, energy, and interaction processes of electrons produced from the interaction of x-rays with a GNP were calculated by Monte Carlo simulations. The GEANT4 code was used to simulate and track electrons generated from a 2, 50, and 100 nm diameter GNP when it is irradiated with a 50 kVp, 250 kVp, cobalt-60, and 6 MV photon beam in water. Results: When a GNP was present, depending on the beam energies used, secondary electron production was increased by 10- to 2000-fold compared to an absence of a GNP. Low-energy photon beams were much more efficient at interacting with the GNP by two to three orders of magnitude compared to MV energies and increased the deflection angle. GNPs with larger diameters also contributed more dose. The majority of the energy deposition was outside the GNP, rather than self-absorbed by the nanoparticle. The mean effective range of electron tracks for the beams tested rangedmore » from approximately 3 {mu}m to 1 mm. Conclusions: These simulated results yield important insights concerning the spatial distributions and elevated dose in GNP-enhanced radiotherapy. The authors conclude that the irradiation of GNP at lower photon energies will be more efficient for cell killing. This conclusion is consistent with published studies.« less

Irradiation of gold nanoparticles by x-rays: Monte Carlo simulation of dose enhancements and the spatial properties of the secondary electrons production

In this paper, we provide an introduction to machine learning tasks that address important problems in genomic medicine. One of the goals of genomic medicine is to determine how variations in the DNA of individuals can affect the risk of different diseases, and to find causal explanations so that targeted therapies can be designed. Here we focus on how machine learning can help to model the relationship between DNA and the quantities of key molecules in the cell, with the premise that these quantities, which we refer to as cell variables, may be associated with disease risks. Modern biology allows high-throughput measurement of many such cell variables, including gene expression, splicing, and proteins binding to nucleic acids, which can all be treated as training targets for predictive models. With the growing availability of large-scale data sets and advanced computational techniques such as deep learning, researchers can help to usher in a new era of effective genomic medicine.

Michael K. K. Leung

Papers

Speckle variance detection of microvasculature using swept-source optical coherence tomography

Deep learning of the tissue-regulated splicing code.

Optimized speckle variance OCT imaging of microvasculature

Irradiation of gold nanoparticles by x-rays: Monte Carlo simulation of dose enhancements and the spatial properties of the secondary electrons production

Machine Learning in Genomic Medicine: A Review of Computational Problems and Data Sets