IMPORTANCEDeep learning is a family of computational methods that allow an algorithm to program itself by learning from a large set of examples that demonstrate the desired behavior, removing the need to specify rules explicitly. Application of these methods to medical imaging requires further assessment and validation.OBJECTIVE To apply deep learning to create an algorithm for automated detection of diabetic retinopathy and diabetic macular edema in retinal fundus photographs. DESIGN AND SETTINGA specific type of neural network optimized for image classification called a deep convolutional neural network was trained using a retrospective development data set of 128 175 retinal images, which were graded 3 to 7 times for diabetic retinopathy, diabetic macular edema, and image gradability by a panel of 54 US licensed ophthalmologists and ophthalmology senior residents between May and December 2015. The resultant algorithm was validated in January and February 2016 using 2 separate data sets, both graded by at least 7 US board-certified ophthalmologists with high intragrader consistency.EXPOSURE Deep learning-trained algorithm. MAIN OUTCOMES AND MEASURESThe sensitivity and specificity of the algorithm for detecting referable diabetic retinopathy (RDR), defined as moderate and worse diabetic retinopathy, referable diabetic macular edema, or both, were generated based on the reference standard of the majority decision of the ophthalmologist panel. The algorithm was evaluated at 2 operating points selected from the development set, one selected for high specificity and another for high sensitivity. RESULTSThe EyePACS-1 data set consisted of 9963 images from 4997 patients (mean age, 54.4 years; 62.2% women; prevalence of RDR, 683/8878 fully gradable images [7.8%]); the Messidor-2 data set had 1748 images from 874 patients (mean age, 57.6 years; 42.6% women; prevalence of RDR, 254/1745 fully gradable images [14.6%]). For detecting RDR, the algorithm had an area under the receiver operating curve of 0.991 (95% CI, 0.988-0.993) for EyePACS-1 and 0.990 (95% CI, 0.986-0.995) for Messidor-2. Using the first operating cut point with high specificity, for EyePACS-1, the sensitivity was 90.3% (95% CI, 87.5%-92.7%) and the specificity was 98.1% (95% CI, 97.8%-98.5%). For Messidor-2, the sensitivity was 87.0% (95% CI, 81.1%-91.0%) and the specificity was 98.5% (95% CI, 97.7%-99.1%). Using a second operating point with high sensitivity in the development set, for EyePACS-1 the sensitivity was 97.5% and specificity was 93.4% and for Messidor-2 the sensitivity was 96.1% and specificity was 93.9%. CONCLUSIONS AND RELEVANCEIn this evaluation of retinal fundus photographs from adults with diabetes, an algorithm based on deep machine learning had high sensitivity and specificity for detecting referable diabetic retinopathy. Further research is necessary to determine the feasibility of applying this algorithm in the clinical setting and to determine whether use of the algorithm could lead to improved care and outcomes compared with curren...
Because of thermal fluctuations, a polymer chain at room temperature will be bent. WLC and our model both predict that the molecule's observed conformations will be drawn from a certain probability distribution of shapes, obtained by combining bends distributed according to Boltzmann statistics with some energy function E(θ). Our task is to evaluate E(θ) from data.A comprehensive theory of DNA bending on short length scales must also include the twist degrees of freedom [1], as well as inhomogeneities from sequence [2,3]. Indeed, recent cyclization experiments suggest that the harmonic-elasticity model for the twist response of DNA also overstates the energetic cost of twist when curvature is high [4]. Thus, it seems likely that the twist energy function must be modified in a manner analogous to the one we have proposed for the bending energy. We leave this generalization to future work. For the random sequences studied here, we expect bending anisotropy to be a small effect for behavior on length scales greater than the helical pitch of 3.5 nm. Sequence dependent curvature, in natural DNA and in modified constructs with nonstandard bases, has been observed in AFM studies [5,6]; again we leave the extension of our model to include sequence to future work. A.2 Scale dependence in equilibrium statistical physicsHere we briefly elaborate on some ideas of scale dependence in equilibrium statistical physics, applied to our problem.The conformation of a macromolecule like DNA can usefully be described on any of several length scales. That is, when describing the molecule's behavior on a length scale ℓ exp larger than the size of individual atoms, we can often simplify our description by imagining the macromolecule to be composed
Microscopy is a central method in life sciences. Many popular methods, such as antibody labeling, are used to add physical fluorescent labels to specific cellular constituents. However, these approaches have significant drawbacks, including inconsistency; limitations in the number of simultaneous labels because of spectral overlap; and necessary perturbations of the experiment, such as fixing the cells, to generate the measurement. Here, we show that a computational machine-learning approach, which we call "in silico labeling" (ISL), reliably predicts some fluorescent labels from transmitted-light images of unlabeled fixed or live biological samples. ISL predicts a range of labels, such as those for nuclei, cell type (e.g., neural), and cell state (e.g., cell death). Because prediction happens in silico, the method is consistent, is not limited by spectral overlap, and does not disturb the experiment. ISL generates biological measurements that would otherwise be problematic or impossible to acquire.
We investigate the statistical mechanics of a torsionally constrained polymer. The polymer is modeled as a fluctuating rod with bend stiffness A kT and twist stiffness C kT. In such a model, thermal bend fluctuations couple geometrically to an applied torque through the relation Lk = Tw + Wr. We explore this coupling and find agreement between the predictions of our model and recent experimental results on single lambda-DNA molecules. This analysis affords an experimental determination of the microscopic twist stiffness (averaged over a helix repeat). Quantitative agreement between theory and experiment is obtained using C=109 nm. The theory further predicts a thermal reduction of the effective twist rigidity induced by bend fluctuations. Finally, we find a small reflection of molecular chirality in the experimental data and interpret it in terms of a twist-stretch coupling of the DNA duplex.Comment: 37 pages RevTeX, 2 postscript figures. Revisions include the analysis of new data and an investigation of non-perturbative effects. Postscript also available at http://www.physics.upenn.edu/~moro
The theory of random walks is one of the most fundamental problems in statistical mechanics, with applications throughout physics, biology, and even finance. The discovery that polymer conformations afford a concrete realization of this mathematical problem, and the understanding that rubber elasticity is inherently an entropic phenomenon, marked the birth of polymer physics (1). Remarkably, it recently has become possible to apply minuscule forces to single molecules of DNA in solution and observe their extension (2). Besides allowing a detailed confirmation of the directed random walk model of entropic elasticity, these experiments allow direct physical measurement of microscopic (nanometer-scale) mechanical properties of DNA relevant to its function, by using mesoscopic (micron-scale) apparatus. Two linear elastic parameters of DNA now have been measured in this way: the bend persistence length A and the intrinsic-stretch modulus ␥ (3-9).DNA and other stiff biopolymers differ from classical polymers, however, in that they exhibit torsional as well as bend stiffness. Thus their conformations reflect not a classical directed walk but a new fundamental problem: the ''torsional directed walk'' (TDW), whose random variables are the direction of each step relative to its predecessor, together with a relative axial twist. In this paper we will formulate the version of the TDW appropriate to DNA, solve it analytically in a regime appropriate to a recent experiment (10), and show that the model quantitatively fits the data over a wide range of applied forces (Fig. 1). Some of these results were announced in ref. 11; related work on the scaling limit of the TDW appeared in ref. 12. Besides being transparent, analytic formulae permit systematic least-squares fitting to experimental data. We fit to obtain three microscopic elastic constants: the twist persistence length C, bend persistence length A, and intrinsic twist-stretch coupling D (13-15). Because A is known independently we have a check on the model. The experiment is not sensitive to the other allowed linear-elastic constants such as twist-bend coupling (16).We find that the existing data (10) yield A ϭ 49 nm, C ϭ 120 nm, and D ϭ 50 nm; future experiments will refine these values when fit to our formula. Many authors have sought to extract the value of C from both cyclization experiments and fluorescence depolarization (17)(18)(19)(20). A key point of this paper is that the force regime we study is free from some vexing physical and mathematical difficulties that have helped make the determination of C from these experiments controversial.In particular, we can use a continuum model with no need for the short-length cutoff required to make Monte Carlo calculations tractable (ref. 21; C. Bouchiat and M. Mézard,. Our value for D is similar to within the large errors to recent estimates (13-15).We also give a simple analytical prediction for the reduction of effective twist stiffness by bend fluctuations. This renormalization may explain why some other determination...
The importance of nonlinearities in material constitutive relations has long been appreciated in the continuum mechanics of macroscopic rods. Although the moment ͑torque͒ response to bending is almost universally linear for small deflection angles, many rod systems exhibit a high-curvature softening. The signature behavior of these rod systems is a kinking transition in which the bending is localized. Recent DNA cyclization experiments by Cloutier and Widom have offered evidence that the linear-elastic bending theory fails to describe the high-curvature mechanics of DNA. Motivated by this recent experimental work, we develop a simple and exact theory of the statistical mechanics of linear-elastic polymer chains that can undergo a kinking transition. We characterize the kinking behavior with a single parameter and show that the resulting theory reproduces both the low-curvature linear-elastic behavior which is already well described by the wormlike chain model, as well as the high-curvature softening observed in recent cyclization experiments.
We give a theoretical analysis of bead motion in tethered-particle experiments, a single-molecule technique that has been used to explore the dynamics of a variety of macromolecules of biological interest. Our analysis reveals that the proximity of the tethered bead to a nearby surface gives rise to a volumeexclusion effect, resulting in an entropic stretching-force on the molecule that changes its statistical properties. In addition, volume exclusion brings about intriguing scaling relations between key observables (statistical moments of the bead) and parameters such as bead size and contour length of the molecule. We present analytic and numerical results for these effects in both flexible and semiflexible tethers. Finally, our results give a precise, experimentally testable prediction for the probability distribution of the bead center measured from the polymer attachment point. DOI: 10.1103/PhysRevLett.96.088306 PACS numbers: 82.37.Rs, 36.20.Ey, 82.35.Pq, 87.14.Gg Single-molecule biophysics has rapidly become an experimental centerpiece in the dissection of cellular machinery. This part of the biophysics repertoire often relies, in turn, on the use of micron-scale beads as a reporter of underlying molecular motions in these single-molecule systems. Thus, a key part of the theoretical infrastructure of this field is a clear understanding of the role that these beads play in altering the statistical properties of the macromolecules which are the real target of interest in such experiments. Beyond interest in the in vitro consequences of tethered-particle motions, many processes within the cell themselves involve tethering. The statisticalmechanical analysis performed here may prove useful for understanding in vivo processes, in addition to the in vitro consequences that form the main motivation for the work.Figure 1 sketches the tethered-particle method (TPM). The main idea is that a macromolecule (for example DNA or some protein that translocates DNA or RNA) is anchored at one end to a surface, while the other end of the molecular complex is attached to an otherwise free microsphere (''bead''). In contrast to classic DNA-stretching experiments, no external stretching force is applied to the bead; instead its motion is passively observed, for example, using single-particle tracking. Thus, the observed motion of the bead serves as a reporter of the underlying, invisible, macromolecular motion. This technique has been used in a variety of settings, e.g., the examination of nanometerscale motions of motors like kinesin [1] or RNA polymerase [2,3], protein synthesis by ribosomes [4], exonuclease translocation on DNA [5,6], protein mediated deformation [7] and loop formation [8] in DNA, DNA hybridization [9], and DNA motion [10,11]. The main goal of this Letter is to show how the proximity of the reporter bead to the surface affects the interpretation of the reported data and can even alter the conformation of the macromolecule of interest. A theoretical understanding of these effects will improve the ability ...
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.
hi@scite.ai
10624 S. Eastern Ave., Ste. A-614
Henderson, NV 89052, USA
Copyright © 2024 scite LLC. All rights reserved.
Made with 💙 for researchers
Part of the Research Solutions Family.