Visual acuity prediction on real-life patient data using a machine learning based multistage system.

In ophthalmology, intravitreal operative medication therapy (IVOM) is a widespread treatment for diseases related to the age-related macular degeneration (AMD), the diabetic macular edema, as well as the retinal vein occlusion. However, in real-world settings, patients often suffer from loss of vision on time scales of years despite therapy, whereas the prediction of the visual acuity (VA) and the earliest possible detection of deterioration under real-life conditions is challenging due to heterogeneous and incomplete data. In this contribution, we present a workflow for the development of a research-compatible data corpus fusing different IT systems of the department of ophthalmology of a German maximum care hospital. The extensive data corpus allows predictive statements of the expected progression of a patient and his or her VA in each of the three diseases. For the disease AMD, we found out a significant deterioration of the visual acuity over time. Within our proposed multistage system, we subsequently classify the VA progression into the three groups of therapy "winners", "stabilizers", and "losers" (WSL classification scheme). Our OCT biomarker classification using an ensemble of deep neural networks results in a classification accuracy (F1-score) of over 98%, enabling us to complete incomplete OCT documentations while allowing us to exploit them for a more precise VA modelling process. Our VA prediction requires at least four VA examinations and optionally OCT biomarkers from the same time period to predict the VA progression within a forecasted time frame, whereas our prediction is currently restricted to IVOM/no therapy. We achieve a final prediction accuracy of 69% in macro average F1-score, while being in the same range as the ophthalmologists with 57.8 and 50 ± 10.7 % F1-score.

Improving OCT Image Segmentation of Retinal Layers by Utilizing a Machine Learning Based Multistage System of Stacked Multiscale Encoders and Decoders.

Optical coherence tomography (OCT)-based retinal imagery is often utilized to determine influential factors in patient progression and treatment, for which the retinal layers of the human eye are investigated to assess a patient's health status and eyesight. In this contribution, we propose a machine learning (ML)-based multistage system of stacked multiscale encoders and decoders for the image segmentation of OCT imagery of the retinal layers to enable the following evaluation regarding the physiological and pathological states. Our proposed system's results highlight its benefits compared to currently investigated approaches by combining commonly deployed methods from deep learning (DL) while utilizing deep neural networks (DNN). We conclude that by stacking multiple multiscale encoders and decoders, improved scores for the image segmentation task can be achieved. Our retinal-layer-based segmentation results in a final segmentation performance of up to 82.25±0.74% for the Sørensen-Dice coefficient, outperforming the current best single-stage model by 1.55% with a score of 80.70±0.20%, given the evaluated peripapillary OCT data set. Additionally, we provide results on the data sets Duke SD-OCT, Heidelberg, and UMN to illustrate our model's performance on especially noisy data sets.

Metal ions and sugar puckering balance single-molecule kinetic heterogeneity in RNA and DNA tertiary contacts.

The fidelity of group II intron self-splicing and retrohoming relies on long-range tertiary interactions between the intron and its flanking exons. By single-molecule FRET, we explore the binding kinetics of the most important, structurally conserved contact, the exon and intron binding site 1 (EBS1/IBS1). A comparison of RNA-RNA and RNA-DNA hybrid contacts identifies transient metal ion binding as a major source of kinetic heterogeneity which typically appears in the form of degenerate FRET states. Molecular dynamics simulations suggest a structural link between heterogeneity and the sugar conformation at the exon-intron binding interface. While Mg2+ ions lock the exon in place and give rise to long dwell times in the exon bound FRET state, sugar puckering alleviates this structural rigidity and likely promotes exon release. The interplay of sugar puckering and metal ion coordination may be an important mechanism to balance binding affinities of RNA and DNA interactions in general.

Reliable State Identification and State Transition Detection in Fluorescence Intensity-Based Single-Molecule Förster Resonance Energy-Transfer Data.

Single-molecule Förster resonance energy transfer (smFRET) is a powerful technique to probe biomolecular structure and dynamics. A popular implementation of smFRET consists of recording fluorescence intensity time traces of surface-immobilized, chromophore-tagged molecules. This approach generates large and complex data sets, the analysis of which is to date not standardized. Here, we address a key challenge in smFRET data analysis: the generation of thermodynamic and kinetic models that describe with statistical rigor the behavior of FRET trajectories recorded from surface-tethered biomolecules in terms of the number of FRET states, the corresponding mean FRET values, and the kinetic rates at which they interconvert. For this purpose, we first perform Monte Carlo simulations to generate smFRET trajectories, in which a relevant space of experimental parameters is explored. Then, we provide an account on current strategies to achieve such model selection, as well as a quantitative assessment of their performances. Specifically, we evaluate the performance of each algorithm (change-point analysis, STaSI, HaMMy, vbFRET, and ebFRET) with respect to accuracy, reproducibility, and computing time, which yields a range of algorithm-specific referential benchmarks for various data qualities. Data simulation and analysis were performed with our MATLAB-based multifunctional analysis software for handling smFRET data (MASH-FRET).

Simulations of camera-based single-molecule fluorescence experiments.

Single-molecule microscopy has become a widely used technique in (bio)physics and (bio)chemistry. A popular implementation is single-molecule Förster Resonance Energy Transfer (smFRET), for which total internal reflection fluorescence microscopy is frequently combined with camera-based detection of surface-immobilized molecules. Camera-based smFRET experiments generate large and complex datasets and several methods for video processing and analysis have been reported. As these algorithms often address similar aspects in video analysis, there is a growing need for standardized comparison. Here, we present a Matlab-based software (MASH-FRET) that allows for the simulation of camera-based smFRET videos, yielding standardized data sets suitable for benchmarking video processing algorithms. The software permits to vary parameters that are relevant in cameras-based smFRET, such as video quality, and the properties of the system under study. Experimental noise is modeled taking into account photon statistics and camera noise. Finally, we survey how video test sets should be designed to evaluate currently available data analysis strategies in camera-based sm fluorescence experiments. We complement our study by pre-optimizing and evaluating spot detection algorithms using our simulated video test sets.

Explicit analytic equations for multimolecular thermal melting curves.

The analysis of thermal melting curves requires the knowledge of equations for the temperature dependence of the relative fraction of folded and unfolded components. To implement these equations as standard tools for curve fitting, they should be as explicit as possible. From the van't Hoff formalism it is known that the equilibrium constant and hence the folded fraction is a function of the absolute temperature, the van't Hoff transition enthalpy, and the melting temperature. The work presented here is devoted to the mathematically self-contained derivation and the listing of explicit equations for the folded fraction as a function of the thermodynamic parameters in the case of arbitrary molecularities. Part of the results are known, others are new. It is in particular shown for the first time that the folded fraction is the composition of a universal function which depends solely on the molecularity and a dimensionless function which is governed by the concrete thermodynamic regime but is independent of the molecularity. The results will prove useful for extracting the thermodynamic parameters from experimental data on the basis of regression analysis. As supporting information, open-source Matlab scripts for the computer implementation of the equations are provided.

Cation-induced kinetic heterogeneity of the intron-exon recognition in single group II introns.

RNA is commonly believed to undergo a number of sequential folding steps before reaching its functional fold, i.e., the global minimum in the free energy landscape. However, there is accumulating evidence that several functional conformations are often in coexistence, corresponding to multiple (local) minima in the folding landscape. Here we use the 5'-exon-intron recognition duplex of a self-splicing ribozyme as a model system to study the influence of Mg(2+) and Ca(2+) on RNA tertiary structure formation. Bulk and single-molecule spectroscopy reveal that near-physiological M(2+) concentrations strongly promote interstrand association. Moreover, the presence of M(2+) leads to pronounced kinetic heterogeneity, suggesting the coexistence of multiple docked and undocked RNA conformations. Heterogeneity is found to decrease at saturating M(2+) concentrations. Using NMR, we locate specific Mg(2+) binding pockets and quantify their affinity toward Mg(2+). Mg(2+) pulse experiments show that M(2+) exchange occurs on the timescale of seconds. This unprecedented combination of NMR and single-molecule Förster resonance energy transfer demonstrates for the first time to our knowledge that a rugged free energy landscape coincides with incomplete occupation of specific M(2+) binding sites at near-physiological M(2+) concentrations. Unconventional kinetics in nucleic acid folding frequently encountered in single-molecule experiments are therefore likely to originate from a spectrum of conformations that differ in the occupation of M(2+) binding sites.

Kinetic subpopulations detected by single-molecule spectroscopy: fundamental property of functional nucleic acids or experimental artefact?

Single-molecule spectroscopy allows the direct observation of conformational dynamics in individual biomolecules. Here, we describe how single-molecule Förster resonance energy transfer (smFRET) reveals heterogeneous kinetics in the EBS1*/IBS1* interaction, two RNA sequences that play an important role in group II intron mediated self-cleavage. Further examples of dynamic heterogeneity in functional nucleic acids are provided and the possible origins of this phenomenon are discussed.

BOBA FRET: bootstrap-based analysis of single-molecule FRET data.

Time-binned single-molecule Förster resonance energy transfer (smFRET) experiments with surface-tethered nucleic acids or proteins permit to follow folding and catalysis of single molecules in real-time. Due to the intrinsically low signal-to-noise ratio (SNR) in smFRET time traces, research over the past years has focused on the development of new methods to extract discrete states (conformations) from noisy data. However, limited observation time typically leads to pronounced cross-sample variability, i.e., single molecules display differences in the relative population of states and the corresponding conversion rates. Quantification of cross-sample variability is necessary to perform statistical testing in order to assess whether changes observed in response to an experimental parameter (metal ion concentration, the presence of a ligand, etc.) are significant. However, such hypothesis testing has been disregarded to date, precluding robust biological interpretation. Here, we address this problem by a bootstrap-based approach to estimate the experimental variability. Simulated time traces are presented to assess the robustness of the algorithm in conjunction with approaches commonly used in thermodynamic and kinetic analysis of time-binned smFRET data. Furthermore, a pair of functionally important sequences derived from the self-cleaving group II intron Sc.ai5γ (d3'EBS1/IBS1) is used as a model system. Through statistical hypothesis testing, divalent metal ions are shown to have a statistically significant effect on both thermodynamic and kinetic aspects of their interaction. The Matlab source code used for analysis (bootstrap-based analysis of smFRET data, BOBA FRET), as well as a graphical user interface, is available via http://www.aci.uzh.ch/rna/.

Efficient simultaneous fluorescence orientation, spectrum, and lifetime detection for single molecule dynamics.

We report on the simultaneous detection of the fluorescence lifetime, spectrum, and three-dimensional dipole orientation determination of single perylene diimide molecules deposited on a silica surface as a model system for studying fluorophore internal and orientational dynamics. We employ a multi-parameter detection scheme to demonstrate how jumps in the orientation of the molecule can be disentangled from spectral jumps, both leading to changes of the detected total fluorescence intensity. The fluorescence lifetime determined simultaneously from the same photons is also sensitive to the orientation of the dipole with respect to the interface between media with different refractive indices. The correlated changes of the lifetime and orientation we observe are in good agreement with theory.

Formation principles and ligand dynamics of nanoassemblies of CdSe quantum dots and functionalised dye molecules.

Functional dye molecules, such as porphyrins, attached to CdSe quantum dots (QDs) through anchoring meso-pyridyl substituents, form quasi-stable nanoassemblies. This fact results in photoluminescence (PL) quenching of the QDs both due to Förster resonance energy transfer (FRET) and the formation of non-radiative surface states under conditions of quantum confinement (non-FRET). The formation process is in competition with the ligand dynamics. At least two timescales are found for the formation of the assemblies: 1) one faster than 60 s attributed to saturation of empty attachment sites and 2) one slower than 600 s, which is attributed to a reorganisation of the tri-n-octylphosphine oxide (TOPO) ligand shell. Finally, this process results in almost complete exchange of the TOPO shell by porphyrin dye molecules. Following a Stern-Volmer analysis, we established a microscopic description of PL quenching and assembly formation. Based on this formalism, we determined the equilibrium constant for assembly formation between QDs and the pyridyl-functionalised dye molecules to be K ≈ 10(5) - 10(7) M(-1), which is several orders of magnitude larger than that of the TOPO ligands. Our results give additional insights into the non-FRET PL quenching processes involved and show that the QD surface is inhomogeneous with respect to the involved attachment and detachment processes. In comparison with other methods, such as NMR spectroscopy, the advantage of our approach is that ligand dynamics can be investigated at extremely low ratios of dye molecules to QDs.

Spectral diffusion of single molecules in a hierarchical energy landscape.

Spectral diffusion as a result of both the transitions between different molecular conformers and the ''molecular softness'' of quasi-free perylene diimides on a SiO(2) surface is investigated by means of single-molecule spectroscopy, which reveals the time dependence of both the fluorescence spectra and the three-dimensional orientation. Spectral wavelengths of all single emitters cover a wide energy range of about 0.27 eV, which is due to different types of conformers with large differences in optical transition energy. Time-dependent spectral trajectories of single emitters within this wavelength manifold are evaluated with a model transcribed from the analysis of spatial diffusion. Spectral diffusion processes are closely correlated with fluorescence emission and excitation power. The overall analysis of spectral diffusion reveals, similar to proteins, a hierarchy of energy barriers in a broad energy landscape.

FRET and ligand related NON-FRET processes in single quantum dot-perylene bisimide assemblies.

Nanoassemblies are formed via self-assembly of ZnS capped CdSe quantum dots (QD) and perylene bisimide dyes (PBI). Upon assembly formation with functionalized dye molecules the QD photoluminescence (PL) is quenched. Quenching has been assigned partly to FRET (fluorescence resonance energy transfer) and NON-FRET processes. By means of time resolved single particle spectroscopy of immobilized QD-dye assemblies, it is demonstrated that NON-FRET processes are due to new non-radiative decay channels caused by the assembly formation process itself. Immobilized (single) assemblies exhibit the same processes as ensembles of assemblies in toluene solution. Only one dye molecule on a QD quenches the PL up to 50%, which is much stronger than is expected when replacing a volume related number of ligands. NON-FRET processes are distinct from photoinduced charge and/or energy transfer. A combination of a Stern-Volmer and FRET analysis of ensemble experiments supports the investigation of the dynamics of assembly formation at extremely low concentration ratios of PBI to QD. This allows us to distinguish between the effects of PBI and ligands on PL quenching on a single molecule level which is not possible in conventional ligand dynamic experiments.

Identification of different donor-acceptor structures via Förster Resonance Energy Transfer (FRET) in quantum-dot-perylene bisimide assemblies.

Nanoassemblies are formed via self-assembly of ZnS capped CdSe quantum dots (QD) and perylene bisimide (PBI) dyes. Upon assembly formation the QD photoluminescence is quenched, as can be detected both via single particle detection and ensemble experiments in solution. Quenching has been assigned to FRET and NON-FRET processes. Analysis of FRET allows for a distinction between different geometries of the QD dye assemblies. Time-resolved single molecule spectroscopy reveals intrinsic fluctuations of the PBI fluorescence lifetime and spectrum, caused by rearrangement of the phenoxy side groups. The distribution of such molecular conformations and their changed dynamics upon assembly formation are discussed in the scope of FRET efficiency and surface ligand density.