Powerful, transferable representations for molecules through intelligent task selection in deep multitask networks.

Chemical representations derived from deep learning are emerging as a powerful tool in areas such as drug discovery and materials innovation. Currently, this methodology has three major limitations - the cost of representation generation, risk of inherited bias, and the requirement for large amounts of data. We propose the use of multi-task learning in tandem with transfer learning to address these limitations directly. In order to avoid introducing unknown bias into multi-task learning through the task selection itself, we calculate task similarity through pairwise task affinity, and use this measure to programmatically select tasks. We test this methodology on several real-world data sets to demonstrate its potential for execution in complex and low-data environments. Finally, we utilise the task similarity to further probe the expressiveness of the learned representation through a comparison to a commonly used cheminformatics fingerprint, and show that the deep representation is able to capture more expressive task-based information.

PyLDM - An open source package for lifetime density analysis of time-resolved spectroscopic data.

Ultrafast spectroscopy offers temporal resolution for probing processes in the femto- and picosecond regimes. This has allowed for investigation of energy and charge transfer in numerous photoactive compounds and complexes. However, analysis of the resultant data can be complicated, particularly in more complex biological systems, such as photosystems. Historically, the dual approach of global analysis and target modelling has been used to elucidate kinetic descriptions of the system, and the identity of transient species respectively. With regards to the former, the technique of lifetime density analysis (LDA) offers an appealing alternative. While global analysis approximates the data to the sum of a small number of exponential decays, typically on the order of 2-4, LDA uses a semi-continuous distribution of 100 lifetimes. This allows for the elucidation of lifetime distributions, which may be expected from investigation of complex systems with many chromophores, as opposed to averages. Furthermore, the inherent assumption of linear combinations of decays in global analysis means the technique is unable to describe dynamic motion, a process which is resolvable with LDA. The technique was introduced to the field of photosynthesis over a decade ago by the Holzwarth group. The analysis has been demonstrated to be an important tool to evaluate complex dynamics such as photosynthetic energy transfer, and complements traditional global and target analysis techniques. Although theory has been well described, no open source code has so far been available to perform lifetime density analysis. Therefore, we introduce a python (2.7) based package, PyLDM, to address this need. We furthermore provide a direct comparison of the capabilities of LDA with those of the more familiar global analysis, as well as providing a number of statistical techniques for dealing with the regularization of noisy data.

X-ray Free Electron Laser Determination of Crystal Structures of Dark and Light States of a Reversibly Photoswitching Fluorescent Protein at Room Temperature.

The photochromic fluorescent protein Skylan-NS (Nonlinear Structured illumination variant mEos3.1H62L) is a reversibly photoswitchable fluorescent protein which has an unilluminated/ground state with an anionic and cis chromophore conformation and high fluorescence quantum yield. Photo-conversion with illumination at 515 nm generates a meta-stable intermediate with neutral trans-chromophore structure that has a 4 h lifetime. We present X-ray crystal structures of the cis (on) state at 1.9 Angstrom resolution and the trans (off) state at a limiting resolution of 1.55 Angstrom from serial femtosecond crystallography experiments conducted at SPring-8 Angstrom Compact Free Electron Laser (SACLA) at 7.0 keV and 10.5 keV, and at Linac Coherent Light Source (LCLS) at 9.5 keV. We present a comparison of the data reduction and structure determination statistics for the two facilities which differ in flux, beam characteristics and detector technologies. Furthermore, a comparison of droplet on demand, grease injection and Gas Dynamic Virtual Nozzle (GDVN) injection shows no significant differences in limiting resolution. The photoconversion of the on- to the off-state includes both internal and surface exposed protein structural changes, occurring in regions that lack crystal contacts in the orthorhombic crystal form.

Radical-Triggered Reaction Mechanism of the Green-to-Red Photoconversion of EosFP.

Reaction intermediates in the green-to-red photoconversion of the photochromic fluorescent protein EosFP have been observed using high-intensity continuous blue illumination. An intermediate was identified through light-induced accumulation that continues to convert the green form in subsequent darkness, putatively containing a tyrosyl radical, albeit with anomalously shifted features in both the electronic and FTIR spectra. Lowering the pH to 5.5 significantly delays the decay of this tyrosyl intermediate, which is accompanied by Stark-shifted features in the electronic spectra of reactants and products. Vibrational mode assignments for the high-frequency and fingerprint FTIR spectral regions of the reaction intermediates support a proposed sequence of events where the newly formed Cα═Cß ethylenic bond precedes modifications on the His-62 imidazole ring and confirms a C═O(NH2) product group on Phe-61. We propose a reaction mechanism that involves tyrosyl generation via singlet excited-state-mediated oxidation which subsequently triggers the covalent reactions by oxidation of the green chromophore.

Self-organizing and stochastic behaviors during the regeneration of hair stem cells.

Stem cells cycle through active and quiescent states. Large populations of stem cells in an organ may cycle randomly or in a coordinated manner. Although stem cell cycling within single hair follicles has been studied, less is known about regenerative behavior in a hair follicle population. By combining predictive mathematical modeling with in vivo studies in mice and rabbits, we show that a follicle progresses through cycling stages by continuous integration of inputs from intrinsic follicular and extrinsic environmental signals based on universal patterning principles. Signaling from the WNT/bone morphogenetic protein activator/inhibitor pair is coopted to mediate interactions among follicles in the population. This regenerative strategy is robust and versatile because relative activator/inhibitor strengths can be modulated easily, adapting the organism to different physiological and evolutionary needs.