Improved protein structure prediction using potentials from deep learning
https://www.nature.com/articles/s41586-019-1923-7
Deep leaning is used to predict the interaction energy (distance) between two arbitrary residues. The mean force potential generated from the interaction energy was used to describe and optimize the structure.
Learning about learning by many-body systems
The authors explore the efficacy with which the states of a spin glass can be used to train a machine learning algorithm to classify time-dependent driving fields. It transpires that the algorithm is more effective if it has access to the detailed configurations rather than just global properties such as the absorbed power (which is effectively useless in this setting). The authors describe this phenomenon as the spins "learning" the applied drive, but really the states of the spins are just a communication channel from the applied drive to the machine learning algorithm; the detailed states have a higher capacity than the absorbed power, which is not much of a surprise.
Fundamentally, to really learn, it feels to me like the state of the learning system should be updated permanently (or at least long-term) in a way so that it is better at recognising patterns in the future. In the case discussed here, the spins respond to a drive but this information is not retained long-term by the spins and used to improve the response to future drives.
Sequence information transfer using covalent template-directed synthesis https://doi.org/10.1039/c9sc01460h
Template-directed polymer synthesis is the basis for the replication of information stored in D/RNA in living systems. Here the authors look to a synthetic oligomeric system and present a method by which information stored on the oligomer can be copied by synthesis of a new oligomer. The authors describe a general process and implement copying for a 3 unit long template. A trimer template oligomer is prepared with a binary sequence. One type of symbol (0's) is protected (covered up) along the template. The template is presented with the complement for the other symbol type and covalent ester binding occurs. The 0's are then un-protected and their complementary symbol also forms ester bonds with the template. Each site on the template is now occupied by its complementary monomer. The monomers are polymerised together, and then, as the ester bonds are broken, the copy is released from the template. Protection of sites on the template during synthesis is required to generate accuracy in this system, as there is no kinetic difference between the binding of the two different monomer types. A limitation of this procedure, for the purpose of building a synthetic molecular copier, is that it requires external manipulation of the environment of the template at each stage - it is not autonomous.
Landauer's principle at zero temperature
https://arxiv.org/pdf/1911.00910.pdf
The fundamental cost of setting a bit of information to a definite value is \Delta Q= -T \Delta S . This relationship scales with temperature, and so when T=0, the limit is trivial. In this paper the authors derive a tighter bound by considering the relationship between the system and the background, allowing the background to have thermal properties.
Solving the chemical master equation for monomolecular reaction systems analytically: a Doi-Peliti path integral view.
The manuscript considers the problem of obtaining time-dependent solutions of the chemical master equation (CME) by using so-called Doi-Peliti path-integral method. The method formulates the CME as an operator equation for an associated generating function, which is solved via a sequence of integration steps, providing time-dependent probability mass-functions and the underlying moments. The Doi-Peliti path-integral approach has been utilized in the manuscript to recover previously obtained results for so-called monomolecular networks (which consist of reactions whose complexes are single species), and to provide novel result for a more general (non-monomolecular) one-species first-order network, which includes an auto-catalytic reaction.
Deep Learning for RNA Synthetic Biology
Toehold switches are a class of versatile prokaryotic riboregulators inducible by the presence of a fully programmable trans-RNA trigger sequence. These RNA synthetic biology modules hold great promise for a variety of in vitro and in vivo applications. Then, considering the wide applicability and general challenges of toehold switch design, the objective of this work is to develop a deep learning platform to predict toehold switch function as a canonical RNA switch model in synthetic biology. The authors demonstrated the benefits of using deep learning methods (a tenfold improvement) that directly analyse sequence rather than relying on calculations from mechanistic thermodynamic and kinetic models.
Implementing digital computing with DNA-based switching circuits
This work features a strand displacement implementation of a switching circuits formalism first described by Shannon in 1938. In this approach, the different strand displacement reactions implement switches that can be in two different states and can implement functions of different complexity ranging from the construction of Boolean logic gates to the now-classic example of the square root function of a number in a generalized manner without the need for dual-rail logic. As a result, the approach reduces the number of strands required to implement the circuit considerably from the previous implementations.
Will any crap we we put into graphene increase its electrocatalytic effect?
In the present work, Pumera and collaborators took an unorthodox spin on current trends on graphene functionalization research. They demonstrated that, in concordance to previous research in which any kind of addition of dopant heteroatoms would enhance the performance of the material in electrocatalysis applications and the use of different heteroatoms produces a synergistic effect, the use of bird guano as such dopant does indeed improve the electrocatalytical performance of graphene for oxygen reduction reactions as well as hydrogen evolution reactions. Moreover, it is an affordable methodology for the development of metal-free catalysts for fuel cells and electrolysers
Ordered three-dimensional nanomaterials using DNA-prescribed and valence-controlled material voxels
The formation of self-assembled 3D nano-structures is a challenging problem that heavily depends on the molecules used and their interactions. The authors present a generalizable approach to assemble molecules using DNA cubes, pyramids and rhomboids that mimic crystalline unit cells. Each of the DNA unit cells can contain one of the molecules of interest bound by base pairing. Once the DNA unit cells polymerise to form a 3D structure, the contained molecules of interest become arranged in 3D.
Fast and compact DNA logic circuits based on single-stranded gates using strand-displacing polymerase
The authors generated OR and AND logic gates for computing square-root function of 4-bit numbers. The gates are single stranded which can reduce the potential for leakage. Fuel strands anneal to the gates and followed by polymerisation; the input strands can then anneal to this complex and there is displacement of the output strand.
