User feedback on the NHS test & Trace Service during COVID-19: The use of machine learning to analyse free-text data from 37,914 England adults.

Objectives: The UK government's approach to the pandemic relies on a test, trace and isolate strategy, mainly implemented via the digital NHS Test & Trace Service. Feedback on user experience is central to the successful development of public-facing Services. As the situation dynamically changes and data accumulate, interpretation of feedback by humans becomes time-consuming and unreliable. The specific objectives were to 1) evaluate a human-in-the-loop machine learning technique based on structural topic modelling in terms of its Service ability in the analysis of vast volumes of free-text data, 2) generate actionable themes that can be used to increase user satisfaction of the Service. Methods: We evaluated an unsupervised Topic Modelling approach, testing models with 5-40 topics and differing covariates. Two human coders conducted thematic analysis to interpret the topics. We identified a Structural Topic Model with 25 topics and metadata as covariates as the most appropriate for acquiring insights. Results: Results from analysis of feedback by 37,914 users from May 2020 to March 2021 highlighted issues with the Service falling within three major themes: multiple contacts and incompatible contact method and incompatible contact method, confusion around isolation dates and tracing delays, complex and rigid system. Conclusions: Structural Topic Modelling coupled with thematic analysis was found to be an effective technique to rapidly acquire user insights. Topic modelling can be a quick and cost-effective method to provide high quality, actionable insights from free-text feedback to optimize public health Services.

Comparative study of selected parallel tempering methods.

We review several parallel tempering schemes and examine their main ingredients for accuracy and efficiency. The present study covers two selection methods of temperatures and several choices for the exchange of replicas, including a recent novel all-pair exchange method. We compare the resulting schemes and measure specific heat errors and efficiency using the two-dimensional (2D) Ising model. Our tests suggest that an earlier proposal for using numbers of local moves related to the canonical correlation times is one of the key ingredients for increasing efficiency, and protocols using cluster algorithms are found to be very effective. Some of the protocols are also tested for efficiency and ground state production in 3D spin-glass models where we find that a simple nearest-neighbor approach using a local n-fold-way algorithm is the most effective. Finally, we present evidence that the asymptotic limits of the ground state energy for the isotropic case and for an anisotropic case of the 3D spin-glass model are very close and may even coincide.

Critical behavior of the three-dimensional Ising model with anisotropic bond randomness at the ferromagnetic-paramagnetic transition line.

We study the ±J three-dimensional (3D) Ising model with a spatially uniaxial anisotropic bond randomness on the simple cubic lattice. The ±J random exchange is applied on the xy planes, whereas, in the z direction, only a ferromagnetic exchange is used. After sketching the phase diagram and comparing it with the corresponding isotropic case, the system is studied at the ferromagnetic-paramagnetic transition line using parallel tempering and a convenient concentration of antiferromagnetic bonds (p(z)=0;p(xy)=0.176). The numerical data clearly point out a second-order ferromagnetic-paramagnetic phase transition belonging in the same universality class with the 3D random Ising model. The smooth finite-size behavior of the effective exponents, describing the peaks of the logarithmic derivatives of the order parameter, provides an accurate estimate of the critical exponent 1/ν=1.463(3), and a collapse analysis of magnetization data gives an estimate of ß/ν=0.516(7). These results are in agreement with previous papers and, in particular, with those of the isotropic ±J three-dimensional Ising model at the ferromagnetic-paramagnetic transition line, indicating the irrelevance of the introduced anisotropy.

Universality aspects of the d = 3 random-bond Blume-Capel model.

The effects of bond randomness on the universality aspects of the simple cubic lattice ferromagnetic Blume-Capel model are discussed. The system is studied numerically in both its first- and second-order phase transition regimes by a comprehensive finite-size scaling analysis. We find that our data for the second-order phase transition, emerging under random bonds from the second-order regime of the pure model, are compatible with the universality class of the 3d random Ising model. Furthermore, we find evidence that the second-order transition emerging under bond randomness from the first-order regime of the pure model belongs to a new and distinctive universality class. The first finding reinforces the scenario of a single universality class for the 3d Ising model with the three well-known types of quenched uncorrelated disorder (bond randomness, site and bond dilution). The second amounts to a strong violation of the universality principle of critical phenomena. For this case of the ex-first-order 3d Blume-Capel model, we find sharp differences from the critical behaviors, emerging under randomness, in the cases of the ex-first-order transitions of the corresponding weak and strong first-order transitions in the 3d three-state and four-state Potts models.

Universality of the Ising and the S=1 model on Archimedean lattices: a Monte Carlo determination.

The Ising models S=1/2 and S=1 are studied by efficient Monte Carlo schemes on the (3,4,6,4) and the (3,3,3,3,6) Archimedean lattices. The algorithms used, a hybrid Metropolis-Wolff algorithm and a parallel tempering protocol, are briefly described and compared with the simple Metropolis algorithm. Accurate Monte Carlo data are produced at the exact critical temperatures of the Ising model for these lattices. Their finite-size analysis provide, with high accuracy, all critical exponents which, as expected, are the same with the well-known 2D Ising model exact values. A detailed finite-size scaling analysis of our Monte Carlo data for the S=1 model on the same lattices provides very clear evidence that this model obeys, also very well, the 2D Ising model critical exponents. As a result, we find that recent Monte Carlo simulations and attempts to define effective dimensionality for the S=1 model on these lattices are misleading. Accurate estimates are obtained for the critical amplitudes of the logarithmic expansions of the specific heat for both models on the two Archimedean lattices.

Multicritical points and crossover mediating the strong violation of universality: Wang-Landau determinations in the random-bond d=2 Blume-Capel model.

The effects of bond randomness on the phase diagram and critical behavior of the square lattice ferromagnetic Blume-Capel model are discussed. The system is studied in both the pure and disordered versions by the same efficient two-stage Wang-Landau method for many values of the crystal field, restricted here in the second-order phase-transition regime of the pure model. For the random-bond version several disorder strengths are considered. We present phase diagram points of both pure and random versions and for a particular disorder strength we locate the emergence of the enhancement of ferromagnetic order observed in an earlier study in the ex-first-order regime. The critical properties of the pure model are contrasted and compared to those of the random model. Accepting, for the weak random version, the assumption of the double-logarithmic scenario for the specific heat we attempt to estimate the range of universality between the pure and random-bond models. The behavior of the strong disorder regime is also discussed and a rather complex and yet not fully understood behavior is observed. It is pointed out that this complexity is related to the ground-state structure of the random-bond version.

Expression of protein tyrosine phosphatase-like molecule ICA512/IA-2 induces growth arrest in yeast cells and transfected mammalian cell lines.

The ICA512/IA-2 molecule, a protein with similarity to receptor-type protein tyrosine phosphatases, was discovered during studies to identify autoantigens in Type 1 diabetes. The biological function of ICA512/IA-2 is unknown. We describe striking effects of ICA512/IA-2 on viability and growth of both yeast cells and cultured mammalian cells. In transformed yeast Saccharomyces cerevisiae cells, expression of ICA512/IA-2 induced growth retardation as judged by measurements of optical density and counts of colony-forming units. In contrast, expression of the intracellular domain (amino acids 600-979) of ICA512/IA-2 in yeast or mammalian cells had no such effects. In investigations on apoptosis, expression of ICA512/IA-2 in yeast cells caused loss of plasma membrane asymmetry, but not release of cytochrome c from mitochondria which did occur in a control system after expression of the pro-apoptotic molecule Bax. Possible interactions between ICA512/IA-2 and components of the cytoskeleton were not supported by studies on staining of fixed yeast cells with phalloidin-Texas Red. With transfected mammalian cell lines COS-7 and NIH3T3, expression of ICA512/IA-2 likewise induced growth arrest, with some of the morphological features of apoptosis. Thus obligatory expression of ICA512/IA-2 in eukaryotic cells causes disruption of cellular activities, with growth arrest in yeast and nuclear pycnosis/fragmentation in mammalian cells. A possible explanation is that growth inhibition reflects a part of the presently unknown function of ICA512/IA-2.

Comparative expression and purification of human glutamic acid decarboxylase from Saccharomyces cerevisiae and Pichia pastoris.

The yeast cell factory is a potentially useful source of proteins in general. They include glutamic acid decarboxylase (GAD), which is one of the major autoantigens for Type 1 diabetes. We have created a hybrid form of GAD consisting of amino acids 1-101 of the human GAD67 protein fused to amino acids 96-585 of the human GAD65 protein, and have modified this to include a C-terminal hexa-Histidine (H6) tag sequence. This hybrid GAD67/65-H6 was expressed in two yeast hosts: constitutively under the control of the plasmid phosphoglycerate kinase promoter (PGK1) in Saccharomyces cerevisiae, and inducibly under the control of the chromosomal alcohol oxidase promoter (AOX1) in Pichia pastoris. Enzymatically active hybrid GAD was prepared from yeast lysates by purification either on an affinity column based on the GAD-1 monoclonal antibody, or by metal-affinity chromatography. The purified GAD67/65-H6 was radiolabelled with iodine-125 and tested with Type 1 diabetes sera in a radioimmunoprecipitation assay, and results were compared with those using untagged GAD67/65 and those using porcine brain GAD. The results of enzymatic and immunological assays show hybrid GAD67/65 is isolated at high specific activity and moderate yield, and the addition of the H6 tag sequences or the choice of yeast strain did not appreciably affect enzyme activity, percentage recovery of GAD, protein purification, or the utility in diagnosis of diabetes in terms of specificity and sensitivity to the various sera.

Molecular genetic analysis of the central hydrophobic domain of subunit 8 of yeast mitochondrial ATP synthase.

Subunit 8 (Y8) of yeast mitochondrial ATP synthase (mtATPase) is a hydrophobic component of the membrane Fo sector. Encoded by the mitochondrial aap1 gene, Y8 is a 48-amino-acid polypeptide having a central hydrophobic domain (CHD) spanning 19 residues. Site-directed mutagenesis was carried out on a nuclear code-equivalent gene encoding Y8, to introduce either adjacent charged amino acids (positive or negative) or proline residues into the CHD, or to alter the length of this domain by deletion or insertion of additional non-polar residues. We report a functional resilience of Y8 in tolerating the introduction of charged residues implanted within the CHD. Thus, expression of variants having adjacent positively charged amino acids (arginines) in Y8-deficient cells restored growth on the non-fermentable substrate ethanol, though in some cases this was impaired compared to that conferred by the parent Y8 construct. Introduction of adjacent negative charges (aspartate residues) was less well tolerated, but in all cases a measurable rate of cell growth on ethanol was retained. These results underscore the interpretation that it is not necessary for Y8 to maintain a transmembrane stem in its role as an integral component of functional mtATPase. Further, the impaired growth properties of cells expressing variants of Y8 having changes designed to perturb the structure (proline substitutions) and length (insertions or deletions) of the CHD lead us to conclude that the overall shape and dimensions of Y8 are important for its function in mtATPase.

Non-functional variants of yeast mitochondrial ATP synthase subunit 8 that assemble into the complex.

Subunit 8 (Y8) is a component of the proton channel of yeast (Saccharomyces cerevisiae) mitochondrial ATP synthase (mtATPase), whose function in the complex remains to be precisely defined. Y8 variants truncated at residue 46 (Lys47-->STP), or in which each of three conserved C-terminal amino acid residues (Arg37, Arg42 and Lys47) were substituted with isoleucine, have defects in assembly and function. The additional positive charge substitution (Gln29-->Lys) was introduced into each of the variants to determine whether functional compensation for these defects could be achieved. In the case of the (Lys47-->STP) variant, the additional positive charge restored the ability of cells to grow on non-fermentable substrate. By contrast, for the (Lys47-->Ile) variant the additional positive charge did not confer any improvement in cellular growth rate compared to that of cells expressing the Lys47-->Ile substitution alone. For the (Arg42-->Ile) and (Arg37-->Ile) variants, the presence of the Gln29-->Lys substitution failed to restore growth of host cells lacking endogenous subunit 8 on non-fermentable substrate. However, use of an in vitro assembly assay revealed that, unlike their respective parents (Arg42-->Ile or Arg37-->Ile), the (Gln29-->Lys Arg42-->Ile) and (Gln29-->Lys Arg37-->Ile) variants assemble into mtATPase. Thus we conclude that Arg42 and Arg37 have a role in mtATPase function, in addition to being required for assembly of Y8 into mtATPase.

Relationship of subunit 8 of yeast ATP synthase and the inner mitochondrial membrane. Subunit 8 variants containing multiple lysine residues in the central hydrophobic domain retain function.

A molecular genetic approach has been used to test the proposition that the central hydrophobic domain of yeast mitochondrial ATP synthase subunit 8 represents a transmembrane stem in contact with the lipid bilayer. The rationale for this approach is the general inability of membrane bilayers to accomodate unshielded charged residues of polypeptide chains. Non-polar residues at several positions within the central hydrophobic domain of subunit 8 were replaced with the positively charged amino acid lysine. This was done in an attempt to disrupt subunit 8 function, and thereby determine the boundaries of the putative transmembrane stem. Each subunit 8 variant was allotopically expressed in vivo as a mitochondrial import precursor encoded by a nuclear gene. It was found that all variants, which included proteins carrying two lysines at various positions in the hydrophobic domain, exhibited the ability to restore growth of subunit-8-deficient cells on the non-fermentable substrate ethanol. This indicated that the function of none of these subunit 8 variants was severely compromised. There was also no detectable change in the proteolipid characteristics of subunit 8, as defined by the chloroform/methanol solubility properties of variant proteins extracted from membranes following import into isolated mitochondria. These data suggest that subunit 8 is located in a hydrophobic niche in the mitochondrial ATP synthase, probably in contact with other protein subunits of the complex. We conclude that the function of subunit 8 does not necessarily require it to be integrated within the inner mitochondrial membrane, in contact with the lipid bilayer. Our findings also suggest that hydropathy plots, indicating hydrophobic domains within polypeptides, cannot reliably be interpreted as transmembrane helices in the absence of independent evidence.

Each of three positively-charged amino acids in the C-terminal region of yeast mitochondrial ATP synthase subunit 8 is required for assembly.

Each of three conserved positively-charged residues in the C-terminal region of subunit 8 of yeast (Saccharomyces cerevisiae) mitochondrial ATP synthase was replaced with isoleucine. The assembly and functional properties of the resulting variants (substituted at Arg-37, Arg-42 and Lys-47) were examined using in-vitro systems to assay import into isolated mitochondria and to monitor assembly into ATP synthase, as well as an in-vivo rescue system using host yeast cells lacking endogenous subunit 8. Each such variant was found to be impaired in assembly in vitro, after import in the form of a chimaeric protein bearing a leader sequence with mitochondrial targeting function. Import precursors bearing a duplicated-leader sequence, engendering enhanced delivery to mitochondria of the passenger variant subunit-8 proteins, enabled assembly of the (Lys-47-->Ile) variant to be detected in vitro but not that of (Arg-37-->Ile) or (Arg-42-->Ile) variants. The respiratory growth of subunit 8-deficient host cells could be rescued with the (Lys-47-->Ile) variant expressed allotopically in the nucleus. Such rescued cells were found to have an enhanced growth rate (comparable to that produced by non-mutagenized parental subunit 8) when delivered to mitochondria with the duplicated-leader sequence, as compared to the single-leader sequence. This confirms that the impediment in the (Lys-47-->Ile) variant lies in the efficiency of its assembly, rather than a functional defect, as such, arising from the loss of that positive charge. In contrast, host cells were unable to be rescued by the (Arg-37-->Ile) and (Arg-42-->Ile) variants, even when they were endowed with the duplicated leader sequence. It is concluded that the positively-charged C-terminal domain of subunit 8, common to fungal and mammalian homologues of this protein, plays a key role in its assembly into mitochondrial ATP synthase.

Structure/function analysis of yeast mitochondrial ATP synthase subunit 8.

Subunit 8 of yeast mitochondrial ATP synthase is a small hydrophobic component of the membrane-associated F0 sector. Structure/function relations in subunit 8 were studied by focusing on three structural domains: a highly conserved NH2-terminal region, a central hydrophobic region (previously suggested to be a transmembrane stem), and a COOH-terminal region bearing a conserved array of three positively charged residues. A combined approach was used, which encompasses site-directed mutagenesis, in vitro import and assembly tests, and an in vivo allotopic expression system (using host cells unable to synthesise subunit 8 in mitochondria). The results indicate that the NH2-terminal region of subunit 8 is involved functionally in the F0 sector. As the central hydrophobic region can functionally tolerate the introduction of multiple, positively charged residues (which abolishes the proteolipid solubility characteristics of the entire subunit), the role of this hydrophobic region as a transmembrane stem is brought into question. Each of the three positively charged residues toward the COOH-terminus of subunit 8 is required for the efficient assembly of this subunit into the F0 sector. Removal of the more proximal charged residues Arg37 or Arg42 has a more severe impact on subunit 8 assembly than does removal of the most distal residue Lys47 in terms of both in vitro import and assembly as well as the ability of the subunit 8 variant to function in mitochondrial ATP synthase in vivo.