Personalized Telehealth: Redesigning Complex Care Delivery for the 65+ During the COVID Pandemic: a Survey of Patients, Caregivers, and Health-care Providers.

Background: In the context of the ongoing COVID-19 pandemic, rapid transitions have been made towards telehealth. Optimal use of telehealth in elderly patients remains poorly understood and adaptation challenges persist. Our study aimed at identifying perceptions, barriers, and possible facilitators to telehealth use amongst elderly patients with comorbidities, their caregivers, and health-care providers (HCPs). Methods: Health-care providers, patients 65 years and older with multiple comorbidities, and caregivers were recruited from outpatient clinics and invited to complete an electronic self-administered or telephone-administered survey on their perceptions of telehealth and of barriers to its implementation. Results: A total of 39 health-care providers, 40 patients, and 22 caregivers responded to the survey. Most patients (90%), caregivers (82%), and HCPs (97%) had experienced telephone visits, but few were conducted via videoconference platforms. Patients and caregivers showed interest in pursuing some future telehealth visits (68%, 86%, respectively), but felt they lacked access to technology and skills (n=8, 20%), and some felt that telehealth visits may be inferior to in-person visits (n=9, 23%). HCPs showed interest in incorporating telehealth visits into practice (n=32, 82%), but identified challenges in lack of administrative support (n=37), lack of HCP (n=28) and patient (n=37) technological skills, and limited infrastructure (n=37)/internet access (n=33). Conclusions: Older patients, caregivers, and HCPs show interest in pursuing future telehealth visits but elucidate similar barriers. Facilitating access to technology, as well as to administrative and technology support guides, could promote high quality and equal access to virtual care for the older adult.

ProteInfer, deep neural networks for protein functional inference.

Predicting the function of a protein from its amino acid sequence is a long-standing challenge in bioinformatics. Traditional approaches use sequence alignment to compare a query sequence either to thousands of models of protein families or to large databases of individual protein sequences. Here we introduce ProteInfer, which instead employs deep convolutional neural networks to directly predict a variety of protein functions - Enzyme Commission (EC) numbers and Gene Ontology (GO) terms - directly from an unaligned amino acid sequence. This approach provides precise predictions which complement alignment-based methods, and the computational efficiency of a single neural network permits novel and lightweight software interfaces, which we demonstrate with an in-browser graphical interface for protein function prediction in which all computation is performed on the user's personal computer with no data uploaded to remote servers. Moreover, these models place full-length amino acid sequences into a generalised functional space, facilitating downstream analysis and interpretation. To read the interactive version of this paper, please visit https://google-research.github.io/proteinfer/.

Using deep learning to annotate the protein universe.

Understanding the relationship between amino acid sequence and protein function is a long-standing challenge with far-reaching scientific and translational implications. State-of-the-art alignment-based techniques cannot predict function for one-third of microbial protein sequences, hampering our ability to exploit data from diverse organisms. Here, we train deep learning models to accurately predict functional annotations for unaligned amino acid sequences across rigorous benchmark assessments built from the 17,929 families of the protein families database Pfam. The models infer known patterns of evolutionary substitutions and learn representations that accurately cluster sequences from unseen families. Combining deep models with existing methods significantly improves remote homology detection, suggesting that the deep models learn complementary information. This approach extends the coverage of Pfam by >9.5%, exceeding additions made over the last decade, and predicts function for 360 human reference proteome proteins with no previous Pfam annotation. These results suggest that deep learning models will be a core component of future protein annotation tools.

Generalizing Few-Shot Classification of Whole-Genome Doubling Across Cancer Types.

The study and treatment of cancer is traditionally specialized to the cancer's site of origin. However, certain phenotypes are shared across cancer types and have important implications for clinical care. To date, automating the identification of these characteristics from routine clinical data - irrespective of the type of cancer - is impaired by tissue-specific variability and limited labeled data. Whole-genome doubling is one such phenotype; whole-genome doubling events occur in nearly every type of cancer and have significant prognostic implications. Using digitized histopathology slide images of primary tumor biopsies, we train a deep neural network end-to-end to accurately generalize few-shot classification of whole-genome doubling across 17 cancer types. By taking a meta-learning approach, cancer types are treated as separate but jointly-learned tasks. This approach outperforms a traditional neural network classifier and quickly generalizes to both held-out cancer types and batch effects. These results demonstrate the unrealized potential for meta-learning to not only account for between-cancer type variability but also remedy technical variability, enabling real-time identification of cancer phenotypes that are too often costly and inefficient to obtain.

Growth Resilience of Subarctic Rhodoliths (Lithothamnion glaciale, Rhodophyta) to Chronic Low Sea Temperature and irradiance.

Rhodolith beds are pervasive marine biological systems in the subarctic North Atlantic. Limited knowledge about effects of temperature and irradiance on rhodolith growth limits the ability to anticipate the response of rhodolith beds to this ocean's chronic low, yet changing sea temperature and irradiance regimes. We carried out a 149-d laboratory experiment with Newfoundland Lithothamnion glaciale rhodoliths to test the predictions that growth (i) is inhibited at temperatures of ~0.5°C and (ii) resumes as temperature increases above 0.5°C, albeit at a higher rate under high than low irradiances. Rhodoliths were grown in experimental tanks at near-zero (~0.7°C) seawater temperatures during the first 85 d and at temperatures increasing naturally to ~6°C for the remaining 64 d. Rhodoliths in those tanks were exposed to either low (0.02 mol photons·m-2 ·d-1 ) or high (0.78 mol photons·m-2 ·d-1 ) irradiances during the entire experiment. Rhodoliths grew at a linear rate of ~281 µm·year-1 (0.77 µm·d-1 ) throughout the experiment under both irradiance treatments despite daily seawater temperature variation of up to 3°C. Near-zero temperatures of ~0.5 to 1.0°C did not inhibit rhodolith growth. Model selection showed that PAR-day (a cumulative irradiance index) was a better predictor of growth variation than Degree-day (a cumulative thermal index). Our findings extend to ~0.5°C the lower limit of the known temperature range (~1 to at least 16°C) over which growth in L. glaciale rhodoliths remains unaffected, while suggesting that the growth-irradiance relationship in low-light environments at temperatures below 6°C is less irradiance-driven than recently proposed.

Critiquing Protein Family Classification Models Using Sufficient Input Subsets.

In many application domains, neural networks are highly accurate and have been deployed at large scale. However, users often do not have good tools for understanding how these models arrive at their predictions. This has hindered adoption in fields such as the life and medical sciences, where researchers require that models base their decisions on underlying biological phenomena rather than peculiarities of the dataset. We propose a set of methods for critiquing deep learning models and demonstrate their application for protein family classification, a task for which high-accuracy models have considerable potential impact. Our methods extend the Sufficient Input Subsets (SIS) technique, which we use to identify subsets of features in each protein sequence that are alone sufficient for classification. Our suite of tools analyzes these subsets to shed light on the decision-making criteria employed by models trained on this task. These tools show that while deep models may perform classification for biologically relevant reasons, their behavior varies considerably across the choice of network architecture and parameter initialization. While the techniques that we develop are specific to the protein sequence classification task, the approach taken generalizes to a broad set of scientific contexts in which model interpretability is essential.

Rapid Prediction of Electron-Ionization Mass Spectrometry Using Neural Networks.

When confronted with a substance of unknown identity, researchers often perform mass spectrometry on the sample and compare the observed spectrum to a library of previously collected spectra to identify the molecule. While popular, this approach will fail to identify molecules that are not in the existing library. In response, we propose to improve the library's coverage by augmenting it with synthetic spectra that are predicted from candidate molecules using machine learning. We contribute a lightweight neural network model that quickly predicts mass spectra for small molecules, averaging 5 ms per molecule with a recall-at-10 accuracy of 91.8%. Achieving high-accuracy predictions requires a novel neural network architecture that is designed to capture typical fragmentation patterns from electron ionization. We analyze the effects of our modeling innovations on library matching performance and compare our models to prior machine-learning-based work on spectrum prediction.

Design, Synthesis, and Preclinical Characterization of Selective Factor D Inhibitors Targeting the Alternative Complement Pathway.

Complement factor D (FD), a highly specific S1 serine protease, plays a central role in the amplification of the alternative complement pathway (AP) of the innate immune system. Dysregulation of AP activity predisposes individuals to diverse disorders such as age-related macular degeneration, atypical hemolytic uremic syndrome, membranoproliferative glomerulonephritis type II, and paroxysmal nocturnal hemoglobinuria. Previously, we have reported the screening efforts and identification of reversible benzylamine-based FD inhibitors (1 and 2) binding to the open active conformation of FD. In continuation of our drug discovery program, we designed compounds applying structure-based approaches to improve interactions with FD and gain selectivity against S1 serine proteases. We report herein the design, synthesis, and medicinal chemistry optimization of the benzylamine series culminating in the discovery of 12, an orally bioavailable and selective FD inhibitor. 12 demonstrated systemic suppression of AP activation in a lipopolysaccharide-induced AP activation model as well as local ocular suppression in intravitreal injection-induced AP activation model in mice expressing human FD.

Discovery of a highly potent orally bioavailable imidazo-[1, 2-a]pyrazine Aurora inhibitor.

Imidazo-[1, 2-a]pyrazine 1 is a potent inhibitor of Aurora A and B kinase in vitro and is effective in in vivo tumor models, but has poor oral bioavailbility and is unsuitable for oral dosing. We describe herein our effort to improve oral exposure in this class, resulting ultimately in the identification of a potent Aurora inhibitor 16, which exhibited good drug exposure levels across species upon oral dosing, and showed excellent in vivo efficacy in a mouse xenograft tumor model when dosed orally.

The Discovery of ( S)-1-(6-(3-((4-(1-(Cyclopropanecarbonyl)piperidin-4-yl)-2-methylphenyl)amino)-2,3-dihydro-1 H-inden-4-yl)pyridin-2-yl)-5-methyl-1 H-pyrazole-4-carboxylic Acid, a Soluble Guanylate Cyclase Activator Specifically Designed for Topical Ocular Delivery as a Therapy for Glaucoma.

Soluble guanylate cyclase (sGC), the endogenous receptor for nitric oxide (NO), has been implicated in several diseases associated with oxidative stress. In a pathological oxidative environment, the heme group of sGC can be oxidized becoming unresponsive to NO leading to a loss in the ability to catalyze the production of cGMP. Recently a dysfunctional sGC/NO/cGMP pathway has been implicated in contributing to elevated intraocular pressure associated with glaucoma. Herein we describe the discovery of molecules specifically designed for topical ocular administration, which can activate oxidized sGC restoring the ability to catalyze the production of cGMP. These efforts culminated in the identification of compound (+)-23, which robustly lowers intraocular pressure in a cynomolgus model of elevated intraocular pressure over 24 h after a single topical ocular drop and has been selected for clinical evaluation.

Sequential regulatory activity prediction across chromosomes with convolutional neural networks.

Models for predicting phenotypic outcomes from genotypes have important applications to understanding genomic function and improving human health. Here, we develop a machine-learning system to predict cell-type-specific epigenetic and transcriptional profiles in large mammalian genomes from DNA sequence alone. By use of convolutional neural networks, this system identifies promoters and distal regulatory elements and synthesizes their content to make effective gene expression predictions. We show that model predictions for the influence of genomic variants on gene expression align well to causal variants underlying eQTLs in human populations and can be useful for generating mechanistic hypotheses to enable fine mapping of disease loci.

Synthesis and SAR studies of imidazo-[1,2-a]-pyrazine Aurora kinase inhibitors with improved off-target kinase selectivity.

The structure-activity relationships of new Aurora A/B kinase inhibitors derived from the previously identified kinase inhibitor 12 are described. Introduction of acetic acid amides onto the pyrazole of compound 12 was postulated to influence Aurora A/B selectivity and improve the profile against off-target kinases. The SAR of the acetic acid amides was explored and the effect of substitution on enzyme inhibition as well as mechanism-based cell activity was studied. Additionally, several of the more potent inhibitors were screened for their off-target kinase selectivity.

Discovery of novel imidazo[1,2-a]pyrazin-8-amines as Brk/PTK6 inhibitors.

A series of substituted imidazo[1,2-a]pyrazin-8-amines were discovered as novel breast tumor kinase (Brk)/protein tyrosine kinase 6 (PTK6) inhibitors. Tool compounds with low-nanomolar Brk inhibition activity, high selectivity towards other kinases and desirable DMPK properties were achieved to enable the exploration of Brk as an oncology target.

Bioisosteric approach to the discovery of imidazo[1,2-a]pyrazines as potent Aurora kinase inhibitors.

Our continued effort toward the development of the imidazo[1,2-a]pyrazine scaffold as Aurora kinase inhibitors is described. Bioisosteric approach was applied to optimize the 8-position of the core. Several new potent Aurora A/B dual inhibitors, such as 25k and 25l, were identified.

Discovery of orally bioavailable imidazo[1,2-a]pyrazine-based Aurora kinase inhibitors.

We report a series of potent imidazo[1,2-a]pyrazine-based Aurora kinase inhibitors. Optimization of the solvent accessible 8-position led to improvements in both oral bioavailability and off-target kinase inhibition. Compound 25 demonstrates anti-tumor activity in an A2780 ovarian tumor xenograft model.

Discovery of imidazo[1,2-a]pyrazine-based Aurora kinase inhibitors.

The synthesis and structure-activity relationships (SAR) of novel, potent imidazo[1,2-a]pyrazine-based Aurora kinase inhibitors are described. The X-ray crystal structure of imidazo[1,2-a]pyrazine Aurora inhibitor 1j is disclosed. Compound 10i was identified as lead compound with a promising overall profile.

Structure and activity relationships of tartrate-based TACE inhibitors.

The syntheses and structure-activity relationships of the tartrate-based TACE inhibitors are discussed. The optimization of both the prime and non-prime sites led to compounds with picomolar activity. Several analogs demonstrated good rat pharmacokinetics.

Discovery of a Potent, Injectable Inhibitor of Aurora Kinases Based on the Imidazo-[1,2-a]-Pyrazine Core.

The imidazo-[1,2-a]-pyrazine (1) is a dual inhibitor of Aurora kinases A and B with modest cell potency (IC50 = 250 nM) and low solubility (5 µM). Lead optimization guided by the binding mode led to the acyclic amino alcohol 12k (SCH 1473759), which is a picomolar inhibitor of Aurora kinases (TdF K d Aur A = 0.02 nM and Aur B = 0.03 nM) with improved cell potency (phos-HH3 inhibition IC50 = 25 nM) and intrinsic aqueous solubility (11.4 mM). It also demonstrated efficacy and target engagement in human tumor xenograft mouse models.

The discovery of novel tartrate-based TNF-alpha converting enzyme (TACE) inhibitors.

A novel series of TNF-alpha convertase (TACE) inhibitors which are non-hydroxamate have been discovered. These compounds are bis-amides of L-tartaric acid (tartrate) and coordinate to the active site zinc in a tridentate manner. They are selective for TACE over other MMP's. We report the first X-ray crystal structure for a tartrate-based TACE inhibitor.