Disease variant prediction with deep generative models of evolutionary data

Quantifying the pathogenicity of protein variants in human disease-related genes would have a marked effect on clinical decisions, yet the overwhelming majority (over 98%) of these variants still have unknown consequences^1-3. In principle, computational methods could support the large-scale interpretation of genetic variants. However, state-of-the-art methods^4-10 have relied on training machine learning models on known disease labels. As these labels are sparse, biased and of variable quality, the resulting models have been considered insufficiently reliable¹¹. Here we propose an approach that leverages deep generative models to predict variant pathogenicity without relying on labels. By modelling the distribution of sequence variation across organisms, we implicitly capture constraints on the protein sequences that maintain fitness. Our model EVE (evolutionary model of variant effect) not only outperforms computational approaches that rely on labelled data but also performs on par with, if not better than, predictions from high-throughput experiments, which are increasingly used as evidence for variant classification^12-16. We predict the pathogenicity of more than 36 million variants across 3,219 disease genes and provide evidence for the classification of more than 256,000 variants of unknown significance. Our work suggests that models of evolutionary information can provide valuable independent evidence for variant interpretation that will be widely useful in research and clinical settings.

Eosinophil hematopoietins antagonize the programmed cell death of eosinophils. Cytokine and glucocorticoid effects on eosinophils maintained by endothelial cell-conditioned medium

Granulocyte-macrophage colony-stimulating factor (GM-CSF) was established as the constitutive and elicited human umbilical vein endothelial cell-derived eosinophil viability-sustaining factor. Stimulation of endothelium cell monolayers with IL-1 alpha (5 U/ml) increased the 48-h elaboration of GM-CSF from a mean of 3.2 to a mean of 8.2 pM (P less than 0.05). Dexamethasone (100 nM) decreased the constitutive GM-CSF elaboration by 49% (P less than 0.001) but did not diminish production by IL-1 alpha-stimulated endothelium. However, eosinophil viability decreased by 21% in dexamethasone-pretreated IL-1 alpha-stimulated endothelial cell-conditioned medium (P less than 0.05), which suggested viability antagonism by glucocorticoids. After 24 h of culture, eosinophil viability for replicate cells in enriched medium alone or with 1 pM GM-CSF decreased from means of 43 and 75% to means of 21 and 54%, respectively, when dexamethasone was included (P less than 0.05). However, 10 pM GM-CSF, IL-3, or IL-5 protected the cells against dexamethasone and against endonuclease-specific DNA fragmentation. In this model system of eosinophil-tissue interactions, dexamethasone prevents the endothelial cells from inducing a pathobiologic phenotypic change in the eosinophil by suppression of GM-CSF elaboration to concentrations that are not cytoprotective. Cytokine priming by GM-CSF, IL-3, or IL-5 may account for the differential responsiveness of select eosinophilic disorders to glucocorticoids.

Whitefly transmission and efficient ssDNA accumulation of bean golden mosaic geminivirus require functional coat protein

Bean golden mosaic geminivirus (BGMV) has a bipartite genome composed of two circular ssDNA components (DNA-A and DNA-B) and is transmitted by the whitefly, Bemisia tabaci. DNA-A encodes the viral replication proteins and the coat protein. To determine the role of BGMV coat protein systemic infection and whitefly transmission, two deletions and a restriction fragment inversion were introduced into the BGMV coat protein gene. All three coat protein mutants produced systemic infections when coinoculated with DNA-B onto Phaseolus vulgaris using electric discharge particle acceleration "particle gun." However, they were not sap transmissible and coat protein was not detected in mutant-infected plants. In addition, none of the mutants were transmitted by whiteflies. With all three mutants, ssDNA accumulation of DNA-A and DNA-B was reduced 25- to 50-fold and 3- to 10-fold, respectively, as compared to that of wild-type DNA. No effect on dsDNA-A accumulation was detected and there was 2- to 5-fold increase in dsDNA-B accumulation. Recombinants between the mutated DNA-A and DNA-B forms were identified when the inoculated coat protein mutant was linearized in the common region.

A controlled, prospective, 1-year trial of citalopram in the treatment of panic disorder

BACKGROUND

The objective of this study was to evaluate the efficacy and tolerability of citalopram in the long-term treatment of adult outpatients with panic disorder with or without agoraphobia.

METHOD

Patients in this double-blind, parallel-group trial were assigned to 1 of 3 fixed dosage ranges of citalopram (10 or 15 mg/day, 20 or 30 mg/day, or 40 or 60 mg/day), 1 dosage range of clomipramine (60 or 90 mg/day), or placebo. After the completed 8-week acute treatment period, the eligible patients could continue the treatment for up to 1 year. Of the 475 patients who were randomly assigned for the short-term trial, 279 agreed to continue double-blind treatment at their assigned doses. The primary efficacy measure used was the Clinical Anxiety Scale panic attack item, and the response was defined as no panic attacks (score of 0 or 1). The other key measures used were the Physician's Global Improvement Scale, the Patient's Global Improvement Scale, and the Hamilton Rating Scale for Anxiety (HAM-A).

RESULTS

In all drug-treated groups, except the group receiving the lowest citalopram dose, the treatment outcome was generally better than with placebo. As determined by a life table analysis of response, the probability of response during the 12 months was significantly greater with all treatment regimens than with placebo (p < .05), with citalopram 20 or 30 mg/day demonstrating the best response. Panic attacks tended to disappear in all patients remaining in the study until the end of follow-up. Analysis of the difference in the number of patients in different treatment groups remaining in the study (perhaps the best measure of long-term efficacy) also demonstrated that the patients treated with citalopram in dosage ranges of 20 or 30 mg/day and 40 or 60 mg/day had better response than placebo-treated patients (p < .0002 and p < .004, respectively). HAM-A and Global Improvement Scale scores also showed that patients treated with active drug showed greater improvement than placebo-treated patients. All treatment groups showed no new or exceptional adverse event clusters.

CONCLUSION

Citalopram in the dosage range of 20 to 60 mg/day is effective, well tolerated, and safe in the long-term treatment of patients who have panic disorder.

Discovery of a potent and orally bioavailable benzolactam-derived inhibitor of Polo-like kinase 1 (MLN0905)

This article describes the discovery of a series of potent inhibitors of Polo-like kinase 1 (PLK1). Optimization of this benzolactam-derived chemical series produced an orally bioavailable inhibitor of PLK1 (12c, MLN0905). In vivo pharmacokinetic-pharmacodynamic experiments demonstrated prolonged mitotic arrest after oral administration of 12c to tumor bearing nude mice. A subsequent efficacy study in nude mice achieved tumor growth inhibition or regression in a human colon tumor (HT29) xenograft model.

ProteinGym: Large-Scale Benchmarks for Protein Design and Fitness Prediction

Predicting the effects of mutations in proteins is critical to many applications, from understanding genetic disease to designing novel proteins that can address our most pressing challenges in climate, agriculture and healthcare. Despite a surge in machine learning-based protein models to tackle these questions, an assessment of their respective benefits is challenging due to the use of distinct, often contrived, experimental datasets, and the variable performance of models across different protein families. Addressing these challenges requires scale. To that end we introduce ProteinGym, a large-scale and holistic set of benchmarks specifically designed for protein fitness prediction and design. It encompasses both a broad collection of over 250 standardized deep mutational scanning assays, spanning millions of mutated sequences, as well as curated clinical datasets providing high-quality expert annotations about mutation effects. We devise a robust evaluation framework that combines metrics for both fitness prediction and design, factors in known limitations of the underlying experimental methods, and covers both zero-shot and supervised settings. We report the performance of a diverse set of over 70 high-performing models from various subfields (eg., alignment-based, inverse folding) into a unified benchmark suite. We open source the corresponding codebase, datasets, MSAs, structures, model predictions and develop a user-friendly website that facilitates data access and analysis.

Effect of erythrocyte storage and oxyhemoglobin affinity changes on cardiac function

Storage of blood can depress erythrocyte 2,3-diphosphoglycerate (DPG) levels and thereby increase oxyhemoglobin affinity and potentially decrease capillary-to-tissue oxygen transport. We measured myocardial function and metabolism in isolated rabbit hearts with fixed coronary flow under basal conditions and during isoproterenol stress at 37 and 30 degrees C, comparing high and low oxyhemoglobin affinity (OHA) erythrocytes. The high OHA state resulted from standard storage conditions, which caused depressed values of DPG and P50 (the oxygen tension at which hemoglobin is 50% saturated). The low OHA erythrocytes were initially stored and then underwent biochemical treatment to restore the DPG and P50 values to normal. The low OHA cells released more oxygen, and myocardial oxygen consumption and contractile function were increased relative to the high OHA cells during both the basal and stress states at both 37 and 30 degrees C. These observations may be relevant for patients with limited coronary flow when such patients receive large transfusions of stored blood.

Proton NMR of human breast tumors: correlation with clinical prognostic parameters

Proton NMR spectroscopy and imaging of human breast tissue have provided new methods in studying breast carcinomas. Continuous wave proton NMR spectroscopy in this study is able to discriminate breast carcinomas from normal breast tissue on the basis of the integrated area under the water and lipid peaks, width at half height of the water peak, and chemical shift of the lipid peak. In addition, the NMR parameters were correlated with the following clinical and pathologic prognostic indices: TNM tumor stage, nuclear grade, and estrogen receptor status (ER). Width at half height of the lipid peak (1/2 delta lipid) correlated with tumor content and ER. Studies using higher resolution proton or phosphorus NMR spectra may separate signals that can correlate with biological information on breast neoplasms useful to the clinician. Chemical shift of the lipid peak may be used to sharpen contrast on MRI of breast tumors.

Nuclear magnetic resonance assessment of adenosine triphosphate (ATP) dynamics in ischemic mouse livers perfused with adenine and ribose

Hepatic energy stores are essential to liver viability. We used a mouse liver perfusion model and MR spectroscopy to study the effect of adding two precursors of ATP (adenine and ribose) on ATP dynamics during ischemia and reperfusion. Using Krebs-Henseleit buffer with or without added adenine and ribose made little difference in the ATP decay rate during ischemia, but the recovery of ATP during reperfusion was more complete when adenine and ribose were added to the buffer. These findings suggest that the addition of the precursors of ATP, adenine and ribose, to perfusate after ischemia can accelerate and enhance ATP recovery.

Characteristics predicting outcomes of allogeneic stem-cell transplantation in relapsed acute myelogenous leukemia

BACKGROUND

Allogeneic hematopoietic stem-cell transplantation (ahsct) is associated with significant morbidity and mortality, but it can cure carefully selected patients with acute myeloid leukemia (aml) in second remission (cr2). In a cohort of patients with aml who underwent ahsct in cr2, we determined the pre-transplant factors that predicted for overall survival (os), relapse, and non-relapse mortality. We also sought to validate the prognostic risk groups derived by Michelis and colleagues in this independent population.

METHODS

In a retrospective chart review, we obtained data for 55 consecutive patients who underwent ahsct for aml in cr2. Hazard ratios were used to describe the independent effects of pre-transplant variables on outcome, and Kaplan-Meier curves were used to assess outcomes in the three prognostic groups identified by Michelis and colleagues.

RESULTS

At 1, 3, and 5 years post-transplant, os was 60%, 45.5%, and 37.5% respectively. Statistically significant differences in os, relapse mortality, and non-relapse mortality were not identified between the prognostic risk groups identified by Michelis and colleagues. Women were less likely than men to relapse, and a modified European Society for Blood and Marrow Transplantation (mebmt) score of 3 or less was associated with a lower non-relapse mortality.

CONCLUSIONS

The 37.5% 5-year os in this cohort suggests that, compared with other options, ahsct offers patients with aml in cr2 a better chance of cure. Our study supports the use of the mebmt score to predict non-relapse mortality in this population.

Liposomes of enviroxime and phosphatidylcholine: definition of the drug-phospholipid interactions

Interaction of the antiviral compound, enviroxime (E), with natural and synthetic phosphatidylcholines in organic and aqueous media was studied. Although insoluble in chloroform, E dissolved in chloroform solutions containing phosphatidylcholines. Solvation was directly related to the length of the fatty acid chains of the phospholipid. Proton spin resonance studies suggested an interaction of the fatty acid chains with the aromatic rings of E. Suspension of E-phosphatidylcholine mixtures of molar ratios up to 0.7:1.0 in aqueous media resulted in the formation of multilamellar liposomes. Liposomes containing E were more stable permeability barriers than those prepared with phospholipid alone, a property previously observed with cholesterol. Competition experiments suggested that E bound to the same sites in lipid bilayers as does cholesterol. These data indicate that E is incorporated into lipid bilayers of liposomes and that it alters the physical properties of the liposomes in a manner similar to that of cholesterol.

An animal model of occlusion-hyperthermia of the liver

An animal model of partial hepatic artery occlusion and radiofrequency (RF)-induced hepatic hyperthermia was developed in the New Zealand white rabbit. Seventy-seven percent (10/13) of animals survived partial hepatic artery occlusion, 82% (41/50) survived RF-induced hepatic hyperthermia, and 66% (24/36) survived combined occlusion-hyperthermia. Mesenteric infarction secondary to inadvertent embolization of intestinal arteries was the principal cause of death in animals undergoing partial hepatic artery occlusion. Extrahepatic thermal toxicity appeared to be the major cause of death in animals subjected to hepatic hyperthermia or occlusion hyperthermia although some animals showed evidence of hepatic necrosis as well. Recent developments in our laboratory hold the promise that extrahepatic thermal toxicity can be eliminated by selectively focusing heat into the liver allowing exploration of the efficacy of occlusion-hyperthermia in controlling VX-2 tumors implanted in the rabbit liver.

Responses of Adaxial and Abaxial Stomata of Normally Oriented and Inverted Leaves of Vicia faba L. to Light

Stomatal conductances of normally oriented and inverted leaves were measured as light levels (photosynthetic photon flux densities) were increased to determine whether abaxial stomata of Vicia faba leaves were more sensitive to light than adaxial stomata. Light levels were increased over uniform populations of leaves of plants grown in an environmental chamber. Adaxial stomata of inverted leaves reached maximum water vapor conductances at a light level of 60 micromoles per square meter per second, the same light level at which abaxial stomata of normally oriented leaves reached maximum conductances. Abaxial stomata of inverted leaves reached maximum conductances at a light level of 500 micromoles per square meter per second, the same light level at which adaxial stomata of normally oriented leaves reached maximum conductances. Maximum conductances in both normally oriented and inverted leaves were about 200 millimoles per square meter per second for adaxial stomata and 330 millimoles per square meter per second for abaxial stomata. Regardless of whether leaves were normally oriented or inverted, when light levels were increased to values high enough that upper leaf surfaces reached maximum conductances (about 500 micromoles per square meter per second), light levels incident on lower, shaded leaf surfaces were just sufficient (about 60 micromoles per square meter per second) for stomata of those surfaces to reach maximum conductances. This ;coordinated' stomatal opening on the separate epidermes resulted in total leaf conductances for normally oriented and inverted leaves that were the same at any given light level. We conclude that stomata in abaxial epidermes of intact Vicia leaves are not more sensitive to light than those in adaxial epidermes, and that stomata in leaves of this plant do not respond to light alone. Additional factors in bulk leaf tissue probably produce coordinated stomatal opening on upper and lower leaf epidermes to optimally meet photosynthetic requirements of the whole leaf for CO(2).

The isolation of an X-dependent strain of haemophilus from otitis media identified as H. haemoglobinophilus (canis)

The growth factor requirements, together with the serological results, would seem to justify the view that strain M.G., isolated from a child, M.G., with otitis media, in pure culture has to be regarded as H. haemoglobinophilus (canis). It is interesting to note that two out of the three strains of H. haemoglobinophilus (canis) available in this laboratory agglutinate to a titre of 1:80 with the serum produced against H. aphrophilus, the only other X-dependent haemophilus we had used for the preparation of antiserum.

Left ventricular aneurysm formation: an iatrogenic cause for the third mogul

Two cases of left ventricular pseudoaneurysm formation developing post-operatively after mitral valve replacement are reported. The chest radiographs showed an abnormal protuberance on the left heart border at the site of the "third mogul". The definitive diagnosis of this protuberance, resulting from the development of a left ventricular aneurysm, was made on angiography in each case. This site is unusual for left ventricular aneurysm formation, with the exception of the annular subvalvular aneurysm described in the negro population. Relevant aetiological factors in the development of these post-operative left ventricular pseudoaneurysms are considered. Since pseudoaneurysms are more prone to rupture than true aneurysms, it is concluded that early diagnosis of development of these left ventricular pseudoaneurysms should be made.

The agglutination reactions of Haemophilus paraphrophilus and H. paraphrohaemolyticus, and some observations on the agglutination of H. Aphrophilus and H. haemoglobinophilus (H. canis)

Agglutination tests were used to study the surface antigens of two recently described species of the genus Haemophilus, H. paraphrophilus and H. paraphrohaemolyticus, and their antigenic relationship to other members of the genus. The results obtained with a few strains of H. haemoglobinophilus and H. aphrophilus are also reported. The species H. paraphrophilus appears to be homogeneous; no major cross-reactions were observed. The species H. paraphrohaemolyticus contains at least three serotypes, of which two have been defined in terms of agglutination reactions. Cross-agglutinations occurred between one strain of H. paraphrohaemolyticus and strains of the other V-dependent species, H. parainfluenzae and H. parahaemolyticus. Of the X-dependent species, H. haemoglobinophilus seems to be homogeneous, and the species H. aphrophilus is not. A non-specific antibody against horse blood in the medium occurred erratically and was present in only two antisera, those raised to H. aphrophilus strain Khairat and H. haemoglobinophilus strain no. NCTC8540.

Simple predictions from multifield inflationary models

We explore whether multifield inflationary models make unambiguous predictions for fundamental cosmological observables. Focusing on N-quadratic inflation, we numerically evaluate the full perturbation equations for models with 2, 3, and O(100) fields, using several distinct methods for specifying the initial values of the background fields. All scenarios are highly predictive, with the probability distribution functions of the cosmological observables becoming more sharply peaked as N increases. For N=100 fields, 95% of our Monte Carlo samples fall in the ranges ns∈(0.9455,0.9534), α∈(-9.741,-7.047)×10-4, r∈(0.1445,0.1449), and riso∈(0.02137,3.510)×10-3 for the spectral index, running, tensor-to-scalar ratio, and isocurvature-to-adiabatic ratio, respectively. The expected amplitude of isocurvature perturbations grows with N, raising the possibility that many-field models may be sensitive to postinflationary physics and suggesting new avenues for testing these scenarios.