Monthly Archives: January 2021

Natural language predicts viral escape

Viral mutations that evade neutralizing antibodies, an occurrence known as viral escape, can occur and may impede the development of vaccines. To predict which mutations may lead to viral escape, Hie et al. used a machine learning technique for natural language processing with two components: grammar (or syntax) and meaning (or semantics) (see the Perspective by Kim and Przytycka). Three different unsupervised language models were constructed for influenza A hemagglutinin, HIV-1 envelope glycoprotein, and severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) spike glycoprotein. Semantic landscapes for these viruses predicted viral escape mutations that produce sequences that are syntactically and/or grammatically correct but effectively different in semantics and thus able to evade the immune system.

Science, this issue p. 284; see also p. 233

The ability for viruses to mutate and evade the human immune system and cause infection, called viral escape, remains an obstacle to antiviral and vaccine development. Understanding the complex rules that govern escape could inform therapeutic design. We modeled viral escape with machine learning algorithms originally developed for human natural language. We identified escape mutations as those that preserve viral infectivity but cause a virus to look different to the immune system, akin to word changes that preserve a sentences grammaticality but change its meaning. With this approach, language models of influenza hemagglutinin, HIV-1 envelope glycoprotein (HIV Env), and severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) Spike viral proteins can accurately predict structural escape patterns using sequence data alone. Our study represents a promising conceptual bridge between natural language and viral evolution.

Viral mutations that allow an infection to escape from recognition by neutralizing antibodies have prevented the development of a universal antibody-based vaccine for influenza (1, 2) or HIV (3) and are a concern in the development of therapies for severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) infection (4, 5). Escape has motivated high-throughput experimental techniques that perform causal escape profiling of all single-residue mutations to a viral protein (14). Such techniques, however, require substantial effort to profile even a single viral strain, and testing the escape potential of many (combinatorial) mutations in many viral strains remains infeasible.

Instead, we sought to train an algorithm that learns to model escape from viral sequence data alone. This approach is not unlike learning properties of natural language from large text corpuses (6, 7) because languages such as English and Japanese use sequences of words to encode complex meanings and have complex rules (for example, grammar). To escape, a mutant virus must preserve infectivity and evolutionary fitnessit must obey a grammar of biological rulesand the mutant must no longer be recognized by the immune system, which is analogous to a change in the meaning or the semantics of the virus.

Currently, computational models of protein evolution focus either on fitness (8) or on functional or semantic similarity (911), but we want to understand both (Fig. 1A). Rather than developing two separate models of fitness and function, we developed a single model that simultaneously achieves these tasks. We leveraged state-of-the-art machine learning algorithms called language models (6, 7), which learn the probability of a token (such as an English word) given its sequence context (such as a sentence) (Fig. 1B). Internally, the language model constructs a semantic representation, or an embedding, for a given sequence (6), and the output of a language model encodes how well a particular token fits within the rules of the language, which we call grammaticality and can also be thought of as syntactic fitness (supplementary text, note S2). The same principles used to train a language model on a sequence of English words can train a language model on a sequence of amino acids. Although immune selection occurs on phenotypes (such as protein structures), evolution dictates that selection is reflected within genotypes (such as protein sequences), which language models can leverage to learn functional properties from sequence variation.

(A) Constrained semantic change search (CSCS) for viral escape prediction is designed to search for mutations to a viral sequence that preserve fitness while being antigenically different. This corresponds to a mutant sequence that is grammatical (conforms to the structure and rules of a language) but has high semantic change with respect to the original (for example, wild type) sequence. (B) A neural language model with a bidirectional long short-term memory (BiLSTM) architecture was used to learn both semantics (as a hidden layer output) and grammaticality (as the language model output). CSCS combines semantic change and grammaticality to predict escape (12). (C) CSCS-proposed changes to a news headline (implemented by using a neural language model trained on English news headlines) makes large changes to the overall semantic meaning of a sentence or to the part-of-speech structure. The semantically closest mutated sentence according to the same model is largely synonymous with the original headline.

We hypothesize that (i) language modelencoded semantic change corresponds to antigenic change, (ii) language model grammaticality captures viral fitness, and (iii) both high semantic change and grammaticality help predict viral escape. Searching for mutations with both high grammaticality and high semantic change is a task that we call constrained semantic change search (CSCS) (Fig. 1C) (12). Our language model implementation of CSCS uses sequence data alone (which is easier to obtain than structure) and requires no explicit escape information (is completely unsupervised), does not rely on multiple sequence alignment (MSA) preprocessing (is alignment-free), and captures global relationships across an entire sequence (for example, because word choice at the beginning of a sentence can influence word choice at the end) (supplementary text, notes S2 and S3).

We assessed the generality of our approach across viruses by analyzing three proteins: influenza A hemagglutinin (HA), HIV-1 envelope glycoprotein (Env), and SARS-CoV-2 spike glycoprotein (Spike). All three are found on the viral surface, are responsible for binding host cells, are targeted by antibodies, and are drug targets (15). We trained a separate language model for each protein using a corpus of virus-specific amino acid sequences (12).

We initially sought to understand the semantic patterns learned by our viral language models. We therefore visualized the semantic embeddings of each sequence in the influenza, HIV, and coronavirus corpuses using Uniform Manifold Approximation and Projection (UMAP) (13). The resulting two-dimensional semantic landscapes show clustering patterns that correspond to subtype, host species, or both (Fig. 2), suggesting that the model was able to learn functionally meaningful patterns from raw sequence.

(A and B) UMAP visualization of the high-dimensional semantic embedding landscape of influenza HA. (C) A cluster consisting of avian sequences from the 2009 flu season onward also contains the 1918 pandemic flu sequence, which is consistent with their antigenic similarity (15). (D) Louvain clusters of the HA semantic embeddings have similar purity with respect to subtype or host species compared with phylogenetic sequence clustering (Phylo). Bar height, mean; error bars, 95% confidence. (E and F) The HIV Env semantic landscape shows subtype-related distributional structure and high Louvain clustering purity. Bar height, mean; error bars, 95% confidence. (G) Sequence proximity in the semantic landscape of coronavirus spike proteins is consistent with the possible zoonotic origin of SARS-CoV-1, MERS-CoV, and SARS-CoV-2.

We quantified these clustering patterns, which are visually enriched for particular subtypes or hosts, with Louvain clustering (14) to group sequences on the basis of their semantic embeddings (fig. S1, A to C). We then measured the clustering purity on the basis of the percent composition of the most represented metadata category (sequence subtype or host species) within each cluster (12). Average cluster purities for HA subtype, HA host species, and Env subtype are 99, 96, and 95%, respectively, which are comparable with or higher than the clustering purities obtained with MSA-based phylogenetic reconstruction (Fig. 2, D and F, and fig. S1D) (12).

Within the HA landscape, clustering patterns suggest interspecies transmissibility. The sequence for 1918 H1N1 pandemic influenza belongs to the main avian H1 cluster, which contains sequences from the avian reservoir for 2009 H1N1 pandemic influenza (Fig. 2C and fig. S1, A to C). Antigenic similarity between H1 HA from 1918 and 2009, although nearly a century apart, is well supported (15). Within the landscape of SARS-CoV-2 Spike and homologous proteins, clustering proximity is consistent with the suggested zoonotic origin of several human coronaviruses (Fig. 2G), including bat and civet for SARS-CoV-1, camel for Middle East respiratory syndrome-related coronavirus (MERS-CoV), and bat and pangolin for SARS-CoV-2 (16). Analysis of these semantic landscapes strengthens our hypothesis that our viral sequence embeddings encode functional and antigenic variation.

We then assessed the relationship between viral fitness and language model grammaticality using high-throughput deep mutational scan (DMS) characterization of hundreds or thousands of mutations to a given viral protein. We obtained datasets measuring replication fitness of all single-residue mutations to A/WSN/1933 (WSN33) HA H1 (17), combinatorial mutations to antigenic site B in six HA H3 strains (18), or all single-residue mutations to BG505 and BF520 HIV Env (19), as well as a dataset measuring the dissociation constant (Kd) between combinatorial mutations to yeast-displayed SARS-CoV-2 Spike receptor-binding domain (RBD) and human ACE2 (20), which we used to approximate the fitness of Spike.

Language model grammaticality was significantly correlated (table S1, t-distribution P values) with viral fitness across all viral strains and across studies that examined single or combinatorial mutations (Fig. 3A), even though our language models were not given any explicit fitness-related information nor trained on the DMS mutants. When we compared viral fitness with the magnitude of mutant semantic change (rather than grammaticality), we observed significant negative correlation (table S1, t-distribution P values) in 8 out of 10 strains tested (Fig. 3A). This makes sense biologically because a mutation with a large effect on function is on average more likely to be deleterious and result in a loss of fitness. These results suggest that grammaticality of a given mutation captures fitness information and add an additional dimension to our understanding of how semantic change encodes perturbed protein function.

(A) Whereas grammaticality is positively correlated with fitness, semantic change has negative correlation, suggesting that most semantically altered proteins lose fitness. (B and C) However, a mutation with both high semantic change and high grammaticality is more likely to induce escape. Considering both semantic change and grammaticality enables identification of escape mutants that is consistently higher than that of previous fitness models or generic functional embedding models. (D) Across 891 surveilled SARS-CoV-2 Spike sequences, only three have both higher semantic change and grammaticality than a Spike sequence with four mutations that is associated with a potential reinfection case.

We then tested whether combining semantic change and grammaticality enables us to predict mutations that lead to viral escape. Our experimental setup involved making, in silico, all possible single-residue mutations to a given viral protein sequence; next, each mutant was ranked according to the CSCS objective that combines semantic change and grammaticality. We validated this ranking on the basis of enrichment of experimentally verified mutants that causally induce escape from neutralizing antibodies. Three of these causal escape datasets used a DMS with antibody selection to identify escape mutations to WSN33 HA H1 (1), A/Perth/16/2009 (Perth09) HA H3 (2), and BG505 Env (3). The fourth identified escape mutations to SARS-CoV-2 Spike by using natural replication error after in vitro passages under antibody selection (5), whereas the fifth performed a DMS to identify mutants that affect antibody binding to yeast-displayed Spike RBD (4).

We computed the area under the curve (AUC) of acquired escape mutations versus the total acquired mutations (12). In all five cases, escape prediction with CSCS resulted in both statistically significant and strong AUCs of 0.83, 0.77, 0.69, 0.85, and 0.71 for H1 WSN33, H3 Perth09, Env BG505, Spike, and Spike RBD, respectively (one-sided permutation-based P < 1 105 in all cases) (Fig. 3B and table S2). We did not provide the model with any information on escape, a setup in machine learning referred to as zero-shot prediction (7). The AUC decreased when ignoring either grammaticality or semantic change, evidence that both are useful in predicting escape (Fig. 3C, fig. S2A, and table S2). Although semantic change is negatively correlated with fitness, it is positively predictive (along with grammaticality) of escape (table S2), indicating that functional mutations are often deleterious, but when fitness is preserved, they are associated with antigenic change and subsequent escape from immunity.

We also tested how well alternative models of fitness (each requiring MSA preprocessing) (8, 21) or of semantic change (pretrained on generic protein sequence) (911) predict escape, although these models are not explicitly designed for escape prediction. Fitness models associate more frequently observed patterns with higher fitness and greater escape potential, whereas semantic models associate larger functional changes with escape (12). CSCS with our viral language models was more predictive of escape across all five datasets (Fig. 3B and fig. S2A). Moreover, the individual grammaticality or semantic change components of our language models often outperformed benchmark models (table S2).

Language modeling can also characterize sequence changes beyond single-residue mutations, such as from accumulated replication error or recombination (22), although our approach is agnostic to how a sequence acquires its mutations. We therefore estimated the antigenic change and fitness of a set of four mutations to the SARS-CoV-2 Spike associated with a reported reinfection event (23). Among 891 other distinct, surveilled Spike sequences, we found that only three (0.34%) represent both higher semantic change and grammaticality (Fig. 3D). We estimate significant escape potential of these four mutations (random mutant null distribution P < 1 108) (12), and we observed similar patterns for known antigenically dissimilar sequences (fig. S2B) (12). Our analysis suggests a way to quantify the escape potential of interesting combinatorial sequence changes, such as those from possible reinfection (23), and calls for more information that relates combinatorial mutations to reinfection and escape.

To further assess whether our model could learn structurally relevant patterns from sequence alone, we scored each residue on the basis of the CSCS objective, visualized escape potential across the protein structure, and quantified enrichment or depletion of escape (12). Escape potential is significantly enriched in the HA head (permutation-based P < 1 105) and significantly depleted in the HA stalk (permutation-based P < 1 105) (Fig. 4, A and B; fig. S3; and table S3), which is consistent with literature on HA mutation rates and supported by the successful development of antistalk broadly neutralizing antibodies (24). We also detected, consistent with existing knowledge, a significant enrichment (permutation-based P < 1 105) of escape mutations in the V1/V2 hypervariable regions of the HIV Env (Fig. 4, C and D; fig. S3; and table S3) (3). Our model only learns escape patterns linked to mutations, rather than posttranslational changes such as glycosylation that contribute to HIV escape (3), which may explain the lack of escape potential specifically assigned to Env glycosylation sites (Fig. 4C and table S3).

(A and B) HA trimer colored by escape potential. (C) Escape potential P values for HIV Env. The gray dashed line indicates the statistical significance threshold. (D) The Env trimer colored by escape potential, oriented to show the V1/V2 regions. (E and F) Potential for escape in SARS-CoV-2 Spike is significantly enriched at the N-terminal domain and receptor binding domain (RBD) and significantly depleted at multiple regions in the S2 subunit. The gray dashed line indicates the statistical significance threshold.

The escape potential within the SARS-CoV-2 Spike is significantly enriched in both the RBD (permutation-based P = 2.7 103) and N-terminal domain (permutation-based P < 1 105), whereas escape potential is significantly depleted in the S2 subunit (permutation-based P < 1 105) (Fig. 4, E and F; fig. S3; and table S3). These results are supported by the greater evolutionary conservation at S2 antigenic sites (25). Our model of Spike escape therefore suggests that immunodominant antigenic sites in S2 (5, 25) may be more stable target antibody epitopes and underscores the need for more exhaustive causal escape profiling of Spike in regions beyond the RBD.

Our study leverages the principle that evolutionary selection is reflected in sequence variation. This principle may allow CSCS to generalize beyond viral escape to different kinds of natural selection (such as T cell selection) or drug selection. CSCS and its components could be used to select elements of a multivalent or mosaic vaccine. Our techniques also lay the foundation for more complex modeling of sequence dynamics. We anticipate that the distributional hypothesis from linguistics (26), in which co-occurrence patterns can model complex concepts and on which language models are based, can further inform viral evolution.

M. Peters, M. Neumann, M. Iyyer, M. Gardner, C. Clark, K. Lee, L. Zettlemoyer, Deep contextualized word representations. Proc. NAACL-HLT, 22272237 (2018).

R. Rao, N. Bhattacharya, N. Thomas, Y. Duan, P. Chen, J. Canny, P. Abbeel, Y. Song, Evaluating protein transfer learning with TAPE. Proc. Adv. Neural Inf. Process. Syst., 96869698 (2019).

B. Hie, brianhie/viral-mutation: viral-mutation release 0.3. Zenodo (2020).doi:10.5281/zenodo.4034681

B. Foley, C. Apetrei, I. Mizrachi, A. Rambaut, B. Korber, T. Leitner, B. Hahn, J. Mullins, S. Wolinsky, HIV Sequence Compendium 2018, technical report LA-UR 18-2 (2018).

T. Mikolov, I. Sutskever, K. Chen, G. Corrado, J. Dean, Distributed representations of words and phrases and their compositionality. Proc. Adv. Neural Inf. Process. Syst., 31113119 (2013).

A. M. Dai, Q. V. Le, Semi-supervised sequence learning. Proc. Adv. Neural Inf. Process. Syst., 30793087 (2015).

C. C. Aggarwal, A. Hinneburg, D. A. Keim, in Proceedings of the International Conference on Database Theory (2001), vol. 1973, pp. 420434.

Excerpt from:

Learning the language of viral evolution and escape - Science Magazine

In the final, darkest days of the deadliest year in U.S. history, the world received ominous news of a mutation in the SARS-CoV-2 coronavirus. Scientists in the U.K. had identified a form of the virus that was spreading rapidly throughout the nation. Then, on January 4, Prime Minister Boris Johnson announced a lockdown that began almost immediately and will last until at least the middle of February. Its been both frustrating and alarming to see the speed with which the new variant is spreading, he said in an address, noting that our scientists have confirmed this new variant is between 50 and 70 percent more transmissible than previous strains.

Those figures, based on an early estimate by British government scientists in late December, made for terrifying push alerts and headlines. Though this strain of the virus (officially called B.1.1.7) quickly became known as the U.K. variant, it has already been found in 45 countries, suggesting that the opportunity to contain it with travel restrictions has passed. On January 8, Australia locked down Brisbane, a city of 2.3 million people, after discovering a single case.

Each day, B.1.1.7 is being found in more people in more places, including all around the United States. Experts have raised dire warnings that a 70 percent more transmissible form of the virus would overwhelm already severely stretched medical systems. Daily deaths have already tripled in recent months, and the virus is killing more than 3,000 Americans every day. From a purely mathematical perspective, considering exponential growth, a significantly more transmissible strain could theoretically lead to tens of thousands of daily deaths, with hospital beds lining sidewalks and filling parking lots.

Read: The problem with stories about dangerous coronavirus mutations

To make matters worse, the warnings from Britain were followed by headlines about yet another variant, B.1.351, in South Africa. Then another concerning variant was identified in Brazil. News reports speculated that these strains may resist vaccines. Some experts cautioned that the mutations could render current treatments less effective. Scott Gottlieb, the former director of the FDA, said last week: The South Africa variant is very concerning right now because it does appear that it may obviate some of our medical countermeasures, particularly the antibody drugs. On Tuesday, Anthony Fauci echoed that concern, calling the variant disturbing.

These new variants demand to be taken seriously. Skyrocketing case counts in the U.K. suggest a potential to do enormous damage, and the identification of B.1.1.7 in so many countries is noteworthy. Still, we dont yet know whether either variant will become as dominant worldwide as they have in their respective countries. They might spread widely and cause tremendous harm. They might also do neither.

The sheer scale and capacity of this virus are challenging many things we thought we knew, but the basic laws governing its evolution are not among them. All viruses are constantly evolving and changing, just as human populations are. When a virus is spreading as widely and rapidly as SARS-CoV-2, spinning through trillions of generations each minute, adaptation is inevitable. The transmissibility of the virus will change. The severity of the disease it causes will change. Its ability to evade our immune system will change. It very well may evolve to circumvent our current vaccines.

Thanks to genetic-sequencing technology, we can watch this evolution in real time. We can see the changes in a viruss genes before we even know what they mean for the spread of disease. Charting the course of this evolution, and assessing its significance, has quickly become a foremost challenge of the pandemic. The peril is not that the virus will suddenly change in an extraordinary way that transforms the pandemic, but that it is changing in small, ordinary ways that are playing out on a vast scale, and whose significance we may not appreciate until its too late.

Almost exactly a year ago, in January of 2020, a flight attendant warned the renowned Chinese virologist Zhang Yongzhen: It was time to turn off all portable electronic devices. He was sitting with his phone to his ear. On the other end of the line, his Australian collaborator Eddie Holmes was pleading with him to publish the genetic code of the novel coronavirus.

The Chinese government had forbidden this. Yongzhen was torn. The world did not yet know the cause of the rapidly spreading respiratory infection, and he seemed to have uncovered it in a sample of sputum from a severely ill person in Wuhan. Using genomic sequencing to unravel the code of the virus, he had found what appeared to be the blueprint of a new coronavirus.

He told Holmes to publish the code. When Holmes did so on Twitter, the international scientific community pounced. Within days, researchers in Thailand were able to verify that the same virus had infected patients there. Scientists at the U.S. National Institutes of Health began to work on a vaccine. The code became the backbone of the Pfizer/BioNTech and Moderna vaccines, which owe their development to the speedy identification and sharing of the genome.

Read: The end of the pandemic is now in sight

The exact sequence that Holmes tweeted is now a relic. The virus it represented is gone, replaced by many, many, many subsequent generations. New lineages have arisen in different parts of the world, and hundreds of thousands of slightly different sequences have been added to an international database. There are now thousands of unique SARS-CoV-2 genomes, each the result of myriad permutations of mutations in the code. There is no single, standard genetic code for this coronavirus, any more than there is a standard human genome.

The term variant is misleading, in that it creates the idea that all the other viruses are the same, explains Ramon Lorenzo Redondo, a genomic analyst at Northwestern Universitys Feinberg School of Medicine. Technically, every version of the virus is a variant. Even within a single person, the virus changes and evolves many times. If you were to have your bodily fluids sequenced on different days, the viral strains would show new mutations. Viruses operate as a cloud of mutantsa swarm of mutants, Redondo told me.

This is not a flaw in the system, but rather the way viruses work. When it comes to reproduction, viruses are sloppy. The speed and scale of their replication come at the cost of accuracy; they operate like a spam email marketing scheme, favoring inundation over meticulous grammar. Insofar as a virus can be said to have a goal, the goal is to ensure as many future generations as possible. To that end, it fires off shotgun blasts of imperfect clones, gambling that a few will make their way to other cells and penetrate them.

Almost all of these accidental mutations are inconsequential: The virus still looks and functions just as its parent before it did. Over time, though, sets of mutations can layer on top of one another and accumulate, and the virus begins to function differently. Some of these differences confer an advantage of one sort or anotherfor example, increased transmissibility.

What were observing is very expected, Paul Turner, a professor of ecology and evolutionary biology at Yale, told me. If a population can improve in its environment, evolution lets that happen. The virus population size is expanding, and mutations spontaneously occur.

Although its not news that the virus has mutated, its extremely important to keep an eye on the general direction of the changesand what they mean for the humans whose cells are being hijacked. If you see a mutation that could allow the virus to escape detection by the immune system, or escape vaccine coverage, thats very worrisome, Turner said. We dont have evidence of that yet.

The likelihood of these scenarios depends on a few factors. Some viruses mutate more readily than others: Influenza mutates so quickly that new strains spread around the world each year, requiring the creation of new vaccines. Measles, by contrast, mutates slowly, so people who were vaccinated decades ago are very likely still protected. Coronaviruses typically dont mutate very quickly, Turner said. I dont see any evidence that this coronavirus is going to suddenly become like influenza. But right now there are so many people infected, and the virus is in a new environment [humans instead of bats], so Im not surprised that evolution is pushing it to improve.

In the long run, he believes, the spread of this coronavirus will more closely resemble measles than flu. Although we may need to update our vaccines occasionally, we wont need to do so every year. But as long as rates of infection remain high, the coronavirus is likely to acquire, over months or years, the ability to at least partially bypass our immune responses. Second-time infections may be less severe, but their severity also depends on how the virus evolves. And we may develop immunity to one variant but not to another.

In anticipation of such complexities, Redondo and others have been creating and updating phylogenetic mapsessentially, family treesfor this coronavirus. Groups with a common ancestor are referred to as a lineage. A lineage is something like a human family: different individuals sharing a common ancestor. (B.1.1.7 and B.1.351 are separate lineages that evolved similar changes in their spike protein independently.)

Genetic commonalities can also define broader groups called clades. Last spring, a clade known as D614G came to dominate the world. This was attributed to a mutation in the spike protein that made this group more transmissible than previous strains. And this was just one part of a family tree thats now more like a forest. The two first clades that were defined have disappeared, Redondo said. Right now five major clades are jockeying for dominance, he said, but the picture is constantly shifting.

The emergence of a new clade can be as difficult to predict as any rise to global domination. Obsolescence and dominance are determined by the qualities of the lineages, the characteristics of the host populations, and the legacies of previous microbial invaders. The fact that a lineage or clade is dominant in one place, within one group of humans, does not mean it will be in others. So far, the U.K. and South African variants are dominating local surges, but they are not expansive enough to be considered clades. The South African variant, for example, accounts for about 90 percent of the genetic sequences analyzed in the country, but remains a minor player elsewhere.

Similarly, the B.1.1.7 variant was identified in the U.K. in September, yet so far dominates only one geographic region. Although genomic testing in the U.S. is relatively sparse, were doing enough sequencing that we know its not that common in the U.S., says Nathan Grubaugh, a microbial epidemiologist at the Yale School of Public Health. This variant doesnt seem to be more than 1 or 2 percent of cases at the moment. Its here, and its very widespread, but its low in frequency. I think, for the most part, this is true globally.

For now, these variants may be thought of like weeds in a garden. They have shown that they have the capacity to take over in some areas. There is a plausible mechanism that could allow them to do so elsewhere: Both the U.K. and South African variants share a mutation that manifests as a subtle change in a key site where the virus binds to human cells. But weeds take over for many reasons, and sometimes they have more to do with the garden than the weed. Human populations vary in so many waysbehaviorally, genetically, immunologically, geographically, environmentallythat the degree to which a regional surge in cases is due to a change in the virus itself is extremely difficult to discern. And that leads to some uncertainty. We may have it wrong in the end, [and] its not actually more transmissible, Grubaugh told me. It seems to be, but we may have been fooled.

Even if these variants are indeed as transmissible as the rising case numbers suggest, transmissibility is only one determinant of a viruss overall potential to do harm. Sometimes viruses become more transmissible but ultimately less dangerous. And of course, each new variant is but an intermediate step toward some other form of the virus. The real challenge is understanding how any given change fits into all of these larger patterns, and what that means for us.

The plot of Michael Crichtons 1969 novel, The Andromeda Strain, hinges on an extraterrestrial microorganism that mutates its way out of containment. A similar narrative device drives the film Outbreak, in which a bleed-from-the-eyes virus suddenly becomes airborne. What amounts to a lazy screenwriting clich has loaded the word mutation with such horrifying subtext that its almost unusable. The process of viral evolution is much subtler, and requires a careful eye to detect. That subtlety is what makes it dangerous.

There are two basic ways that a coronavirus can become more transmissible. One is by binding more effectively to human cells. When this happens, a person who inhales viral particles becomes slightly more likely to develop an infection. The other is by replicating more efficiently, creating higher numbers of viral particles (higher viral load) in an infected person, so that they exhale more particles with each breath (making it statistically more likely that one of the particles will infect someone else). If a breath contains 10 percent more viral particles, it is that much more likely that one will land in someone elses nose.

Its unclear whether one or both of these mechanisms are at play in the U.K. and South African lineages, but we know their effects can be complex. If a person is carrying a much higher viral load, for example, they may get sick more quickly. That sounds badand it certainly is for that person. But a shorter asymptomatic period could ultimately make the virus easier to contain. This was the case with the first SARS coronavirus, in 2003 (SARS-CoV-1), which caused a more severe disease than SARS-CoV-2 does, but killed far fewer people in total because each case was identifiable.

By the same token, this coronavirus could evolve to cause a somewhat less severe illnesssomething slightly closer to that caused by the other four endemic coronaviruses. The common cold is extremely transmissible, yet rarely fatal. This makes sense from an evolutionary perspective: Viruses that kill their hosts are less likely to become dominant than those that dont. It could be that transmissibility correlates with being kinder to your host, Turner said. Weve observed that in other realms of virus evolution. Natural selection would, hypothetically, favor the versions that leave people feeling well enough to be out and about, spreading the virus to other hosts.

Major changes in the severity of the diseasein either directionare unlikely, but the scale at which this virus is operating means that small differences in things like transmissibility are amplified, and can manifest as significant changes in how many people get sick. Experts widely agree that playing it safe in the coming weeks is prudent. Oliver Pybus, a professor of evolution and infectious diseases at the University of Oxford, emphasizes that understanding why B.1.1.7 took over the U.K. is extremely scientifically difficult. He has been at the forefront of identifying and tracking the variant, but says huge questions remain unanswered. Theres still considerable uncertainty as to the long-term consequences of B.1.1.7, Pybus told me. We dont even know whether this lineage truly originated in the U.K., with so many countries not doing this surveillance.

Though much of the world is now on alert for this particular variant, Pybus said that very few places are sequencing genomes as comprehensively as the U.K. is. In some places, institutions are sampling but not sharing findings in the public domain. Both elements are crucial. Testing with PCR or antigen tests alone is no longer sufficient. Positive tests must be followed by analyses of the genomes of the virus. The more genomes we have, the more effectively we can identify anomalous patterns, both to raise alarms early and to avoid raising false ones.

The U.S. is especially far behind the U.K. in this regard. Without a baseline level of genomic surveillance, Yales Grubaugh told me, we do not know if a city like New York would be as devastated by B.1.1.7 as London has been. The forecast for any given variant depends on context that we lack. I dont think any one state is doing enough sequencing yet, Grubaugh said. Sequencing is the most important thing. We dont have a big organized project like in the U.K. What we have is a bunch of individual labs, mostly at academic medical centers. Sporadic sequencing is arguably as bad as none at all, in that it can fail to represent how and why variants are spreading. And focusing too narrowly on hunting one particular variant can mean failing to notice other, possibly more consequential warning signs.

The hunt for any one variant also introduces selection bias, making it hard to know if the variant is truly spreading more readily than others, or if we are just looking harder for it. After finding a person carrying B.1.1.7 in New York last week, for example, state health officials sequenced the genomes of nearby casesan approach that is likely to find a disproportionate number of B.1.1.7 cases. Without constant, widespread surveillance testing, Pybus said, its difficult to discern an accurate overall picture.

The field of genomic epidemiology is going through its adolescence in public, developing in full view of the most extraordinary event of the century, Pybus told me. The ability to identify new viral lineages before we even understand how they will affect people may allow epidemiologists to warn of consequential variantsbut may undermine their credibility when strains dont prove as dangerous as headlines predict.

Even if we cannot contain this particular variant, were learning from its spread. Preventing more virulent strains from becoming dominantwhen they inevitably do arisemay be possible if we can track genomic patterns more widely, so that we have the context needed to determine whether a strain is indeed uniquely dangerous. If we can take steps to contain a new threat early enough, it may never become widespread. If we miss these opportunities, we risk repeating the kind of mistake that allowed the original SARS-CoV-2 strain to escape China in the first place.

Last week, Eddie Holmes reflected on the fateful moment when he tweeted the viruss original genetic code. It was a moment of triumph for collaborative science, but the work was just beginning. The triumph must be repeated daily. What worries me most of all is if politics gets in the way of data sharing and science, he told Medscape. Step one has to be immediate, rapid, open data sharing. Speed is of the essence in a pandemic. Any barrier to working together makes this a much less safe world. That should be the lesson of this outbreak.

The changing genetic code of the coronavirus will not nullify our fundamental strategies for ending the pandemic. With better data, we can keep our vaccines and antibody treatments up to date and our shutdown measures as minimal as possible, and we can sever any ominous new chains of transmission. The spread of new variants is a stark reminder that we all have an immediate part to play in this. If you carry the virus, it will mutate within you. You could be the person in whom a new, even more threatening variant emerges. You could seed the entire world with it. But no matter how the virus mutates, the same basic preventive measuresthe unglamorous ones weve been lectured about for nearly a yearwill still have the power to ensure that you dont.

See the rest here:

The Evolution of the Coronavirus - The Atlantic

Andrew Young, Vice President of Project Engineering and Service Delivery at Akselos, gives an insight into the evolution of the digital twin technology and its application in the hydro industry.

Awarded as Technology Pioneer 2020 by the World Economic Forum, Akselos provides real-time predictive digital twins for large asset integrity management. Founded in 2012 and with operations in Europe, the US, and South East Asia, the companys products are designed specifically to help protect the worlds critical infrastructure with real-time, condition-based monitoring for large critical assets.

Using patented algorithms developed over 15 years of research at Massachusetts Institute of Technology (MIT) and funded by the US Department of Defence, Akselos software solution is helping the energy sector optimise operations, lower operating costs, and extend the life of ageing assets.

The Akselos software is based on a newly developed algorithm that is 1000 times faster than legacy technology. The result of 15 years of research in the mechanical engineering lab at MIT, it changes the process of structural assessment to allow for real-time, continuous monitoring of large assets something that has never been done before. Its new mathematical approach, called Reduced-Basis Finite Element Analysis (RB-FEA), scales up and accelerates the legacy approach to model large-scale assets in detail and accuracy. Described as the worlds most powerful structural analysis tool and the only one fast enough to cope with the speed of a sensor feed, it has been validated by over 100 scientific publications, an MIT patent, years of co-development with Shell, and subsequent software validation with the American Bureau of Shipping.

The technology changes how we look at structural integrity management, it empowers asset operators to monitor condition continuously, rather than the current workflow which relies on periodic inspection data to update conventional finite element models. This means that the understanding of the assets condition is on an as is basis - rather than at the time of the last inspection. The result is eyes on the asset 24/7, reassuring operators that structures are safe to operate and efficiency is increased through a move to risk-based inspection.

Akselos technology allows maintenance engineers to really pinpoint, using rules of science and engineering, where structural integrity issues are likely to occur. It is also possible to establish what the operational asset life is in terms of fatigue assessment while supporting predictive maintenance.

The digital twin is essentially a tool with which plant operator ESB can simulate many parameters around the day-to-day operation of the plant and visually pan through the entire structure in 3D and see how it reacts.

Akselos can simulate future events because of the nature of the physics-based twin. Whereas traditional engineering approaches are assessing the probability of a known and identified established defect reoccurring, Akselos can assess and identify the risk of any type of failure based on the structural assessment. Results can be made available within an operations time frame supporting the asset manager as a fault does not have to already be pre-identified.

Akselos' underground asset uses the rules of science and engineering

Turlough Hill pumped storage plant in Ireland had been in operation for almost five decades but the operator ESB had many questions on the actual structural condition of the main penstock and water distribution manifold. In addition, an in-depth assessment of the consumed structural capacity of plant had never been previously performed.

Akselos technology was used to build a structural digital twin of the large penstock and manifold for the plant.Thanks to Akselos high-fidelity structural modelling capabilities, ESB identified the digital twin technology as being suited to performing such an analysis for the very first time. Not only did ESB want to understand the consumed fatigue life of the station, but they desired a tool to run what-if scenarios for future and more demanding operating conditions.

Engineers at Akselos digitised all available data on the 700m long asset, from which a 3D rendered structural digital model was built. First of all they had to establish the geometries, interactions and stresses in the plant system to create an intuitive heat map. The navigable 3D model shows which, when and where stresses are forming through the power station and so help direct inspections to specific areas of concern, rather than just doing a random walk through.

The structural digital twin, built up from a fully digitised asset, takes in the live feeds of the current loading of the system, cycles of exploitation such as generation, pumping stoppages and allows a real time assessment of the stress distribution in the system. In addition, post-processing of the cycles allows for calculation of accumulated fatigue and residual structural life.

Akselos technology allows for a speed-up of these assessments into operational time frames, meaning that the operator can be fully informed on the state of the structure without stoppages. Typical approaches require a stoppage to drain down the entire system that can take weeks while inspections are conducted.

Engineering drawing (left) and digital replica (right)

The main objective of the ESB team was to get a new, up-to-date engineering understanding of the as-designed water distribution system of the hydro plant. In the original concept the main design parameter was static pressure containment with safety factors that were fixed in time. Time based degradation mechanisms that might impact safety margins were not considered as part of the original design, such as fatigue or changes to the thickness of the supporting surrounding structure. There was no original design life, but an anticipated operational number of hours had long passed in the current exploitation of the hydro system.

Another objective was to optimise the integrity management process and strategy: failure of this part of the system is not an option for the operator. The extreme size of the structure (for example over 8km of welding alone), observed cracking in some welded zones, low or zero accessibility on much of the structure, and high costs of drain downs, are part of the considerations that made the business case easily attainable.

With the advent of a tool such as the Akselos modeller there can be a new approach to the inspection strategy and better understanding of the required duration of the outages for inspection purposes due to greater control of the inspection scopes.

Finally, as the hydro plan enters the age of low carbon power generation with increasing renewables as part of the energy mix, there are also changes to the operational demands of the system. Stop-start generation to market prices or due to grid instability are giving a higher frequency and demand usage. These more frequent and more varied cycles of exploitation were not considered in the original design.

Identifying stress pressure loads with the digital twin

When the starting point is an existing asset operating beyond design life there are two immediate wins for the operator:

In the case of the Turlough Hill station another advantage is for the assessment of the impact of the new modes of operation, mainly higher cycle times with more demand on the hydro pumped station. With market driven operations, and the possibility for generation at times of highest price or storage at times of low or negative pricing, the pumped storage unit has become strategic in the operators portfolio.

If the hydro station or asset is a new build then there can be lower one-time engineering costs, along with reduced capital expenditure, as the simulation engine allows the design team to explore more options in terms of material types, assembly and welding approaches to assess optimal costing for the different design options.

In addition, the structural digital twin can be a powerful tool for the commissioning process when sensor data can be plugged into the digital twin to assess the stress distribution and any potential issues for operation. This can reduce the commissioning time and costs while also providing a good collaborative platform to involve all stakeholders in the process.

Finally, once in operation the digital twin can provide continuous monitoring of the asset and also give inputs to the inspection planning process.

In general, the earlier in the life of the asset that the digital twin is created, the sooner and the larger the value that can be generated. Essentially there is only a one-time set-up cost for digitisation. The earlier the digital twin is active and enabled for an asset then the greater the number of value frames can be recouped across the asset management life cycle.

The Turlough Hill digital twin has secured operator confidence for continued operation of the plant an effective life extension case. It has also provided an informed basis on which to assess the ongoing changes and evolution of loading of the asset in an increasingly dynamic energy generation context. Furthermore, the digital twin has demonstrated its ability to identify areas that need inspection by highlighting areas of defects. These hotspots were independently validated against cracking found during earlier inspections.

As a measure of its success, the digital twin initiative won the internal ESB performance award for 2020 and is considered to be a major improvement over alternatives in the marketplace to assess plant structural viability and forward management.

Currently Akselos is working with ESB and other hydro plant operators in Europe on further digital twin enhancements, as well as applications for critical areas such as the turbines blades and generators in addition to the water generation system (penstock and manifold).

The key benefits come from a highly transparent method for structural integrity assessment, grounded in recognised engineering standards and usable by all key stakeholders across the entire ecosystem.

The digital twin had high fidelity structural modelling capabilities

Looking to the future, Akselos sees hydro as a key part of the renewable energy mix where the digital twin can add a lot of value. So Akselos is working to improve the feature set supporting operations and inspection in a very dynamic way. In addition the company is also looking for partners across the energy ecosystem to improve the offering further by leveraging an open standards based software platform with application programming interfaces. This will help provide the best and most integrated solutions for plant operators.

See the rest here:

Evolution of the digital twin - International Water Power - International Water Power and Dam Construction

By: Thomas Szocinski, Director of Vapor Intrusion,Land Science

Since the early 1970s, environmental professionals and health experts have called attention to contaminant vapor intrusion and have alerted the public to the serious health problems caused by poor indoor air quality.

Beginning with radon, the understanding that chemicals could enter occupied spaces and cause adverse health effects has steadily expanded over the ensuing decades. It is now widely known that hydrocarbons and industrial solvents are commonly found on brownfield properties where former industrial sites once operated many of which are now undergoing redevelopment within major metropolitan areas. Even in very low concentrations, these compounds can pose significant health issues, such as cancers and birth defects.

In order to safeguard human health, it is necessary to prevent vapors from entering occupied spaces, but it was not until recently that a vapor barrier was created specifically for that purpose.

Installation of a TerrShield barrier in downtown Sacramento as part of a large brownfield redevelopment project.REGENESISDespite the increased knowledge related to the problem, effective mitigation solutions still lagged behind. Historically, vapor barrier solutions were adopted directly from the waterproofing industry. These barriers were typically comprised of plastic sheets (e.g. polyethylene, polypropylene, geotextiles, etc.) bonded together by a spray-applied asphalt latex mixture containing styrene butadiene rubber (SBR).

Looking to create a better, safer and more long-term solution, scientists and engineers have been conducting intensive scientific research and development to create a first-of-its-kind suite of safe, cost-effective passive vapor barrier systems designed specifically for vapor intrusion. To do so, they examined two main components the base layer and spray-applied seal and reimagined them from the ground up.

To upgrade the base layer, vapor barrier scientists decided to incorporate aluminum, which is well known to be able to prevent the permeation of organic volatile compounds. By sandwiching it between layers of flexible polyethylene, they developed an easy-to-install base layer with chemical resistance 100 times higher than the simple waterproofing sheets composed of high-density polyethylene (HDPE) used in the past.

Installation of a TerrShield barrier in downtown Sacramento as part of a large brownfield redevelopment project.REGENESISTurning to the spray-applied seal, scientists focused on finding a replacement for the generic latex-asphalt spray components previously adapted from the waterproofing industry. These sprays historically included synthetic rubber components such as SBR which was hydrophobic and thus very effective at repelling water, but also had an unfortunate tendency to sorb industrial solvent and hydrocarbon contamination, concentrating them until they would eventually break through the membrane and potentially enter the indoor air space. Additionally, applicators of the SBR material would need to clean their equipment with petroleum hydrocarbon solvents which may cause even greater damage to the affected properties where it is applied.

After evaluating several potential options, scientists developed a Nitra-Core spray-applied barrier material, designed to replace SBR-based latex waterproofing material with nitrile, a key component of chemically resistant personal protective equipment used in handling hazardous materials (e.g. blue disposable gloves).

Studies comparing the chemical resistance of the new nitrile-advanced asphalt barrier material to generic latex-asphalt barriers determined the nitrile-advanced material to be 10X more resistant as a contaminant vapor barrier against the common industrial solvent contaminant trichloroethylene with the generic asphalt-latex material allowing 10X more contaminant vapor diffusion across the barrier.

Land Science, a division of REGENESIS

In addition to the increased chemical resistance, both the base layer and spray-applied seal were designed to be easily installed, thus reducing construction timelines and saving money. They were also built to withstand the stresses of a construction site both during installation and afterwards (i.e. foot traffic, heavy equipment, and the occasional dropped tool), thus preserving its structural integrity and maintaining its efficacy.

Instead of just repurposing inferior technology that was developed for a different industry, scientists and engineers understood that in order to effectively tackle the issue of vapor intrusion, they needed to approach the problem with a full understanding of its challenges to effectively meet the needs of the environmental professionals tasked with guarding human health. The ideal vapor barrier would be easy and quick to apply, physically tough, and chemically resistant.

Through the innovative use of barrier materials such as aluminum and nitrile, researchers have created a suite of barrier systems that meet all those criteria and, unlike the vapor intrusion solutions of the past, are specifically designed to handle the vapor intrusion problems of today.

Land Science, a division of REGENESISThomas Szocinski is the director of vapor intrusion for the Land Science division of REGENESIS, Inc.the global leader in advanced technologies for contaminated site remediation. Since 1994, REGENESIS has developed and commercialized a range of proven soil, groundwater and vapor intrusion products to treat a wide variety of contaminants. Land Science was established in 2008 to address the increasing occurrence and regulation of vapor intrusion. It provides scientifically proven solutions that offer environmental engineering firms and developers cost-effective and innovative vapor mitigation technologies for environmentally compromised properties. For more information visit:www.landsciencetech.com.

See original here:

Vapor Intrusion Mitigation Part 1: Reimagining the Next Evolution in Concrete Vapor Protection - ForConstructionPros.com