Korber, Fischer, Yoon and Theiler are members of a rarified team that the National Institute of Allergy and Infectious Diseases assembled in January 2021, drawing on experts from around the world who specialize in relevant research fields such as viruses, the immune system, vaccines, epidemiology, structural biology, bioinformatics, virus genetics, and evolution. The team is called SAVE, for SARS-CoV-2 Assessment of Viral Evolution.

As noted in the Nature paper, the authors state, “This effort was designed to provide a real-time risk assessment of SARS-CoV-2 variants potentially impacting transmission, virulence, and resistance to convalescent and vaccine-induced immunity. The SAVE program serves as a critical data-generating component of the United States Government SARS-CoV-2 Interagency Group to assess implications of SARS-CoV-2 variants on diagnostics, vaccines and therapeutics and for communicating public health risk.”

Broad model for rapid response

SAVE focuses on mutations in SARS-CoV-2 and emerging virus variants. But its members say the global collaborative concept “is a broad model for rapidly responding to evolving pathogens with pandemic potential.”

“Over the past two decades, we have witnessed the emergence/re-emergence of several RNA viruses, including West Nile virus, H1N1 influenza virus, chikungunya virus, Zika virus, SARS-CoV-1, MERS-CoV and Ebola virus, that have threatened global public health,” the paper’s summary states. “Developing collaborative programs between academic, industry and commercial partners is essential to respond to rapidly evolving viruses,” said Marciela DeGrace of NIAID, the paper’s lead author.

SAVE members represent 58 different research sites located in the United States and around the world. Members participate within three sub-groups:

• Early Detection and Analysis
• In Vitro — what they can learn using flasks, beakers and tubes
• In Vivo — what they can learn in animal models that mimic human disease

Early detection methods

Korber’s team was part of the Early Detection and Analysis team, where such high-impact work as the initial identification of mutations in the virus made waves in the scientific community before its capacity for mutation had been clearly understood and accepted.

The Nature paper notes, “The process is collaborative and iterative, with seven teams using independent models and methodologies to prioritize mutations and lineages as well as rank importance for downstream testing. While the focus is on human infections, the Early Detection group also monitors variants circulating in animal populations, such as mink and deer, since they represent a potential reservoir source.”

On a weekly basis, the SAVE Early Detection and Analysis team reviews downloads of SARS-CoV-2 genomes from the international initiative for sequence sharing, GISAID. They search for variant and co-variant signatures in the genomes, then divide the work into two approaches:

• one based on convergent evolution as the main signal for selection and functional impact of mutations (done byCambridge and Walter Reed Army Institute of Research teams)
• the other anchored on prevalence and growth patterns of mutations and defined lineages (the role ofLos Alamos, Icahn School of Medicine at Mount Sinai, J. Craig Venter Institute/Bacterial Viral Bioinformatic Resource Center, UC-Riverside and Broad Institute teams)

Highlights of Los Alamos impact

At Los Alamos, the Korber team identifies emergent mutational patterns within the SARS-CoV-2 spike protein to track newly emerging and expanding variants and determine transitions in global and regional sampling frequencies over time, which is the specialty area in which Los Alamos has made a huge impact.

They pay particular attention to mutations in parts of the spike protein known to be highly targeted by antibodies, or that might impact infectivity. They also systematically define the most commonly circulating form of each emerging variant of interest or concern against the backdrop of the continuously evolving virus.

“Identifying the emerging variants, and obtaining accurate sequences for those variants, required continued wrangling of burgeoning data,” said Theiler. “There are now close to 10 million SARS-CoV-2 sequences in GISAID. These sequences, however, are non-uniformly sampled, are often partial and some contain errors, and of course it is the newest variants that give the sequencers the most trouble.”

“The tools we developed, along with our colleagues on the LANL COVID-19 Viral Genome Analysis Pipeline (cov.lanl.gov), provided the infrastructure that enabled us to follow this pandemic though its various waves,” he added.

Korber noted that “by working with the SAVE Early Detection team, we were able to be part of a synergistic collaborative effort, where our results in terms of early detection could be cross-checked with those of others.”

She added, “The real beauty of being part of the larger SAVE project was the knowledge that our analysis pipeline could provide foundational support for the many experimental teams in SAVE, and that we could help the scientific community get the best version of newly emergent variants into their laboratories as quickly and accurately as possible. In this way the science needed to understand the immunological and virological characteristics of new variants was rapidly obtained, in time to help inform public health decisions.”

The findings, which are reported in the journal Science Advances, are based on an analysis of more than 630 billion words drawn from internet web pages, using artificial intelligence tools to measure what words mean based on how they are used by millions of individuals.

“Many forms of bias, such as the tendency to associate ‘science’ with men more than women, have been studied in the past, but there has been much less work on how we view a ‘person,’ ” says April Bailey, a postdoctoral researcher in New York University’s Department of Psychology and the lead author of the paper.

“Our findings show that even when using gender-neutral terms, we prioritize men over women,” adds co-author Adina Williams, a research scientist at Meta AI and a graduate of NYU’s doctoral program in linguistics.

Bias at such a foundational level — our word choices — is potentially consequential, the researchers note.

“Conceptions of ‘people’ form the basis of many societal decisions and policymaking,” observes Andrei Cimpian, a professor in New York University’s Department of Psychology and the senior author of the paper. “Because men and women are each about half of the species, prioritizing men in our collective idea of a ‘person’ creates inequity for women in decisions based on this idea.”

The research team examined words’ meanings by considering how they are used by individuals. Specifically, the team studied how we use words expressing the concept of “person” and its gender-specific counterparts, “woman” and “man.”

To test whether we’re likely to think of men more often than we are of women when writing about “people,” the team used artificial intelligence algorithms that learn the meaning of words based on how they are used, drawing from a language repository collected by the non-profit Common Crawl in May 2017. This repository included more than 630 billion mostly English-language words appearing on nearly three billion web pages.

The researchers considered how word meaning is related to word context and use. For example, if you hear, “Each morning, Joe boiled water in the balak for tea,” you might guess that “balak” means something similar to “kettle,” even though “balak” is unfamiliar, because the words alongside “balak” (“tea,” “boiled,” and “water”) also frequently co-occur with “kettle.”

In the Science Advances paper, the researchers investigated, in three studies, the meaning of “person” and related words (e.g., “people”) by taking into account adjacent words — the linguistic context.

In the first study, they compared the similarity in meaning (inferred via linguistic context) between words for people (e.g., “individual”) and words for men (e.g., “he” and “male”) to the similarity in meaning between words for people and words for women (e.g., “she” and “female”).

They found that words for people were used more similarly, and were thus more similar in their meaning, to words for men than to words for women — and by a statistically significant margin. Put another way, the collective concept “people” overlapped more with the concept “men” than with the concept “women” in the studied words.

In the second study, instead of focusing on words for people, the team examined words denoting features central to this concept — specifically, words for traits that commonly describe what people are like. They compared hundreds of trait words identified in past research as common descriptors of people (e.g., “extroverted,” “analytical,” and “superstitious”) to the same lists of words for women and for men from the paper’s initial study.

They found that the meaning of these descriptor words in the second study were, overall, more similar to the meaning of words for men than to the meaning of words for women, with a statistically significant difference between the two. That is, common words that describe what people are like (e.g., “extroverted”) are also used more similarly to words for men than to words for women.

In a third study, the researchers studied the use of verbs — a reasonable area for exploration given the initial findings. Specifically, if the collective concept “people” overlaps more with the concept “men” than with the concept “women,” then words that describe what people do and what is done to them (e.g., “love,” “annoy”) may also be more likely to be similar in their contextual meaning to words denoting men than to words denoting women.

In this study, they compared the meaning similarities between more than 250 verbs that describe actions that people take (e.g., “facilitate,” “smile,” and “threaten”) and words for men vs. words for women.

As with the second study, which focused on common words that describe what people are like, words that describe what people do (e.g., “run”) were also used more similarly to words for men than to words for women — a difference that was again statistically significant.