Two companion studies report new artificial intelligence approaches for identification of molecular “fingerprints” for specific infections. The methods, used here to derive a fingerprint for COVID-19, can advance the development of rapid diagnostics for many infectious diseases.
The first manuscript provides, for the first time, a framework for systematic quantification of the robustness and cross-reactivity of a potential group of blood biomarkers that act as a signature for infectious disease diagnosis. Robustness is the ability of a signature to detect a disease state (e.g., COVID-19) consistently in multiple independent groups. Cross-reactivity is the undesired extent to which a signature incorrectly detects an unintended infection or condition (e.g., a COVID-19 signature falsely detects an influenza infection). The framework was developed through curation and integration of a massive public data collection and application of a standardized signature scoring method. Applying the framework to published and synthetic signatures, the researchers demonstrate an inherent trade-off between robustness and cross-reactivity.
The second manuscript addresses the limitations of previous signature discovery approaches by modeling the robustness/cross-reactivity tradeoff with multi-objective optimization. The researchers apply this new method to identify a highly-specific blood-based signature for SARS-CoV-2 infection, which they validated in multiple independent cohorts. Interpretable signatures are more likely to be robust because they capture a reproducible biological process, such as antiviral response or cytokine signaling. Consonant with this insight, they show that the COVID-19 signature is interpretable as a combination of signals from two kinds of immune cells, plasmablasts (a rapidly proliferating, antibody-producing cell) and memory T cells (antigen-specific T cells that remain long after an infection has been eliminated). In analysis of single cell gene expression data, they found that plasmablasts mediate COVID-19 detection and memory T cells control against cross-reactivity of the signatures with other viral infections.
Standard tests for infection diagnosis involve a variety of technologies such as microbial cultures and PCR assays. These standard tests share a common design principle, which is to directly quantify pathogen material in patient samples. As a consequence, standard tests can have poor detection, particularly early after infection, before the pathogen replicates to detectable levels. For example, due to insufficient viral genetic material, PCR-based tests for SARS-CoV-2 may miss more than 60 percent of cases within the first few days of infection. To overcome these limitations, new tools for infection diagnosis are urgently needed.
The study was conducted by researchers at the Icahn School of Medicine at Mount Sinai, and Yale School of Medicine, and is published December 21 in Cell Systems.
Upon infection, an individual’s immune system rapidly responds. This response includes the transcription of many genes coding for proteins that help combat the pathogen. The activated group of genes serves as a ‘fingerprint’ for the particular infection and is referred to as a host response signature. Host response signatures to infection have emerged as an intense area of research due to their potential both to improve understanding of pathogenesis and to provide a new paradigm for diagnosis.
The performance and overall clinical applicability of host response signatures depend on two main properties, robustness and cross-reactivity. Robustness is the ability of a signature to detect a disease state (e.g., COVID-19) consistently in multiple independent cohorts. Cross-reactivity is the undesired extent to which a signature incorrectly detects an unintended infection or condition (e.g., a COVID-19 signature falsely detects an influenza infection). Achieving the full potential of host response signatures requires both gains in robustness and reduction in cross-reactivity.
The two manuscripts present fundamental advances in the field. The first study curates massive public data and establishes a framework for systematic candidate signature evaluation. The second study leverages this framework and develops a multi-objective optimization approach to identify a highly robust and not cross-reactive COVID-19 signature.
“As we show for COVID-19, the integration of the methods described in both studies helps provide the balance needed for having sensitivity of a diagnostic signature to detect infection while still limiting cross-reactivity with other infections and conditions," said Mount Sinai's Dr. Elena Zaslavsky. "These new methods can help address the need for more rapid diagnosis of infections so that proper treatment can be initiated earlier.”
Source: Mount Sinai Health System