Biology subjects

Sterling, S.

Publications and source records attributed to Sterling, S..

7 recordsLinked to original sources

bioRxiv2026

Vision-Language Models for Image-Based Dietary Assessment: A Benchmark of Accuracy, Cost, and Prompt Strategies Across Ten Models

BackgroundDietary assessment is the cornerstone of clinical management and research studies evaluating diet and health. Traditional methods such as food diaries and 24-hour recalls can be burdensome, prone to recall bias, and difficult to adhere to. Image-based dietary assessment using vision-language models (VLMs) offers a potential solution. ObjectiveOur goal was to benchmark state-of-the-art VLMs for automated food recognition, weight estimation, and calorie estimation using Googles Nutrition5k dataset. MethodsWe evaluated 3,229 food images using ten approaches: proprietary VLMs (Gemini 2.0 Flash, 2.5 Flash, 3.0 Flash, and 3.1 Flash Lite; GPT- 4o, GPT-4o-mini, and GPT-5 Mini; and Claude Haiku 4.5), an open-source VLM (Qwen2-VL-7B), and a commercial food recognition API (FatSecret). We assessed calorie and weight estimation using Lins Concordance Correlation Coefficient (CCC) and component detection using Jaccard similarity. ResultsGemini 3.0 Flash achieved the best calorie estimation (CCC 0.767, MAE 80.7 kcal), while Gemini 3.1 Flash Lite offered very comparable accuracy (CCC 0.754) with the highest ingredient recognition (Jaccard 0.655) at the lowest cost among top-performing models ($0.59/1K images). Among earlier-generation models, Gemini 2.0 Flash remained competitive (CCC 0.742, Jaccard 0.621) at a fraction of the cost ($0.10/1K images). A human validation study in which four annotators reviewed 440 images revealed systematic omissions in the original Nutrition5k labels. After correction, the extrapolated ingredient-overlap score for Gemini 2.0 Flash increased from 0.62 to an estimated 0.82, suggesting that raw Jaccard scores substantially underestimate true model performance. ConclusionsCurrent VLMs can perform automated dietary assessment with reasonable accuracy from single overhead photographs. Our results inform model selection for dietary assessment applications and highlight remaining challenges in calorie estimation and component detection for complex, multi-item meals.

bioinformatics↗

bioRxiv2025

A multi-scale segmentation-free self-supervised AI model to characterize the heterogeneity of the brain tumor microenvironment

Brain tumors affect about 1 million people in the U.S., with aggressive types like glioblastoma having very low survival rates due to complex tumor biology and the protective blood-brain barrier. Current treatments are limited in effectiveness, and our understanding of brain tumor biology remains incomplete. High dimensional multiplexed imaging has enabled us to better understand the tumor microenvironment (TME); however, analyses typically rely on cell segmentation, which is error-prone, may discard useful context outside the cell boundary, and neglects complex tissue-wide features. To address this limitation, we developed a segmentation-free, self-supervised representation learning framework that enables us to train directly on multiplexed images using masked image modeling. We used this approach to analyze 389 imaging mass cytometry images from 185 brain tumor patients. To study tissue-wide features, we first trained our model on 64x64 micron tiles capturing neighborhoods of 10-20 cells, which we termed local tumor microenvironments (LTMEs). To further characterize these LTMEs, we trained our model on 16x16 micron tiles centered on individual cells in our dataset, so that each tile captures a single cell and its surrounding area, which we termed single-cell microenvironments (SCMEs). This multi-scale, self-supervised approach enables a detailed analysis of the heterogeneity within the brain TME, examining single cells in their spatial context. In addition to validating known findings, we identified a novel LTME in GBM patients, composed primarily of tumor cells and a few B and T cells, which strongly correlated with increased survival. By analyzing these B cells with our SCME model, we found they were distinct from other GBM B cells, and higher concentrations of these B cells were linked to improved survival. In conclusion, our study introduces a multi-scale, segmentation-free, self-supervised machine learning model that provides unprecedented insights into brain TMEs, enabling discovery of previously unrecognized cell interactions and spatial features that are predictive of patient survival.

cancer biology↗

bioRxiv2023

Endothelial SARS-CoV-2 infection is not the underlying cause of COVID19-associated vascular pathology in mice

Endothelial damage and vascular pathology have been recognized as major features of COVID-19 since the beginning of the pandemic. Two main theories regarding how Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-CoV-2) damages endothelial cells and causes vascular pathology have been proposed: direct viral infection of endothelial cells or indirect damage mediated by circulating inflammatory molecules and immune mechanisms. However, these proposed mechanisms remain largely untested in vivo. Here, we utilized a set of new mouse genetic tools1 developed in our lab to test both the necessity and sufficiency of endothelial human angiotensin-converting enzyme 2 (hACE2) in COVID19 pathogenesis. Our results demonstrate that endothelial ACE2 and direct infection of vascular endothelial cells does not contribute significantly to the diverse vascular pathology associated with COVID-19.

pathology↗

bioRxiv2021

SARS-CoV-2 infection of olfactory epithelial cells and neurons drives acute lung injury and lethal COVID-19 in mice

Lethal COVID-19 is associated with respiratory failure that is thought to be caused by acute respiratory distress syndrome (ARDS) secondary to pulmonary infection. To date, the cellular pathogenesis has been inferred from studies describing the expression of ACE2, a transmembrane protein required for SARS-CoV-2 infection, and detection of viral RNA or protein in infected humans, model animals, and cultured cells. To functionally test the cellular mechanisms of COVID-19, we generated hACE2fl animals in which human ACE2 (hACE2) is expressed from the mouse Ace2 locus in a manner that permits cell-specific, Cre-mediated loss of function. hACE2fl animals developed lethal weight loss and hypoxemia within 7 days of exposure to SARS-CoV-2 that was associated with pulmonary infiltrates, intravascular thrombosis and patchy viral infection of lung epithelial cells. Deletion of hACE2 in lung epithelial cells prevented viral infection of the lung, but not weight loss, hypoxemia or death. Inhalation of SARS-CoV-2 by hACE2fl animals resulted in early infection of sustentacular cells with subsequent infection of neurons in the neighboring olfactory bulb and cerebral cortex-- events that did not require lung epithelial cell infection. Pharmacologic ablation of the olfactory epithelium or Foxg1Cre mediated deletion of hACE2 in olfactory epithelial cells and neurons prevented lethality and neuronal infection following SARS-CoV-2 infection. Conversely, transgenic expression of hACE2 specifically in olfactory epithelial cells and neurons in Foxg1Cre; LSL-hACE2 mice was sufficient to confer neuronal infection associated with respiratory failure and death. These studies establish mouse loss and gain of function genetic models with which to genetically dissect viral-host interactions and demonstrate that lethal disease due to respiratory failure may arise from extrapulmonary infection of the olfactory epithelium and brain. Future therapeutic efforts focused on preventing olfactory epithelial infection may be an effective means of protecting against severe COVID-19.

microbiology↗

bioRxiv2021

Novel Hendra virus variant detected by sentinel surveillance of Australian horses.

A novel Hendra virus (HeV) variant, not detected by routine testing, was identified and isolated from a Queensland horse that suffered acute, fatal disease consistent with HeV infection. Whole genome sequencing and phylogenetic analysis demonstrated the variant to have ~83% nucleotide identity to the prototype HeV strain. An updated RT-qPCR assay was designed for routine HeV surveillance. In silico and in vitro comparison of the receptor-binding protein with prototypic HeV showed that the human monoclonal antibody m102.4 used for post-exposure prophylaxis, as well as the current equine vaccine, should be effective against this variant. Genetic similarity of this virus to sequences detected from grey-headed flying-foxes suggests the variant circulates at-least in this species. Studies determining infection kinetics, pathogenicity, reservoir-species associations, viral-host co-evolution and spillover dynamics for this virus are urgently needed. Surveillance and biosecurity practices should be updated to appreciate HeV spillover risk across all regions frequented by flying foxes.

molecular biology↗

bioRxiv2021

Structural basis for enhanced infectivity and immune evasion of SARS-CoV-2 variants

Several fast-spreading variants of severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) have become the dominant circulating strains that continue to fuel the COVID-19 pandemic despite intensive vaccination efforts throughout the world. We report here cryo-EM structures of the full-length spike (S) trimers of the B.1.1.7 and B.1.351 variants, as well as their biochemical and antigenic properties. Mutations in the B.1.1.7 protein increase the accessibility of its receptor binding domain and also the binding affinity for receptor angiotensin-converting enzyme 2 (ACE2). The enhanced receptor engagement can account for the increased transmissibility and risk of mortality as the variant may begin to infect efficiently infect additional cell types expressing low levels of ACE2. The B.1.351 variant has evolved to reshape antigenic surfaces of the major neutralizing sites on the S protein, rendering complete resistance to some potent neutralizing antibodies. These findings provide structural details on how the wide spread of SARS-CoV-2 enables rapid evolution to enhance viral fitness and immune evasion. They may guide intervention strategies to control the pandemic.

biochemistry↗

bioRxiv2017

Diverse functions of closely homologous actin isoforms are defined by their nucleotide, rather than their amino acid sequence.

{beta}- and {gamma}-cytoplasmic-actin are nearly indistinguishable in their amino acid sequence, but are encoded by different genes that play non-redundant biological roles. The key determinants that drive their functional distinction are unknown. Here we tested the hypothesis that {beta}- and {gamma}-actin functions are defined by their nucleotide, rather than their amino acid sequence, using targeted editing of the mouse genome. Although previous studies have shown that disruption of {beta}-actin gene critically impacts cell migration and mouse embryogenesis, we demonstrate here that generation of a mouse lacking {beta}-actin protein by editing {beta}-actin gene to encode {gamma}-actin protein, and vice versa, does not affect cell migration and/or organism survival. Our data suggest that the essential in vivo function of {beta}-actin is provided by the gene sequence independent of the encoded protein isoform. We propose that this regulation constitutes a global \"silent code\" mechanism that controls the functional diversity of protein isoforms.

cell biology↗