Functional immune mapping with deep-learning enabled phenomics applied to immunomodulatory and COVID-19 drug discovery

Cuccarese, Michael F.; Earnshaw, Berton; Heiser, Katie; Fogelson, Ben; Davis, Chadwick T.; McLean, Peter F.; Gordon, Hannah; Skelly, Kathleen-Rose; Weathersby, Fiona L.; Rodic, Vlad; Quigley, Ian K.; Pastuzyn, Elissa D.; Mendivil, Brandon M.; Lazar, Nathan H.; Brooks, Carl A.; Carpenter, Joseph; Probst, Brandon L.; Jacobson, Pamela; Glazier, Seth W.; Ford, Jason C.; Jensen, James; Campbell, Nicholas D.; Statnick, Michael A.; Low, Adeline S.; Thomas, Kirk R.; Carpenter, Anne E.; Hegde, Sharath S.; Alfa, Ronald W.; Victors, Mason; Haque, Imran S.; Chong, Yolanda T.; Gibson, Christopher C.

doi:10.1101/2020.08.02.233064

Cited by 38 publications

(74 citation statements)

References 73 publications

(66 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Next, we applied RTG analysis to identify confounders in a dataset of biomedical image embeddings released in the public domain by Recursion Pharmaceuticals (Cuccarese et al2020). The raw images are of cell cultures in 1536-well plates.…”

Section: Resultsmentioning

confidence: 99%

“…Experiment ID, on the other hand, confers significant structure to the data (RTG 0.87), suggesting some difference in conditions between experimental groups. Each experiment was a different “batch” of data (Cuccarese et al 2020), suggesting the presence of confounding effects by batch.…”

Section: Resultsmentioning

confidence: 99%

“…The mean change due to donor debiasing was significantly different from the mean change due to clone debiasing (p < 10 -4 via resampling). Identification of confounding effects in a hierarchical biomedical dataset Next, we applied RTG analysis to identify confounders in a dataset of biomedical image embeddings released in the public domain by Recursion Pharmaceuticals (Cuccarese et al 2020). The raw images are of cell cultures in 1536-well plates.…”

Section: Model Debiasing Examplementioning

confidence: 99%

“…Such conflicting results may impede discovery of the importance of subedge if using linear confounder analysis. Of note, leave-experiment-out cross validation also results in negative values (next to bottom row), but, unlike for subedge, the raw cross validated for experiment is well above zero (0.20, first value (Cuccarese et al 2020). a. RTG scores for potential confounders: experiment, plate, well position, imaging site, unique well, and whether or not a data point is from a subedge well.…”

Section: Model Debiasing Examplementioning

confidence: 99%

“…We then apply our approach to two real-world datasets. First, we analyze a large public dataset of images of cultured cells (Cuccarese et al 2020) and reveal that some features of the experimental design strongly bias the results. We furthermore show that linear techniques fail to discover these confounding effects.…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

Hierarchical confounder discovery in the experiment–machine learning cycle

Rogozhnikov¹,

Ramkumar²,

Bedi³

et al. 2021

Preprint

View full text Add to dashboard Cite

The promise of using machine learning (ML) to extract scientific insights from high dimensional datasets is tempered by the frequent presence of confounding variables, and it behooves scientists to determine whether or not a model has extracted the desired information or instead may have fallen prey to bias. Due both to features of many natural phenomena and to practical constraints of experimental design, complex bioscience datasets tend to be organized in nested hierarchies which can obfuscate the origin of a confounding effect and obviate traditional methods of confounder amelioration. We propose a simple non-parametric statistical method called the Rank-to-Group (RTG) score that can identify hierarchical confounder effects in raw data and ML-derived data embeddings. We show that RTG scores correctly assign the effects of hierarchical confounders in cases where linear methods such as regression fail. In a large public biomedical image dataset, we discover unreported effects of experimental design. We then use RTG scores to discover cross-modal correlated variability in a complex multi-phenotypic biological dataset. This approach should be of general use in experiment–analysis cycles and to ensure confounder robustness in ML models.

show abstract

Section: Resultsmentioning

confidence: 99%

Section: Resultsmentioning

confidence: 99%

Section: Model Debiasing Examplementioning

confidence: 99%

Section: Model Debiasing Examplementioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

Hierarchical confounder discovery in the experiment–machine learning cycle

Rogozhnikov¹,

Ramkumar²,

Bedi³

et al. 2021

Preprint

View full text Add to dashboard Cite

show abstract

An Enantioselective Aminocatalytic Cascade Reaction Affording Bioactive Hexahydroazulene Scaffolds

Faghtmann,

Eugui,

Nygaard Lamhauge

et al. 2024

Chemistry A European J

View full text Add to dashboard Cite

A novel cascade reaction initiated by an enantioselective aminocatalysed 1,3‐dipolar [6+4] cycloaddition between catalytically generated trienamines and 3‐oxidopyridinium betaines is presented. The [6+4] cycloadduct spontaneously undergoes an intramolecular enamine‐mediated aldol, hydrolysis, and E1cb sequence, which ultimately affords a chiral hexahydroazulene framework. In this process, three new C–C bonds and three new stereocenters are formed, enabled by a formal unfolding of the pyridine moiety from the dipolar reagent. The hexahydroazulenes are formed with excellent diastereo‐, regio‐ and periselectivity (>20:1), up to 96% ee, and yields up to 52%. Synthetic elaborations of this scaffold were performed, providing access to a variety of functionalised hydroazulene compounds, of which some were found to display biological activity in U‐2OS osteosarcoma cells in cell painting assays.

show abstract

Nucleoside analog GS‐441524: pharmacokinetics in different species, safety, and potential effectiveness against Covid‐19

Rasmussen

Thomsen

Hansen

2022

Pharmacology Res & Perspec

View full text Add to dashboard Cite

GS‐441524, the parent nucleoside of remdesivir, has been proposed to be effective against Covid‐19 based on in vitro studies and studies in animals. However, randomized clinical trials of the agent to treat Covid‐19 have not been conducted. Here, we evaluated GS‐441524 for Covid‐19 treatment based on studies reporting pharmacokinetic parameters of the agent in mice, rats, cats, dogs, monkeys, and the single individual in the first‐in‐human trial supplemented with information about its activity against severe acute respiratory syndrome coronavirus 2 and safety. A dosing interval of 8 h was considered clinically relevant and used to calculate steady‐state plasma concentrations of GS‐441524. These ranged from 0.27 to 234.41 μM, reflecting differences in species, doses, and administration routes. Fifty percent maximal inhibitory concentrations of GS‐441524 against severe acute respiratory syndrome coronavirus 2 ranged from 0.08 μM to above 10 μM with a median of 0.87 μM whereas concentrations required to produce 90% of the maximal inhibition of the virus varied from 0.18 µM to more than 20 µM with a median of 1.42 µM in the collected data. Most of these concentrations were substantially lower than the calculated steady‐state plasma concentrations of the agent. Plasma exposures to orally administered GS‐441524, calculated after normalization of doses, were larger for dogs, mice, and rats than cynomolgus monkeys and humans, probably reflecting interspecies differences in oral uptake with reported oral bioavailabilities below 8.0% in cynomolgus monkeys and values as high as 92% in dogs. Reported oral bioavailabilities in rodents ranged from 12% to 57%. Using different presumptions, we estimated human oral bioavailability of GS‐441524 at 13% and 20%. Importantly, doses of GS‐441524 lower than the 13 mg/kg dose used in the first‐in‐human trial may be effective against Covid‐19. Also, GS‐441524 appears to be well‐tolerated. In conclusion, GS‐441524 has potential for oral treatment of Covid‐19.

show abstract

Functional immune mapping with deep-learning enabled phenomics applied to immunomodulatory and COVID-19 drug discovery

Cited by 38 publications

References 73 publications

Hierarchical confounder discovery in the experiment–machine learning cycle

Hierarchical confounder discovery in the experiment–machine learning cycle

An Enantioselective Aminocatalytic Cascade Reaction Affording Bioactive Hexahydroazulene Scaffolds

Nucleoside analog GS‐441524: pharmacokinetics in different species, safety, and potential effectiveness against Covid‐19

Contact Info

Product

Resources

About