Journal: Aging Cell

Article Title: Targeting Hyperoxia‐Induced Cellular Senescence in Developing Human Airway Cells: Senomorphics Versus Senolytics Versus Antioxidants

doi: 10.1111/acel.70538

Figure Lengend Snippet: Hyperoxia‐induced senescent fASM secrete higher levels of SASP. Cells cultured till day 7 in normoxic (21% O 2 ) or hyperoxic (50% O 2 ) environment were incubated with fresh growth media in normoxia for 24 h, and the supernatants were collected. Samples were then analyzed for SASP secretion by Eve Technologies Corporation (Calgary, Alberta, Canada) using Luminex xMAP and the Human Cytokine/Chemokine 96‐Plex Discovery Assay Array (HD96) panel. (A) Out of the 96 proteins, 80 were detected and visualized using a heatmap. (B) A detailed analysis of each detected marker is shown as fold‐change normalized to control. For statistical analysis, 2‐way ANOVA with a two‐stage linear step‐up procedure of Benjamini, Krieger and Yekutieli test, with individual variances computed for each comparison was applied. False discovery rate < 0.05 was considered significant. Data are presented as box plot of n = 7 cell lines per group.

Article Snippet: Supernatants collected at day 8 from normoxia and hyperoxia‐exposed fASM were analyzed for SASP secretion by Eve Technologies Corporation (Calgary, Alberta, Canada) using the Human Cytokine/Chemokine 96‐Plex Discovery Assay Array (HD96) panel.

Techniques: Cell Culture, Incubation, Luminex, Marker, Control, Comparison

Journal: bioRxiv

Article Title: Transforming the cytokine literature into a resource for experimental analysis and discovery

doi: 10.64898/2026.05.04.722753

Figure Lengend Snippet: a , Overview of the construction workflow for the Cytokine Effect Database (CytED) for a single example row. b , Cytokine names were standardized using a manually curated list of target cytokines. The bar graph shows the cytokine counts from 110,000 papers for the most common cytokines. c , The most commonly mapped cell types for each target ontology. d , Initial cytokine effect entries are processed into one or, if the initial entry mixes multiple concepts, multiple output entries containing genes, pathways, and cell processes. The bar graph on the right shows precision and recall of DeepSeek-R1 for gene-level statements for a test set of 52 manually processed input rows (117 output rows). e , Distribution of literature counts for mapped cell process categories. f , The initial parsed data undergoes quality control by a DeepSeek-R1-powered comparison between the original text chunk and the extracted table entries. Incorrect entries are removed from the table. g . Table entries pre- and post-QC were evaluated for the presence of clear errors by GPT-5.2 (n=1000 and n=2000, respectively). Incorrect entries were then blindly re-rated by a human evaluator. h , The filtered table was re-evaluated according to the same QC prompt as in (f) by GPT-5. The resulting entries were sampled to increase the expected error rate and presented to human evaluators.

Article Snippet: Here, we present the Cytokine Effect Database (CytED), a repository of 1,012,726 statements of the form ‘cytokine x has effect y on cell type z’ extracted from 110,000 publications using multi-step processing by an LLM reasoning model (DeepSeek-R1, ) .

Techniques: Control, Comparison

Journal: bioRxiv

Article Title: Transforming the cytokine literature into a resource for experimental analysis and discovery

doi: 10.64898/2026.05.04.722753

Figure Lengend Snippet: a , Publication count for some of the most common cytokine-cell type-gene regulation statements. b , Publication counts for the most common cytokine-cell type-cell effect statements. c , Top cytokines that activate chemotaxis in dendritic cells. d , Up- and downregulation counts for the most common cytokine-cell type-gene regulation statements with a low directional agreement. e , The link between CytED and literature fragments was used to formulate potential explanations of a divergent distribution of directions for the effect of IL-6 on FOXP3 expression in Tregs between cited and original statements using LLM queries. f , Downstream cytokines upregulated by IL-6 that upregulate FOXP3 or activate the Treg differentiation process. The shading of each connection indicates log2+1 of relative literature counts and thereby the prominence of a connection. g . UMAP of different cytokines for log10+1-transformed net literature counts of merged cell type-cell effect and cell type-gene relationships after truncated singular value decomposition. Functionally similar cytokines group together. h . Literature weights of different cell processes and genes for the effect of (left) IL-6 and IL-10 in macrophages (MΦ) or T cells and (right) IL-6 and IL-10 in fibroblasts and CD4+ αβ T cells. The Pearson correlation r is shown as an inset.

Article Snippet: Here, we present the Cytokine Effect Database (CytED), a repository of 1,012,726 statements of the form ‘cytokine x has effect y on cell type z’ extracted from 110,000 publications using multi-step processing by an LLM reasoning model (DeepSeek-R1, ) .

Techniques: Chemotaxis Assay, Expressing, Transformation Assay

Journal: bioRxiv

Article Title: Transforming the cytokine literature into a resource for experimental analysis and discovery

doi: 10.64898/2026.05.04.722753

Figure Lengend Snippet: a , CytED entries and transcriptional data are merged by matching cytokine-cell type-gene triplets. b , A high-throughput in vitro screen of human PBMCs response to treatment with 90 cytokines is used as a primary test case . The left bar graph shows the agreement between CytED statements and the experimental data. The right bar graph shows agreement after shuffling cytokine labels within each cell type as a control. c , Literature-experiment agreement by triple confidence derived from literature abundance. d , Workflow to analyze experimental changes in key genes, illustrated for IL-10 stimulation of CD16+ monocytes. For each cell type, a subset of experimental changes agrees or disagrees with CytED entries. Key literature genes for IL-10 stimulation of CD16+ monocytes are shown in the inset. Each triple can be traced back to specific places in origin publications. The experimental results and corresponding text passages are provided to an LLM to generate contextual summaries. Summaries across all key genes are then combined to produce an overall interpretation of the experimental response. e , Same analysis as in (d) for IL-10 stimulation of CD8+ T cells. The distribution of experimental time points for statements is shown on the bottom right. f , Dependence of the agreement (same-opposite) between experimental data and CytED on metadata variables: experimental system (human studies only), species (in vitro studies only), measurement time point, or experimental readout. Statistical significance was determined via a generalized Cochran-Mantel-Haenszel (CMH) test stratified by cytokine identity relative to the reference group (ref.), with the weighted mean score difference (CMHΔ) as the effect size. Multiple testing was corrected using the Benjamini-Hochberg procedure. g , Analysis as in (d) and (f) for a 4-hour in vivo cytokine stimulation experiment in mice. General agreement is much lower, at least partially due to higher noise levels in the data. Overall, we find that literature entries from long term experiments are overwhelmingly less likely to agree.

Article Snippet: Here, we present the Cytokine Effect Database (CytED), a repository of 1,012,726 statements of the form ‘cytokine x has effect y on cell type z’ extracted from 110,000 publications using multi-step processing by an LLM reasoning model (DeepSeek-R1, ) .

Techniques: High Throughput Screening Assay, In Vitro, Control, Derivative Assay, In Vivo

Journal: bioRxiv

Article Title: Transforming the cytokine literature into a resource for experimental analysis and discovery

doi: 10.64898/2026.05.04.722753

Figure Lengend Snippet: a , Count distributions of cytokine-cell type-gene pairs are converted to weighted gene sets to infer cytokine signaling activities via a univariate linear model. b , Ranking of the on-target cytokine for pathway-derived, CytoSig, and CytED-derived weighted gene sets in a subset of perturbation conditions for human PBMC data. c , Distribution of ranks as in (b) for both human PBMC and in vivo mouse cytokine perturbation data. The p-value is calculated using a two-sided Wilcoxon signed-rank test on paired on-target ranks across datasets. d , Gene sets for specific cell processes are generated by identifying genes that co-occur in the same origin row. Co-occurring gene counts across CytED then yield a gene signature for a given process, here shown for cytotoxic effector signaling. Manually curated cytotoxicity genes (CD8A, GZMA, GZMB, GZMH, GZMK, PRF1, NKG7) are highlighted in red. e , Spearman correlation between inferred cytotoxic effector function process activity and the activities of different cytokines in melanoma samples from the TCGM dataset. f , Perturbation experiments measure a mixture of primary and secondary effects. Using CytED, one can infer which effects are likely primary, which secondary, and which might be novel cytokine effects. g , Experimental results of IL-12-cell-cytokine regulation connections in the human PBMC dataset. Each connection indicates a connection with log2FC>1 and padj<0.05. h, i , Inference of primary and secondary effects of (h) IL-12 and (i) IL-1 based on CytED literature annotations. The inset shows how many connections are explained by the primary cytokine, secondary cytokines, or are not annotated in CytED for any of the cytokine perturbations. j , CytED-based cytokine activity analysis for IL-12 and IL-1 stimulation.

Article Snippet: Here, we present the Cytokine Effect Database (CytED), a repository of 1,012,726 statements of the form ‘cytokine x has effect y on cell type z’ extracted from 110,000 publications using multi-step processing by an LLM reasoning model (DeepSeek-R1, ) .

Techniques: Derivative Assay, In Vivo, Generated, Activity Assay

Journal: bioRxiv

Article Title: Transforming the cytokine literature into a resource for experimental analysis and discovery

doi: 10.64898/2026.05.04.722753

Figure Lengend Snippet: a , Overview of the construction workflow for the Cytokine Effect Database (CytED) for a single example row. b , Cytokine names were standardized using a manually curated list of target cytokines. The bar graph shows the cytokine counts from 110,000 papers for the most common cytokines. c , The most commonly mapped cell types for each target ontology. d , Initial cytokine effect entries are processed into one or, if the initial entry mixes multiple concepts, multiple output entries containing genes, pathways, and cell processes. The bar graph on the right shows precision and recall of DeepSeek-R1 for gene-level statements for a test set of 52 manually processed input rows (117 output rows). e , Distribution of literature counts for mapped cell process categories. f , The initial parsed data undergoes quality control by a DeepSeek-R1-powered comparison between the original text chunk and the extracted table entries. Incorrect entries are removed from the table. g . Table entries pre- and post-QC were evaluated for the presence of clear errors by GPT-5.2 (n=1000 and n=2000, respectively). Incorrect entries were then blindly re-rated by a human evaluator. h , The filtered table was re-evaluated according to the same QC prompt as in (f) by GPT-5. The resulting entries were sampled to increase the expected error rate and presented to human evaluators.

Article Snippet: For the quadruple cytokine combinations, around 14% have at least some minimal support in CytED, as compared to an expected value of <1% for random sampling on top of existing combinations.

Techniques: Control, Comparison

Journal: bioRxiv

Article Title: Transforming the cytokine literature into a resource for experimental analysis and discovery

doi: 10.64898/2026.05.04.722753

Figure Lengend Snippet: a , Publication count for some of the most common cytokine-cell type-gene regulation statements. b , Publication counts for the most common cytokine-cell type-cell effect statements. c , Top cytokines that activate chemotaxis in dendritic cells. d , Up- and downregulation counts for the most common cytokine-cell type-gene regulation statements with a low directional agreement. e , The link between CytED and literature fragments was used to formulate potential explanations of a divergent distribution of directions for the effect of IL-6 on FOXP3 expression in Tregs between cited and original statements using LLM queries. f , Downstream cytokines upregulated by IL-6 that upregulate FOXP3 or activate the Treg differentiation process. The shading of each connection indicates log2+1 of relative literature counts and thereby the prominence of a connection. g . UMAP of different cytokines for log10+1-transformed net literature counts of merged cell type-cell effect and cell type-gene relationships after truncated singular value decomposition. Functionally similar cytokines group together. h . Literature weights of different cell processes and genes for the effect of (left) IL-6 and IL-10 in macrophages (MΦ) or T cells and (right) IL-6 and IL-10 in fibroblasts and CD4+ αβ T cells. The Pearson correlation r is shown as an inset.

Article Snippet: For the quadruple cytokine combinations, around 14% have at least some minimal support in CytED, as compared to an expected value of <1% for random sampling on top of existing combinations.

Techniques: Chemotaxis Assay, Expressing, Transformation Assay

Journal: bioRxiv

Article Title: Transforming the cytokine literature into a resource for experimental analysis and discovery

doi: 10.64898/2026.05.04.722753

Figure Lengend Snippet: a , CytED entries and transcriptional data are merged by matching cytokine-cell type-gene triplets. b , A high-throughput in vitro screen of human PBMCs response to treatment with 90 cytokines is used as a primary test case . The left bar graph shows the agreement between CytED statements and the experimental data. The right bar graph shows agreement after shuffling cytokine labels within each cell type as a control. c , Literature-experiment agreement by triple confidence derived from literature abundance. d , Workflow to analyze experimental changes in key genes, illustrated for IL-10 stimulation of CD16+ monocytes. For each cell type, a subset of experimental changes agrees or disagrees with CytED entries. Key literature genes for IL-10 stimulation of CD16+ monocytes are shown in the inset. Each triple can be traced back to specific places in origin publications. The experimental results and corresponding text passages are provided to an LLM to generate contextual summaries. Summaries across all key genes are then combined to produce an overall interpretation of the experimental response. e , Same analysis as in (d) for IL-10 stimulation of CD8+ T cells. The distribution of experimental time points for statements is shown on the bottom right. f , Dependence of the agreement (same-opposite) between experimental data and CytED on metadata variables: experimental system (human studies only), species (in vitro studies only), measurement time point, or experimental readout. Statistical significance was determined via a generalized Cochran-Mantel-Haenszel (CMH) test stratified by cytokine identity relative to the reference group (ref.), with the weighted mean score difference (CMHΔ) as the effect size. Multiple testing was corrected using the Benjamini-Hochberg procedure. g , Analysis as in (d) and (f) for a 4-hour in vivo cytokine stimulation experiment in mice. General agreement is much lower, at least partially due to higher noise levels in the data. Overall, we find that literature entries from long term experiments are overwhelmingly less likely to agree.

Article Snippet: For the quadruple cytokine combinations, around 14% have at least some minimal support in CytED, as compared to an expected value of <1% for random sampling on top of existing combinations.

Techniques: High Throughput Screening Assay, In Vitro, Control, Derivative Assay, In Vivo

Journal: bioRxiv

Article Title: Transforming the cytokine literature into a resource for experimental analysis and discovery

doi: 10.64898/2026.05.04.722753

Figure Lengend Snippet: a , Count distributions of cytokine-cell type-gene pairs are converted to weighted gene sets to infer cytokine signaling activities via a univariate linear model. b , Ranking of the on-target cytokine for pathway-derived, CytoSig, and CytED-derived weighted gene sets in a subset of perturbation conditions for human PBMC data. c , Distribution of ranks as in (b) for both human PBMC and in vivo mouse cytokine perturbation data. The p-value is calculated using a two-sided Wilcoxon signed-rank test on paired on-target ranks across datasets. d , Gene sets for specific cell processes are generated by identifying genes that co-occur in the same origin row. Co-occurring gene counts across CytED then yield a gene signature for a given process, here shown for cytotoxic effector signaling. Manually curated cytotoxicity genes (CD8A, GZMA, GZMB, GZMH, GZMK, PRF1, NKG7) are highlighted in red. e , Spearman correlation between inferred cytotoxic effector function process activity and the activities of different cytokines in melanoma samples from the TCGM dataset. f , Perturbation experiments measure a mixture of primary and secondary effects. Using CytED, one can infer which effects are likely primary, which secondary, and which might be novel cytokine effects. g , Experimental results of IL-12-cell-cytokine regulation connections in the human PBMC dataset. Each connection indicates a connection with log2FC>1 and padj<0.05. h, i , Inference of primary and secondary effects of (h) IL-12 and (i) IL-1 based on CytED literature annotations. The inset shows how many connections are explained by the primary cytokine, secondary cytokines, or are not annotated in CytED for any of the cytokine perturbations. j , CytED-based cytokine activity analysis for IL-12 and IL-1 stimulation.

Article Snippet: For the quadruple cytokine combinations, around 14% have at least some minimal support in CytED, as compared to an expected value of <1% for random sampling on top of existing combinations.

Techniques: Derivative Assay, In Vivo, Generated, Activity Assay

Journal: bioRxiv

Article Title: Transforming the cytokine literature into a resource for experimental analysis and discovery

doi: 10.64898/2026.05.04.722753

Figure Lengend Snippet: a , Overview of the construction workflow for the Cytokine Effect Database (CytED) for a single example row. b , Cytokine names were standardized using a manually curated list of target cytokines. The bar graph shows the cytokine counts from 110,000 papers for the most common cytokines. c , The most commonly mapped cell types for each target ontology. d , Initial cytokine effect entries are processed into one or, if the initial entry mixes multiple concepts, multiple output entries containing genes, pathways, and cell processes. The bar graph on the right shows precision and recall of DeepSeek-R1 for gene-level statements for a test set of 52 manually processed input rows (117 output rows). e , Distribution of literature counts for mapped cell process categories. f , The initial parsed data undergoes quality control by a DeepSeek-R1-powered comparison between the original text chunk and the extracted table entries. Incorrect entries are removed from the table. g . Table entries pre- and post-QC were evaluated for the presence of clear errors by GPT-5.2 (n=1000 and n=2000, respectively). Incorrect entries were then blindly re-rated by a human evaluator. h , The filtered table was re-evaluated according to the same QC prompt as in (f) by GPT-5. The resulting entries were sampled to increase the expected error rate and presented to human evaluators.

Article Snippet: CytED also contains gene sets for many cytokines that are currently missing in CytoSig (∼124 in CytED for non-combinatorial signatures with at least 20 different genes versus 51 in CytoSig).

Techniques: Control, Comparison

Journal: bioRxiv

Article Title: Transforming the cytokine literature into a resource for experimental analysis and discovery

doi: 10.64898/2026.05.04.722753

Figure Lengend Snippet: a , Publication count for some of the most common cytokine-cell type-gene regulation statements. b , Publication counts for the most common cytokine-cell type-cell effect statements. c , Top cytokines that activate chemotaxis in dendritic cells. d , Up- and downregulation counts for the most common cytokine-cell type-gene regulation statements with a low directional agreement. e , The link between CytED and literature fragments was used to formulate potential explanations of a divergent distribution of directions for the effect of IL-6 on FOXP3 expression in Tregs between cited and original statements using LLM queries. f , Downstream cytokines upregulated by IL-6 that upregulate FOXP3 or activate the Treg differentiation process. The shading of each connection indicates log2+1 of relative literature counts and thereby the prominence of a connection. g . UMAP of different cytokines for log10+1-transformed net literature counts of merged cell type-cell effect and cell type-gene relationships after truncated singular value decomposition. Functionally similar cytokines group together. h . Literature weights of different cell processes and genes for the effect of (left) IL-6 and IL-10 in macrophages (MΦ) or T cells and (right) IL-6 and IL-10 in fibroblasts and CD4+ αβ T cells. The Pearson correlation r is shown as an inset.

Article Snippet: CytED also contains gene sets for many cytokines that are currently missing in CytoSig (∼124 in CytED for non-combinatorial signatures with at least 20 different genes versus 51 in CytoSig).

Techniques: Chemotaxis Assay, Expressing, Transformation Assay

Journal: bioRxiv

Article Title: Transforming the cytokine literature into a resource for experimental analysis and discovery

doi: 10.64898/2026.05.04.722753

Figure Lengend Snippet: a , CytED entries and transcriptional data are merged by matching cytokine-cell type-gene triplets. b , A high-throughput in vitro screen of human PBMCs response to treatment with 90 cytokines is used as a primary test case . The left bar graph shows the agreement between CytED statements and the experimental data. The right bar graph shows agreement after shuffling cytokine labels within each cell type as a control. c , Literature-experiment agreement by triple confidence derived from literature abundance. d , Workflow to analyze experimental changes in key genes, illustrated for IL-10 stimulation of CD16+ monocytes. For each cell type, a subset of experimental changes agrees or disagrees with CytED entries. Key literature genes for IL-10 stimulation of CD16+ monocytes are shown in the inset. Each triple can be traced back to specific places in origin publications. The experimental results and corresponding text passages are provided to an LLM to generate contextual summaries. Summaries across all key genes are then combined to produce an overall interpretation of the experimental response. e , Same analysis as in (d) for IL-10 stimulation of CD8+ T cells. The distribution of experimental time points for statements is shown on the bottom right. f , Dependence of the agreement (same-opposite) between experimental data and CytED on metadata variables: experimental system (human studies only), species (in vitro studies only), measurement time point, or experimental readout. Statistical significance was determined via a generalized Cochran-Mantel-Haenszel (CMH) test stratified by cytokine identity relative to the reference group (ref.), with the weighted mean score difference (CMHΔ) as the effect size. Multiple testing was corrected using the Benjamini-Hochberg procedure. g , Analysis as in (d) and (f) for a 4-hour in vivo cytokine stimulation experiment in mice. General agreement is much lower, at least partially due to higher noise levels in the data. Overall, we find that literature entries from long term experiments are overwhelmingly less likely to agree.

Article Snippet: CytED also contains gene sets for many cytokines that are currently missing in CytoSig (∼124 in CytED for non-combinatorial signatures with at least 20 different genes versus 51 in CytoSig).

Techniques: High Throughput Screening Assay, In Vitro, Control, Derivative Assay, In Vivo

Journal: bioRxiv

Article Title: Transforming the cytokine literature into a resource for experimental analysis and discovery

doi: 10.64898/2026.05.04.722753

Figure Lengend Snippet: a , Count distributions of cytokine-cell type-gene pairs are converted to weighted gene sets to infer cytokine signaling activities via a univariate linear model. b , Ranking of the on-target cytokine for pathway-derived, CytoSig, and CytED-derived weighted gene sets in a subset of perturbation conditions for human PBMC data. c , Distribution of ranks as in (b) for both human PBMC and in vivo mouse cytokine perturbation data. The p-value is calculated using a two-sided Wilcoxon signed-rank test on paired on-target ranks across datasets. d , Gene sets for specific cell processes are generated by identifying genes that co-occur in the same origin row. Co-occurring gene counts across CytED then yield a gene signature for a given process, here shown for cytotoxic effector signaling. Manually curated cytotoxicity genes (CD8A, GZMA, GZMB, GZMH, GZMK, PRF1, NKG7) are highlighted in red. e , Spearman correlation between inferred cytotoxic effector function process activity and the activities of different cytokines in melanoma samples from the TCGM dataset. f , Perturbation experiments measure a mixture of primary and secondary effects. Using CytED, one can infer which effects are likely primary, which secondary, and which might be novel cytokine effects. g , Experimental results of IL-12-cell-cytokine regulation connections in the human PBMC dataset. Each connection indicates a connection with log2FC>1 and padj<0.05. h, i , Inference of primary and secondary effects of (h) IL-12 and (i) IL-1 based on CytED literature annotations. The inset shows how many connections are explained by the primary cytokine, secondary cytokines, or are not annotated in CytED for any of the cytokine perturbations. j , CytED-based cytokine activity analysis for IL-12 and IL-1 stimulation.

Article Snippet: CytED also contains gene sets for many cytokines that are currently missing in CytoSig (∼124 in CytED for non-combinatorial signatures with at least 20 different genes versus 51 in CytoSig).

Techniques: Derivative Assay, In Vivo, Generated, Activity Assay