Figure - available from: Nature Communications
This content is subject to copyright. Terms and conditions apply.
GeneVector Framework Overview of GeneVector framework starting from single cell read counts. Mutual information is computed on the joint probability distribution of read counts for each gene pair. Each pair is used to train a single layer neural network where the MSE loss is evaluated from the model output (w1Tw2) with the mutual information between genes. From the resulting weight matrix, a gene embedding, cell embedding, and co-expression similarity graph are constructed. Using vector space arithmetic, downstream analyses include identification of cell-specific metagenes, batch effect correction, and cell type classification.
Source publication
Deciphering individual cell phenotypes from cell-specific transcriptional processes requires high dimensional single cell RNA sequencing. However, current dimensionality reduction methods aggregate sparse gene information across cells, without directly measuring the relationships that exist between genes. By performing dimensionality reduction with...
Similar publications
Single-cell transcriptomic analyses now frequently involve elaborate study designs including samples from multiple individuals, experimental conditions, perturbations, and batches from complex tissues. Dimensionality reduction is required to facilitate integration, interpretation, and statistical analysis. However, these datasets often include subt...
Citations
... One potential solution to address these limitations is deep learning which has shown to excel in learning complex representations from a broad spectrum of tasks involving different kinds of high-dimensional data with proper design and training strategies (Noothout et al. 2022;Krizhevsky, Sutskever, and Hinton 2012;Salehinejad et al. 2018;Zhang et al. 2020;Liu et al. 2021;Wu et al. 2024). Especially, deep learning has found many successful applications in biology and healthcare etc. (Korsunsky et al. 2019;Du et al. 2019;Eraslan et al. 2019;Zhao et al. 2021;Ceglia et al. 2023). Recently Mohamadi et al. presents an attempt to apply deep neural networks to human age estimation from gene expression data without gene selection as a prior step. ...
Understanding the complex biological process of aging is of great value, especially as it can help develop therapeutics to prolong healthy life. Predicting biological age from gene expression data has shown to be an effective means to quantify aging of a subject, and to identify molecular and cellular biomarkers of aging. A typical approach for estimating biological age, adopted by almost all existing aging clocks, is to train machine learning models only on healthy subjects, but to infer on both healthy and unhealthy subjects. However, the inherent bias in this approach results in inaccurate biological age as shown in this study. Moreover, almost all existing transcriptome‐based aging clocks were built around an inefficient procedure of gene selection followed by conventional machine learning models such as elastic nets, linear discriminant analysis etc. To address these limitations, we proposed DeepQA, a unified aging clock based on mixture of experts. Unlike existing methods, DeepQA is equipped with a specially designed Hinge‐Mean‐Absolute‐Error (Hinge‐MAE) loss so that it can train on both healthy and unhealthy subjects of multiple cohorts to reduce the bias of inferring biological age of unhealthy subjects. Our experiments showed that DeepQA significantly outperformed existing methods for biological age estimation on both healthy and unhealthy subjects. In addition, our method avoids the inefficient exhaustive search of genes, and provides a novel means to identify genes activated in aging prediction, alternative to such as differential gene expression analysis.
... In the fine-tuned setting, we constructed the gene similarity network in a similar manner from the scGPT model fine-tuned on the immune human dataset. Following Ceglia et al. 67 , we further performed Leiden 37 clustering on the similarity graph and extracted gene programs from gene clusters that consisted of five or more genes. ...
Generative pretrained models have achieved remarkable success in various domains such as language and computer vision. Specifically, the combination of large-scale diverse datasets and pretrained transformers has emerged as a promising approach for developing foundation models. Drawing parallels between language and cellular biology (in which texts comprise words; similarly, cells are defined by genes), our study probes the applicability of foundation models to advance cellular biology and genetic research. Using burgeoning single-cell sequencing data, we have constructed a foundation model for single-cell biology, scGPT, based on a generative pretrained transformer across a repository of over 33 million cells. Our findings illustrate that scGPT effectively distills critical biological insights concerning genes and cells. Through further adaptation of transfer learning, scGPT can be optimized to achieve superior performance across diverse downstream applications. This includes tasks such as cell type annotation, multi-batch integration, multi-omic integration, perturbation response prediction and gene network inference.
... Tfh-deficient SAP knock-out (KO) and CD4KO:CXCR5KO mixed bone marrow (BM) transplanted RAGKO and control mice were implanted with B16F10, treated with ICB, and 4PD1 hi , 4Dneg, and Tregs were quantified by flow cytometry and FACS-sorted for functional analyses. Results Using prior-knowledge-based signatures and mutualinformation-based cell-type classification, 10 we found that spleen-derived 4PD1 hi cells from tumor-bearing mice polarize toward Tfh, Tregs toward a canonical Treg phenotype, and 4Dneg toward Th1. Conversely, tumor-derived 4PD1 hi were not significantly skewed toward these phenotypes but gained in Th1 polarization after an effective anti-CTLA-4 treatment. ...
Background
In contrast to PD-1⁺CD8⁺ T cells, PD-1⁺CD4⁺ T cells and their impact in tumor progression and immunotherapy response remain relatively unexplored. We previously reported that PD-1hiFoxp3⁻CD4⁺ T cells (4PD1hi) from melanoma-bearing mice and patients with melanoma or non-small cell lung cancer (NSCLC) suppress T-cell function and correlate with unfavorable outcomes upon immune checkpoint blockade (ICB) therapy.¹ CD4⁺PD-1⁺ T cells were also found to correlate with poor prognosis in other NSCLC patient cohorts.2–4 We showed that 4PD1hiup-regulate T-follicular-helper-cell(Tfh)-related genes.¹ 4PD1hi cells suppressing immunotherapy responses were recently described in mouse sarcoma models; however, in this setting, 4PD1hi did not over-express Tfh genes.⁵ Here, we sought to deconvolve the lineage commitment of 4PD1hi tumor-infiltrating lymphocytes (TILs) in relationship with their immune function and impact on ICB outcome.
Methods
Single-cell RNA-sequencing (scRNAseq) was performed in 4PD1hi, PD-1⁻Foxp3⁻CD4⁺ (4Dneg), and Foxp3⁺CD4⁺ T cells (Tregs) FACS-sorted from ICB-treated B16F10-melanoma bearing Foxp3-GFP mice. scRNAseq datasets of TILs from ICB-treated cancer patients6–9 were used to extract 4PD1hi and analyze their profiles. Tfh-deficient SAP knock-out (KO) and CD4KO:CXCR5KO mixed bone marrow (BM) transplanted RAGKO and control mice were implanted with B16F10, treated with ICB, and 4PD1hi, 4Dneg, and Tregs were quantified by flow cytometry and FACS-sorted for functional analyses.
Results
Using prior-knowledge-based signatures and mutual-information-based cell-type classification,¹⁰ we found that spleen-derived 4PD1hi cells from tumor-bearing mice polarize toward Tfh, Tregs toward a canonical Treg phenotype, and 4Dneg toward Th1. Conversely, tumor-derived 4PD1hi were not significantly skewed toward these phenotypes but gained in Th1 polarization after an effective anti-CTLA-4 treatment. In human primary melanoma,⁷ NSCLC,⁶ and squamous/basal cell carcinoma,⁸ 4PD1hi cells over-expressed Tfh-related genes. This was less clear in 4PD1hi from mixed NSCLC samples, encompassing primary tumors and different metastatic sites.⁹ However, 4PD1hi from ICB-non-responder patients in this study⁹ displayed the greatest Tfh-signature scores. Consistently, 4PD1hi TILs from ICB-non-responders in the other NSCLC and melanoma datasets up-regulated Tfh-related genes. To test 4PD1hi TIL Tfh polarization in ICB response, we used SAPKO and CD4KO:CXCR5KO BM chimera mice. Both Tfh-deficient models showed better tumor responses to a suboptimal anti-CTLA-4 treatment¹; however, 4PD1hi TILs did not substantially decrease. In this setting, 4PD1hi TILs lost suppressive function, down-regulated Pdcd1 and Il10, and up-regulated Ifng, suggesting a Th1 phenotypic switch.
Conclusions
These results indicate that 4PD1hi TILs are heterogeneous and their Tfh/Th1 polarization influences immunotherapy responses possibly in a tumor-tissue dependent way. In melanoma, 4PD1hi TIL Tfh polarization drives immunosuppression and ICB resistance.
Acknowledgements
This study was supported in part by the Parker Institute for Cancer Immunotherapy. S. M. and A.O. contributed equally to this work.
References
• Zappasodi R, Budhu S, Hellmann MD, Postow MA, Senbabaoglu Y, Manne S, et al. Non-conventional Inhibitory CD4(+)Foxp3(-)PD-1(hi) T Cells as a Biomarker of Immune Checkpoint Blockade Activity. Cancer Cell. 2018;
33
(6):1017–32 e7.
• Zheng H, Liu X, Zhang J, Rice SJ, Wagman M, Kong Y, et al. Expression of PD-1 on CD4 + T cells in peripheral blood associates with poor clinical outcome in non-small cell lung cancer. Oncotarget. 2016;
7
(35).
• Arrieta O, Montes-Servín E, Hernandez-Martinez J-M, Cardona AF, Casas-Ruiz E, Crispín JC, et al. Expression of PD-1/PD-L1 and PD-L2 in peripheral T-cells from non-small cell lung cancer patients. Oncotarget. 2017;
8
(60).
• Duchemann B, Naigeon M, Auclin E, Ferrara R, Cassard L, Jouniaux JM, et al. CD8(+)PD-1(+) to CD4(+)PD-1(+) ratio (PERLS) is associated with prognosis of patients with advanced NSCLC treated with PD-(L)1 blockers. J Immunother Cancer. 2022;
10
(2).
• Hussein S, Kelly M, Yuang S, Samuel A, Ton S, Yik Andy Y, et al. 578 CD8-targeted IL-2 drives potent anti-tumor efficacy and promotes action of tumor specific vaccines. Journal for ImmunoTherapy of Cancer. 2021;
9
(Suppl 2):A607.
• Caushi JX, Zhang J, Ji Z, Vaghasia A, Zhang B, Hsiue EH, et al. Transcriptional programs of neoantigen-specific TIL in anti-PD-1-treated lung cancers. Nature. 2021;
596
(7870):126–32.
• Schad SE, Chow A, Mangarin L, Pan H, Zhang J, Ceglia N, et al. Tumor-induced double positive T cells display distinct lineage commitment mechanisms and functions. J Exp Med. 2022;
219
(6).
• Yost KE, Satpathy AT, Wells DK, Qi Y, Wang C, Kageyama R, et al. Clonal replacement of tumor-specific T cells following PD-1 blockade. Nat Med. 2019;
25
(8):1251–9.
• Liu B, Hu X, Feng K, Gao R, Xue Z, Zhang S, et al. Temporal single-cell tracing reveals clonal revival and expansion of precursor exhausted T cells during anti-PD-1 therapy in lung cancer. Nature Cancer. 2022;
3
(1):108–21.
• Ceglia N, Sethna Z, Freeman SS, Uhlitz F, Bojilova V, Rusk N, et al. GeneVector: Identification of transcriptional programs using dense vector representations defined by mutual information. bioRxiv. 2023:2022.04.22.487554.
Although allogeneic hematopoietic cell transplantation (allo-HCT) is curative for high-risk pediatric acute myeloid leukemia (AML), disease relapse remains the primary cause of post-transplant mortality. To identify pressures imposed by allo-HCT on AML cells that escape the graft-versus-leukemia effect, we evaluated immune signatures at diagnosis and post-transplant relapse in bone marrow samples from four pediatric patients using a multimodal single-cell proteogenomic approach. Downregulation of MHC class II expression was most profound in progenitor-like blasts and accompanied by correlative changes in transcriptional regulation. Dysfunction of activated NK cells and CD8+ T-cell subsets at relapse was evidenced by loss of response to IFN-γ, TNF-α signaling via NF-kβ, and IL-2/STAT5 signaling. Clonotype analysis of post-transplant relapse samples revealed expansion of dysfunctional T cells and enrichment of T-regulatory and T-helper cells. Using novel computational methods, our results illustrate a diverse immune-related transcriptional signature in post-transplant relapses not previously reported in pediatric AML.
