Characterizing genome-wide binding profiles of transcription factors (TFs) is essential for understanding biological processes. Although techniques have been developed to assess binding profiles within a population of cells, determining them at a single-cell level remains elusive. Here, we report scFAN (single-cell factor analysis network), a deep learning model that predicts genome-wide TF binding profiles in individual cells. scFAN is pretrained on genome-wide bulk assay for transposase-accessible chromatin sequencing (ATAC-seq), DNA sequence, and chromatin immunoprecipitation sequencing (ChIP-seq) data and uses single-cell ATAC-seq to predict TF binding in individual cells. We demonstrate the efficacy of scFAN by both studying sequence motifs enriched within predicted binding peaks and using predicted TFs for discovering cell types. We develop a new metric “TF activity score” to characterize each cell and show that activity scores can reliably capture cell identities. scFAN allows us to discover and study cellular identities and heterogeneity based on chromatin accessibility profiles.

The effects of adjuvants for increasing the immunogenicity of influenza vaccines are well known. However, the effect of adjuvants on increasing the breadth of cross-reactivity is less well understood. In this study we have performed a systematic screen of different toll-like receptor (TLR) agonists, with and without a squalene-in-water emulsion on the immunogenicity of a recombinant trimerized hemagglutinin (HA) vaccine in mice after single-dose administration. Antibody (Ab) cross-reactivity for other variants within and outside the immunizing subtype (homosubtypic and heterosubtypic cross-reactivity, respectively) was assessed using a protein microarray approach. Most adjuvants induced broad IgG profiles, although the response to a combination of CpG, MPLA and AddaVax (termed ‘IVAX-1’) appeared more quickly and reached a greater magnitude than the other formulations tested. Antigen-specific plasma cell labeling experiments show the components of IVAX-1 are synergistic. This adjuvant preferentially stimulates CD4 T cells to produce Th1>Th2 type (IgG2c>IgG1) antibodies and cytokine responses. Moreover, IVAX-1 induces identical homo- and heterosubtypic IgG and IgA cross-reactivity profiles when administered intranasally. Consistent with these observations, a single-cell transcriptomics analysis demonstrated significant increases in expression of IgG1, IgG2b and IgG2c genes of B cells in H5/IVAX-1 immunized mice relative to naïve mice, as well as significant increases in expression of the IFNγ gene of both CD4 and CD8 T cells. These data support the use of adjuvants for enhancing the breath and durability of antibody responses of influenza virus vaccines.

The advent of immune checkpoint therapy for metastatic skin cancer has greatly improved patient survival. However, most skin cancer patients are refractory to checkpoint therapy, and furthermore, the intra-immune cell signaling driving response to checkpoint therapy remains uncharacterized. When comparing the immune transcriptome in the tumor microenvironment of melanoma and basal cell carcinoma (BCC), we found that the presence of memory B cells and macrophages negatively correlate in both cancers when stratifying patients by their response, with memory B cells more present in responders. Moreover, inhibitory immune signaling mostly decreases in melanoma responders and increases in BCC responders. We further explored the relationships between macrophages, B cells and response to checkpoint therapy by developing a stochastic differential equation model which qualitatively agrees with the data analysis. Our model predicts BCC to be more refractory to checkpoint therapy than melanoma and predicts the best qualitative ratio of memory B cells and macrophages for successful treatment.

Many approaches have been developed to overcome technical noise in single cell (and single nucleus) RNA-sequencing (scRNAseq). As researchers dig deeper into data--looking for rare cell types, subtleties of cell states, and details of gene regulatory networks--there is a growing need for algorithms with controllable accuracy and a minimum of ad hoc parameters and thresholds. Impeding this goal is the fact that an appropriate null distribution for scRNAseq cannot simply be extracted from data in the event that ground truth about biological variation is unknown (i.e., most of the time). Here we approach this problem analytically, based on the assumption that scRNAseq data reflect only cell heterogeneity (what we seek to characterize), transcriptional noise (temporal fluctuations randomly distributed across cells), and sampling error (i.e., Poisson noise). We then analyze scRNAseq data without normalization--a step that can skew distributions, particular for sparse data--and calculate p-values associated with key statistics. We develop an improved method for the selection of features for cell clustering and the identification of gene-gene correlations, both positive and negative. Using simulated data, we show that this method, which we call BigSur (Basic Informatics and Gene Statistics from Unnormalized Reads), accurately captures even weak yet significant correlation structures in scRNAseq data. Applying BigSur to data from a clonal human melanoma cell line, we identify tens of thousands of correlations that, when clustered without supervision into gene communities, both align with cellular components and biological processes, and point toward potentially novel cell biological relationships.

Figure S1: Deep learning model in scFAN scheme. The figure shows the overall scheme of the pre-trained model in scFAN. We adopt 3 convolution layers with corresponding Max-pooling layers to extract features, then concatenate the feature map with two fully-connected layers and finally get the output TF probability using Sigmoid function.