EYKTHYR reveals transcriptional regulators of spatial gene programs

EYKTHYR reveals transcriptional regulators of spatial gene programs | bioRxiv /* */ /* */ <!-- <!-- /*! * yepnope1.5.4 * (c) WTFPL, GPLv2 */ (function(a,b,c){function d(a){return"[object Function]"==o.call(a)}function e(a){return"string"==typeof a}function f(){}function g(a){return!a||"loaded"==a||"complete"==a||"uninitialized"==a}function h(){var a=p.shift();q=1,a?a.t?m(function(){("c"==a.t?B.injectCss:B.injectJs)(a.s,0,a.a,a.x,a.e,1)},0):(a(),h()):q=0}function i(a,c,d,e,f,i,j){function k(b){if(!o&&g(l.readyState)&&(u.r=o=1,!q&&h(),l.onload=l.onreadystatechange=null,b)){"img"!=a&&m(function(){t.removeChild(l)},50);for(var d in y[c])y[c].hasOwnProperty(d)&&y[c][d].onload()}}var j=j||B.errorTimeout,l=b.createElement(a),o=0,r=0,u={t:d,s:c,e:f,a:i,x:j};1===y[c]&&(r=1,y[c]=[]),"object"==a?l.data=c:(l.src=c,l.type=a),l.width=l.height="0",l.onerror=l.onload=l.onreadystatechange=function(){k.call(this,r)},p.splice(e,0,u),"img"!=a&&(r||2===y[c]?(t.insertBefore(l,s?null:n),m(k,j)):y[c].push(l))}function j(a,b,c,d,f){return q=0,b=b||"j",e(a)?i("c"==b?v:u,a,b,this.i++,c,d,f):(p.splice(this.i++,0,a),1==p.length&&h()),this}function k(){var a=B;return a.loader={load:j,i:0},a}var l=b.documentElement,m=a.setTimeout,n=b.getElementsByTagName("script")[0],o={}.toString,p=[],q=0,r="MozAppearance"in l.style,s=r&&!!b.createRange().compareNode,t=s?l:n.parentNode,l=a.opera&&"[object Opera]"==o.call(a.opera),l=!!b.attachEvent&&!l,u=r?"object":l?"script":"img",v=l?"script":u,w=Array.isArray||function(a){return"[object Array]"==o.call(a)},x=[],y={},z={timeout:function(a,b){return b.length&&(a.timeout=b[0]),a}},A,B;B=function(a){function b(a){var a=a.split("!"),b=x.length,c=a.pop(),d=a.length,c={url:c,origUrl:c,prefixes:a},e,f,g;for(f=0;f<d;f++)g=a[f].split("="),(e=z[g.shift()])&&(c=e(c,g));for(f=0;f<b;f++)c=x[f](c);return c}function g(a,e,f,g,h){var i=b(a),j=i.autoCallback;i.url.split(".").pop().split("?").shift(),i.bypass||(e&&(e=d(e)?e:e[a]||e[g]||e[a.split("/").pop().split("?")[0]]),i.instead?i.instead(a,e,f,g,h):(y[i.url]?i.noexec=!0:y[i.url]=1,f.load(i.url,i.forceCSS||!i.forceJS&&"css"==i.url.split(".").pop().split("?").shift()?"c":c,i.noexec,i.attrs,i.timeout),(d(e)||d(j))&&f.load(function(){k(),e&&e(i.origUrl,h,g),j&&j(i.origUrl,h,g),y[i.url]=2})))}function h(a,b){function c(a,c){if(a){if(e(a))c||(j=function(){var a=[].slice.call(arguments);k.apply(this,a),l()}),g(a,j,b,0,h);else if(Object(a)===a)for(n in m=function(){var b=0,c;for(c in a)a.hasOwnProperty(c)&&b++;return b}(),a)a.hasOwnProperty(n)&&(!c&&!--m&&(d(j)?j=function(){var a=[].slice.call(arguments);k.apply(this,a),l()}:j[n]=function(a){return function(){var b=[].slice.call(arguments);a&&a.apply(this,b),l()}}(k[n])),g(a[n],j,b,n,h))}else!c&&l()}var h=!!a.test,i=a.load||a.both,j=a.callback||f,k=j,l=a.complete||f,m,n;c(h?a.yep:a.nope,!!i),i&&c(i)}var i,j,l=this.yepnope.loader;if(e(a))g(a,0,l,0);else if(w(a))for(i=0;i (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0];var j=d.createElement(s);var dl=l!='dataLayer'?'&l='+l:'';j.src='//www.googletagmanager.com/gtm.js?id='+i+dl;j.type='text/javascript';j.async=true;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-M677548'); Skip to main content Home About Submit ALERTS / RSS Search for this keyword Advanced Search New Results EYKTHYR reveals transcriptional regulators of spatial gene programs View ORCID Profile Spencer Krieger , Ellie Haber , View ORCID Profile Jian Ma doi: https://doi.org/10.1101/2025.05.19.654884 Spencer Krieger 1 Ray and Stephanie Lane Computational Biology Department, School of Computer Science, Carnegie Mellon University , Pittsburgh, PA 15213, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Spencer Krieger Ellie Haber 2 Machine Learning Department, School of Computer Science, Carnegie Mellon University , Pittsburgh, PA 15213, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site Jian Ma 1 Ray and Stephanie Lane Computational Biology Department, School of Computer Science, Carnegie Mellon University , Pittsburgh, PA 15213, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Jian Ma For correspondence: jianma{at}cs.cmu.edu Abstract Full Text Info/History Metrics Supplementary material Preview PDF Abstract Understanding how transcription factors (TFs) orchestrate gene regulatory networks that define complex tissue structures is central to uncovering tissue organization and disease mechanisms. Although spatial multiome technologies now enable in situ measurement of both transcriptional activity and chromatin accessibility, existing computational methods either overlook spatial tissue context or are hindered by the high dropout rates characteristic of such data. Here, we introduce E ykthyr , a computational framework that integrates gene expression and chromatin accessibility within a spatially aware model to identify TFs driving spatial gene programs. E ykthyr mitigates dropout effects by leveraging interpretable, low-dimensional embeddings of gene expression and chromatin accessibility – both linear with respect to their input – enabling robust identification and scalable inference of spatial transcriptional regulators. Applied across diverse spatial multiome datasets, E ykthyr consistently outperforms existing approaches, accurately identifying TFs that coordinate spatial gene programs in mouse brain development and regulate T-cell states within tumor microenvironments. E ykthyr establishes a foundation for decoding how TFs interpret local intercellular signaling to shape tissue structure, offering insights into the regulatory logic underlying spatial organization in health and disease. Introduction Transcription factors (TFs) orchestrate spatial gene expression within complex tissues, guiding development and mediating cell-cell communication [ 1 – 3 ]. Acting as molecular switches, TFs regulate transcriptional programs that determine cell fate, and their disruption can lead to disease [ 4 – 6 ]. By interpreting extracellular cues and propagating signals through gene regulatory networks, TFs establish localized expression patterns that shape tissue organization. However, single-cell genomic technologies that lack spatial context present a major barrier to identifying TFs that govern spatial gene programs. Recently, spatial multiome technologies have been developed to simultaneously measure transcriptional activity and chromatin accessibility in situ [ 7 – 9 ]. By correlating regulatory regions (e.g., enhancers and promoters) to their target genes, these dual-modality datasets offer new insights into TF-driven spatial gene regulation while reducing the need to infer regulatory activity solely from transcriptomic data, providing clearer potential cause-and-effect relationships in spatially resolved gene regulation. Moreover, because spatial multiome data preserve each cell’s physical location within the tissue, they offer a unique window into how local signaling, cellular neighborhoods, and microenvironmental cues influence TF-driven gene expression. Despite these advantages, analyzing spatial multiome data remains challenging due to high dropout rates, with only 6-11% of the gene expression read count matrix typically containing nonzero values [7– 9], and even greater sparsity in chromatin accessibility data. Existing gene regulation modeling methods [ 10 – 16 ] are not designed for spatial multiome data; they either ignore spatial context, focus solely on cell type-level gene regulation, or struggle with extreme sparsity. As a result, no dedicated computational framework currently exists to identify TFs governing spatial gene expression patterns. Here, we present E ykthyr , a computational method designed to identify TFs governing spatial gene programs (or metagenes) using spatial multiome data. E ykthyr addresses high dropout by generating low-dimensional embeddings of gene expression and chromatin accessibility that remain linear with respect to input, denoising the data while preserving interpretability – shifts in metagene embeddings map directly back to changes in gene expression. E ykthyr uses linear models to learn relationships between TF activity and metagene expression within each cell’s local neighborhood. These two layers of linear mappings – from TF activity to metagenes and from metagenes to gene expression – enable reasoning about how TF activity affects gene expression. We applied E ykthyr to multiple spatial multiome datasets, demonstrating its ability to identify key TFs regulating spatial gene programs in various contexts. E ykthyr ’s predictions align strongly with prior knowledge, capturing TFs that influence cell fate within specific tissue regions during mouse brain development and identifying context-dependent TF effects on T cell states in distinct tumor microenvironments in human metastatic melanoma. Results Overview of E ykthyr E ykthyr takes as input a spatial multiome dataset containing gene expression, chromatin accessibility, and spatial coordinates of each cell or spot within a tissue sample ( Fig . 1 ). It aims to identify TFs that influence spatial gene programs by integrating these data modalities and analyzing region-specific interactions between TF activity and metagene expression. Download figure Open in new tab Figure 1: Overview of E ykthyr . a . E ykthyr processes paired spatial multiome data, including gene expression, chromatin accessibility, and spatial coordinates for each cell or spot. To address data sparsity, gene expression is embedded into interpretable metagenes (gene programs), while chromatin accessibility is summarized as TF activity using a TF-motif scanning approach. For each cell c , E ykthyr fits a regularized linear model linking the expression of metagene m i with the activity of TF j from cells in c ’s spatial neighborhood, estimating the interaction weight of j on m i . These weights are then used to simulate and predict changes in metagene expression following shifts in TF activity. b . Downstream applications of E ykthyr include predicting region-specific shifts in cell identity, identifying TFs driving cellular processes such as differentiation, and pin-pointing pathways influenced by specific TFs. The first step in E ykthyr is generating low-dimensional embeddings for preprocessed gene expression and chromatin accessibility ( Methods ). Each embedding is linear with respect to its input, ensuring interpretability – shifts in the embedding space directly reflect changes in gene expression or chromatin accessibility. This allows us to reason about how changes in TF activity translate into shifts in gene expression patterns. For gene expression, E ykthyr employs P opari [ 17 ], a spatially aware framework of non-negative matrix factorization (NMF) [ 18 ]. P opari extends our prior work S pice M ix [ 19 ], addressing data sparsity through hierarchical binning and learning a shared set of metagenes across samples, where each metagene corresponds to a linear combination of input gene expression. This metagene embedding provides an interpretable low-dimensional representation of the gene expression landscape. To connect chromatin accessibility with TF activity, E ykthyr estimates TF activity based on the number of accessible ATAC-seq peaks containing its binding motif. The assumption is simple but effective: the accessibility of a TF’s binding motifs across the genome correlates with its activity in the cell. This transformation provides a functional connection between chromatin accessibility and TF activity, which we leverage for downstream analysis. Once TF activity is estimated, E ykthyr models the influence of each TF on metagene expression within the spatial context of each cell. Specifically, for each cell or spot in the tissue, E ykthyr applies regularized linear regression to infer the relationships between TF activity and metagene expression using spatially proximal cells or spots. This step effectively learns the influence of each TF on metagene expression, akin to determining edge weights in a TF – metagene regulatory network, where nodes represent TFs or metagenes and edges are directed from TFs to metagenes. By modeling this regulatory network at the metagene level rather than the individual gene level, E ykthyr effectively reduces the number of interaction weights to estimate, making it computationally feasible to infer edge weights that are distinct for each cell in the tissue. A key strength of our method is that the linear relationships between TF activity and metagene expression allow for in silico predictions of how shifts in TF activity alter gene expression patterns across the tissue. This provides a framework for studying TF-driven regulation of spatial gene programs and offers insights into the regulatory mechanisms governing tissue organization. We use these learned relationships in downstream analyses to explore how spatially localized gene programs are controlled by specific TFs. E ykthyr enables visualization of tissue-wide changes in gene expression following an in silico TF knockout, identification of TFs governing particular cellular processes, and discovery of pathways influenced by TF activity in specific tissue regions. These capabilities highlight the novelty of E ykthyr in linking TF activity and gene expression within a unified spatial framework. A detailed description of E ykthyr is provided in Methods , and details on dataset preprocessing and hyperparameter selection are in the Supplementary Note . Decoding regulatory mechanisms in pallium differentiation We first applied E ykthyr to examine embryonic mouse brain development using MISAR-seq data [ 7 ] (see Fig . 3a ), focusing on progenitor cells differentiating into the ventricular region of the dorsal pallium (DPallv), the mantle region of the dorsal pallium (DPallm), and the subpallium ( Fig . 2a ). This developmental pathway serves as an ideal benchmark because it is well-characterized and involves TFs known to drive cell fate decisions. We collected a set of TFs implicated in pallium differentiation from the literature [ 20 ], including Arx, Emx1, Emx2, Gli3, Lef1, Lhx2, Lmx1a, Nr2e1, Nr2f1, Nr2f2, Otx1, Bhlhe22, Pax6 , and Sp8 . Verifying that E ykthyr can recover these known TFs builds confidence in its predictive capabilities. Download figure Open in new tab Figure 2: Evaluation of E ykthyr ’s predictions of TFs driving pallium differentiation. a . Schematic illustrating the differentiation of pallium progenitors into three subtypes during mouse brain development. b . Measurement of each TF’s influence on pallium differentiation, determined by E ykthyr ’s predicted shift in metagene embeddings after simulating TF knockout (setting its expression to zero). c . Top TFs identified by E ykthyr as promoting pallium differentiation, ranked using the metric in b (further analyzed in Fig . 3 ). d . Box plot showing E ykthyr ’s ranking of TFs known from the literature to influence pallium differentiation. The dashed vertical line represents the expected median rank of a random set of TFs. E ykthyr consistently ranks known differentiation-promoting TFs highly, with statistically significant results. To quantify each TF’s effect on differentiation, E ykthyr simulates a TF knockout by setting its estimated activity to zero, propagates this through learned TF-metagene weights to shift the metagene embedding, and recalculates cellular states. We then evaluate changes in cluster separation using silhouette width, iLISI, and principal component regression comparison to assess the TF’s role in maintaining distinct cell fates ( Fig . 2b ). TFs that, when knocked out, lead to collapsed differentiation clusters are inferred to be key regulators. TFs are ranked by their ability to maintain developmental separability and compared against known regulators. E ykthyr successfully prioritized TFs involved in early brain development ( Fig . 2c ). Fourteen of the top twenty ranked TFs have known roles in pallium ( Gli3 and Emx2 ), forebrain ( Msx1, Dlx1 , and Dlx2 ), or visual cortex development ( Sox2, Hoxa6, Sox14, Rax, Barx2, Dbx2, Prrx2, Sox9 , and Prrxl1 ). TFs implicated in pallium differentiation were significantly enriched among the top-ranked TFs across all cluster separation metrics (Silhouette: P = 0.001; iLISI: P = 0.01; PCR: P = 0.005; combined score: P = 1 × 10 −4 ) ( Fig . 2d ). Comparison with CellOracle [ 14 ] and an expression-only variant of E yk - thyr showed that neither baseline prioritized these TFs better than random ( Fig . S1a and Fig . S1b), highlighting E ykthyr ’s ability to pinpoint spatially relevant TFs. Msx1 as a central regulator in pallium development To showcase E ykthyr ’s spatially-informed predictions, we examined its analysis of Msx1 , a TF involved in forebrain development around embryonic day 16 [ 21 , 22 ]. Despite low Msx1 transcript levels in this tissue ( Fig . 3b , left), E ykthyr inferred strong activity in the pallium and diencephalon based on chromatin accessibility ( Fig . 3b , right), underscoring the value of integrating accessibility data. Metagene m11, prominent in DPallv ( Fig . 3c ; other metagene expression in Fig . S2a), was predicted to be strongly regulated by Msx1 ( Fig . S2b). Transforming predicted metagene shifts into gene-level changes allowed us to rank genes likely affected by an Msx1 knockout. These predictions were significantly enriched for validated downstream genes from an existing GRN [ 14 ] ( P = 4 × 10 −7 ; Fig . 3d ), supporting the functional relevance of E ykthyr ’s predictions. Download figure Open in new tab Figure 3: E ykthyr identifies Msx1 as a key regulator of pallium differentiation. a . Spatial plot of a mouse embryo sample at E15.5, with spots annotated using E ykthyr . b . Spatial plots comparing sparse gene expression (left) of Msx1 with denser TF activity (right) derived from chromatin accessibility data. c . Expression of metagene m11 (left) and predicted change in m11 expression following a simulated Msx1 knockout (right). d . Box plot showing the ranking of genes downstream of Msx1 from a generic mouse gene regulatory network [ 14 ]. e . Genomic view of Nrg1 showing common splice variants, with isoform II containing large introns. The upper track shows RNA-seq read coverage from the E15.5 mouse embryo. The zoomed-in section highlights a Msx1 binding site within an annotated proximal enhancer region near the transcription start site of Nrg1 isoform II. f . Spatial expression of Nrg1 isoform II (left) compared to all other isoforms (right). g . Visualizations of cell identity shifts following an in silico knockout of Msx1 , showing transition from Subpallium to DPallv in both spatial (left) and UMAP (right) embeddings. h . Predicted region-specific identity shifts before and after simulated Msx1 knockout. i . Gene set enrichment analysis of genes most affected by Msx1 , revealing pathways associated with pallium development. Notably, E ykthyr identified Nrg1 as a gene strongly influenced by Msx1 , despite not being previously annotated as a downstream target in existing GRNs. Nrg1 spans more than 1 Mb and produces multiple isoforms ( Fig . 3e ). Nrg1 isoform II includes an alternative transcription start site (TSS) located just 650 bp apart from a Msx1 binding motif in an ENCODE [ 23 ] proximal cCRE region ( Fig . 3e ). This isoform has been linked to schizophrenia [ 24 ], a disorder recently associated with glutamatergic neurons in the dorsal-ventral axis [ 25 ]. Nrg1 isoform II shows higher expression near the pallium at embryonic day 15 in mice [ 26 ], suggesting Msx1 may directly regulate its activity during development. Its expression is largely confined to DPallv ( Fig . 3f ), indicating a potential role in regional differentiation. E ykthyr thus highlights a previously uncharacterized regulatory mechanism of Nrg1 isoform II. To further examine Msx1 ’s role, we performed in silico Msx1 knockout and analyzed shifts in cell identity across spatial and metagene embeddings ( Fig . 3g, h ). Cells in the Subpallium shifted toward a DPallv-like identity, indicating Msx1 plays a role in maintaining spatially distinct cellular fates. Gene set enrichment analysis of genes altered by the knockout confirmed Msx1 ’s role in neural development and axon guidance ( Fig . 3i ), including L1CAM upregulation [ 27 , 28 ]. These changes were spatially specific, with minimal effects in other regions after the knockout. This case study of Msx1 highlights four key strengths of E ykthyr : (1) it detects low-expression TFs through chromatin accessibility, identifying regulators that might otherwise be overlooked; (2) it identifies region-specific regulatory effects; (3) its in silico knockout predictions aligh with known developmental roles; and (4) it captures subtle yet spatially impactful regulatory mechanisms. Identifying key regulators of radial glia differentiation We next investigated a spatial ATAC–RNA-seq sample of the mouse embryonic hindbrain [ 8 ] ( Fig . 4a and Fig . 5a ), where radial glia differentiate into postmitotic premature neurons. Unlike the previous system, these cells follow a more continuous developmental trajectory. This setting allowed us to assess E ykthyr ’s predictions under a different evaluation strategy based on pseudotime rather than discrete cluster separation. This trajectory is well suited to a spatial pseudotime approach, as postmitotic neurons spatially drift away from radial glia after differentiation, and the original study showed that spatial distance closely correlates with chromatin accessibility–based pseudotime. Download figure Open in new tab Figure 4: Evaluation of E ykthyr ’s predictions for TFs driving radial glia differentiation. a . Spatial plot of a mouse embryo section from the spatial multiome (scATAC and scRNA-seq) dataset (further analyzed in Fig . 5 ). b-d . Metrics used to quantify a TF’s influence on promoting radial glia differentiation in the ventricle into postmitotic premature neurons in the hindbrain. The metric evaluates the alignment between the shift in cell identity due to ( b ) natural differentiation and ( c ) E ykthyr ’s predicted metagene shift, by calculating ( d ) the dot product between these two vectors. e . Top TFs identified by E ykthyr as promoting differentiation from radial glia to postmitotic premature neurons, ranked according to this custom metric. f . Box plot displaying E ykthyr ’s ranking of TFs known from the literature to influence this differentiation process. The dashed vertical line represents the expected median rank of a random set of TFs. Download figure Open in new tab Figure 5: E ykthyr identifies Hes1 as a key regulator in the differentiation of radial glia into postmitotic premature neurons. a . Spatial plot of a mouse embryo sample with spots annotated using E ykthyr . b . Expression of metagene m13 (top) and the predicted change in m13 expression following a simulated Hes1 knockout (bottom). c . E ykthyr predicts that reducing Hes1 activity would induce a shift in cell identity following the spatial trajectory of differentiation, underscoring its potential role in the differentiation process. d . UMAP visualization showing the transition in cell identity from Ventricle to Hindbrain after a simulated reduction in Hes1 activity. e . Heatmap of metagenes most influenced by Hes1 activity, highlighting regions of highest impact. f . Gene set enrichment analysis of genes most affected by Hes1 in the Ventricle, revealing terms and pathways related to radial glia differentiation into postmitotic premature neurons. We defined pseudotime as the normalized Euclidean distance of each spot to the nearest spot annotated as ventricle ( Fig . 4b ), approximating its progression along the radial glia-to-neuron differentiation path. After simulating E ykthyr ’s TF knockouts ( Fig . 4c ), we computed directional vectors capturing how each cell’s metagene embedding shifts relative to its spatial neighbors. By comparing these vectors to the pseudotime axis using their inner product, we ranked TFs by how strongly their knockouts perturb the differentiation trajectory ( Fig . 4d ). We compiled a list of TFs from the literature associated with the differentiation of radial glia to postmitotic premature neurons [ 29 ], including Ascl1, Dlx1, Dlx2, Emx1, Etv1, Gsx1, Gsx2, Hes1, Neurog2, Olig2, Pax6 , and Sp8 . E ykthyr successfully ranked these TFs with established roles in glial differentiation significantly higher than expected by chance ( P = 0.02) ( Fig . 4f ). In contrast, the baselines – CellOracle and the expression-only variant of E ykthyr – showed weaker concordance with known regulators. Fourteen of E ykthyr ’s top twenty ranked TFs included genes involved in ventricle differentiation ( Hes1 ), neuron development ( Sox9, Foxj1 , and Tbx15 ), or epithelial development ( Foxb1, Sox6, Irx2, Nfib, Sox2, Nfia, Id3, Tgif2, Zic4 , and Id2 ) ( Fig . 4e ). These results reinforce that E ykthyr ’s in silico knockout predictions are biologically grounded and accurately capture TFs responsible for spatially informed differentiation processes in the hindbrain and ventricle. Hes1 as a spatial architect in radial glia differentiation We performed a deeper analysis of E ykthyr ’s predictions regarding Hes1 and its role in the differentiation of radial glia into postmitotic premature neurons ( Fig . 5a ). While the original study emphasized factors like Sox2 as key regulators of this process – which E ykthyr also identified – it did not highlight Hes1 , which E ykthyr ranked highly ( Fig . 4e ). Using E ykthyr ’s framework, we identified metagene m13 as highly expressed in ventricle-proximal cells, with Hes1 exerting strong regulatory influence on m13 in these early-stage populations ( Fig . 5b ). When Hes1 activity was reduced, ventricle cells became more similar to hindbrain cells ( Fig . 5c, d ), recapitulating the known Hes1 -/- phenotypes of premature neuron production and depletion of progenitor cells [ 30 ]. Beyond the ventricle-hindbrain transition, E ykthyr predicted that Hes1 regulates a different metagene (m10) in the central nervous system, suggesting that its regulatory role varies across tissue compartments ( Fig . 5e ). Leveraging the linear metagene embedding, we transformed the shifts in metagene expression back to gene-level changes and ranked genes by their predicted influence to Hes1 perturbation. Comparing this ranked list to an orthogonal mouse GRN [ 14 ], we found significant enrichment of known downstream targets among the top-ranked genes ( P = 8.1 × 10 −37 ) ( Fig . S3). Gene set enrichment analysis of these genes further revealed pathways related to glial differentiation, axon guidance, and nervous system development ( Fig . 5f ). Notably, L1CAM-associated processes were enriched, consistent with its established role in neuronal outgrowth and neurodevelopment [ 31 ]. These findings suggest that Hes1 exerts spatially distinct regulatory effects along the ventricle-hindbrain axis, complementing previous findings on Sox2 . More broadly, this analysis demonstrates how E ykthyr can reveal dynamic, region-specific roles of TFs in shaping developmental trajectories. Context-dependent regulation of T-cell states in tumor microenvironments We applied E ykthyr to a single-cell Slide-tags multiome dataset of human metastatic melanoma [ 9 ], where two tumor compartments were defined based on distinct transcriptomic signatures in tumor cells ( Fig . 6a ). While the original study examined T-cell infiltration into each compartment using T-cell receptor (TCR) sequences, it did not focus on finer-grained T-cell subpopulations. Building on this, our metagene analysis revealed a proliferating T-cell population that was not clearly distinguished in the original clustering ( Fig . 6b ). Identifying this subset allowed us to assess how proliferating T cells respond differently across tumor microenvironment, rather than treating T cells as a single coarse population. Download figure Open in new tab Figure 6: E ykthyr predicts a context-specific influence of YY1 on T-cell states within distinct tumor microenvironments. a . Spatial plot of a human metastatic melanoma sample, with cells annotated by E ykthyr , showing spatial separation into two distinct tumor compartments. b . Comparison of T-cell annotations from the original study and those derived from E ykthyr ’s embedding, revealing additional T-cell states identified by E ykthyr . c . Top TFs identified by E ykthyr ranked by their differential influence on Proliferating T Cells between the two tumor compartments. d . Heatmap displaying the context-specific influence of YY1 on metagene expression across T-cell states, as measured by the difference in YY1 influence between tumor compartments. e . Gene set enrichment analysis of genes with both a YY1 binding motif in their promoter and a cross-tumor difference in promoter accessibility in Proliferating T Cells. To uncover the transcriptional drivers of proliferating T cells, we used E ykthyr ’s in silico knockout framework. This analysis identified KLF4 and YY1 as key regulators with compartment-specific effects on proliferating T cells ( Fig . 6c ). KLF4 has been linked to tumor suppression and improved survival in human cancers [ 32 ], and it can either reverse CD8 T-cell exhaustion or limit T-cell proliferation, depending on context [ 33 ]. E ykthyr ’s predictions reflected this duality: in one compartment, KLF4 knockout increased predicted T-cell proliferative potential, while in the other, it appeared to have an inhibitory effect. Similarly, YY1 has been shown to promote or inhibit T-cell proliferation in a context-dependent manner [ 34 , 35 ]. Analyzing transitions in T cell state before and after an in silico YY1 knockout revealed distinct shifts in metagene expression – particularly among m3 and m4 – across compartments ( Fig . 6d ). These shifts correlated with differential chromatin accessibility at YY1 binding sites, suggesting that YY1 may alter proliferating T-cell function in a compartment-specific manner. To explore this further, we collected genes with a YY1 motif in their promoter region and assessed their chromatin accessibility between tumor compartments. Gene set enrichment analysis on genes with higher accessibility in the bottom tumor compartment showed enrichment for cell-cycle pathways ( Fig . 6e ), whereas no significant cell-cycle enrichment was observed in the top compartment, indicating enhanced proliferation from YY1 only in the bottom compartment. These findings illustrate how E ykthyr identifies TFs that regulate T-cell states in a compartment-specific manner, shaping tumor-immune interactions. Understanding how KLF4 and YY1 drive or restrain T-cell proliferation in distinct microenvironments may inform further immunotherapeutic strategies. Note that to further test the robustness of E ykthyr ’s GRN design, We also evaluated a modified, non-spatial version of E ykthyr on a zebrafish scRNA-seq dataset with in vivo knockout experiments. Results are provided in the Supplementary Note . Together, these analyses highlight how E ykthyr ’s in silico knockout simulations can reveal biologically meaningful regulators in both spatial and non-spatial settings. Discussion We introduced E ykthyr , a computational framework that integrates spatial transcriptomic and chromatin accessibility data to identify location-specific regulators of gene programs in complex tissues. By inferring TF activity directly from chromatin accessibility, E ykthyr overcomes high dropout and low transcript signals that often obscure key regulators. Across diverse applications – including embryonic brain development and compartment-specific T-cell states in melanoma – E ykthyr identified TFs with minimal gene expression but strong regulatory influence. Its in silico TF knockouts further reveal which cells or regions are most affected by a given TF perturbation, underscoring the spatial dependence of gene regulation. Compared to existing methods that rely on single-modality or non-spatial data, E yk - thyr is the first framework explicitly designed to leverage both spatial transcriptomics and chromatin accessibility, offering new insights into how local environments shape cell identity. Despite its strengths, E ykthyr has several limitations. The required paired spatial multiome data – an emerging but still not widely available resource – and our validations have focused on relatively discrete developmental and tumor contexts. Future work should expand to tissues with overlapping lineages or higher cellular complexity, such as adult organs or advanced disease states. Additionally, E ykthyr assumes a fixed spatial map, limiting its ability to simulate large-scale morphological changes such as compartment merging or regional emergence. Addressing these challenges will require more flexible spatial models and expanded datasets that better capture biological complexity. Several extensions could enhance E ykthyr ’s modeling capabilities. Currently, it assumes that metagene expression is a linear combination of TF activity. Yet transcriptional regulation is often non-linear: TFs can act cooperatively or exhibit threshold effects, and chromatin context or epigenetic states can modulate TF activity in complex, context-dependent ways that linear models cannot fully capture. Future version of E ykthyr could incorporate spatio-temporal modeling to analyze tissue samples collected over time. This would allow dynamic models to capture lagged TF effects, identifying how early regulatory signals shape later gene expression. Joint modeling across multiple samples or developmental stages could further improve robustness and reveal consistent regulatory trajectories. Adapting E yk - thyr to unpaired spatial transcriptomic and single-cell multiomic data represents another important direction, broadening applicability to more datasets. Integrating additional modalities, such as proteomics, or shifting from spot-level to true single-cell resolution may refine TF activity estimates, as each data type captures only part of a TF’s regulatory influence. Beyond model enhancement, E ykthyr ’s in silico TF perturbation framework offers a valuable tool for hypothesis generation. By predicting how TF activity shifts reshape tissue architecture and cell state transition, the model can help prioritize candidate regulators for targeted experimental validation. Methods such as PerturbView [ 36 ], Perturb-FISH [ 37 ], and Perturb-map [ 38 ] combine in situ molecular profiling with multiplex antibody staining to measure spatial perturbation effects [ 39 ]. Integrating E yk - thyr ’s predictions with these techniques could accelerate the discovery of key regulatory circuits and support iterative spatial perturbation experiments to uncover causal mechanisms of tissue organization. Overall, by pinpointing region-specific TF effects, E ykthyr provides a solid foundation for precision spatial gene regulatory studies and paves the way for targeted strategies to modulate TF activity for a wide range of tissue contexts in health and disease. Methods The E ykthyr model E ykthyr takes spatial multiome data as input and generates low-dimensional embeddings for both gene expression (as metagenes) and chromatin accessibility (as TF activity) for each cell or spot. Each embedding dimension is a linear combination of the input data, preserving interpretability – shifts in the embedding directly reflect changes in gene expression or chromatin accessibility. E ykthyr then learns the relationship between these embedding spaces for each cell by training a linear model based on the embeddings of spatially proximal cells within the tissue. It can then simulate how changes in TF activity would affect downstream metagene expression using these learned linear relationships. Constructing linear embedding space from spatial transcriptome data E ykthyr addresses spatial transcriptomic sparsity by generating low-dimensional metagene embeddings using P opari [ 17 ]. P opari takes preprocessed read counts (filtered, normalized, and log transformed with SCANPY [ 40 ]) and incorporates spatial information via a hidden Markov random field framework. In short, P opari identifies K distinct metagenes that capture the fundamental transcriptomic factors within one or more samples. A metagene is defined as a latent factor representing a combinatorial signature of coexpressed genes with spatial coherence. These factors are encoded in a G × K metagene matrix, D , where G is the total number of genes measured. Each cell is represented by a K -dimensional latent vector, M i , which describes its state in the metagene space, and its observed gene expression Y i is modeled as a nonnegative linear combination of the metagenes: P opari also estimates a K × K spatial affinity matrix Λ for each sample, quantifying the coexpression of metagene pairs among neighboring cells. The metagene matrix D , the latent embeddings M i , and the spatial affinity matrices Λ are jointly optimized via an alternating maximum a posteriori (MAP) procedure, yielding intuitive representations of latent spatial structure. Each column d k of D provides an interpretable ranked list of genes, with the top-ranked genes forming a spatially coherent module. This approach enables the capture of nuanced combinatorial effects within a denoised, low-dimensional embedding, facilitating spatially-aware interactions between TFs and downstream gene expression even in high-noise or sparse settings. Estimating transcription factor activity from spatial epigenome data To address limited expression of TFs, E ykthyr estimates TF activity directly from chromatin accessibility data using ArchR [ 41 ] and following the MISAR-seq data processing workflow [ 7 ]. Specifically, we generate Arrow files with 5kb-tile matrices and call peaks using MACS2 [ 42 ]. We then apply a latent semantic indexing (LSI) transformation to the most accessible peaks, followed by shared nearest neighbor clustering to form initial clusters. These clusters are then refined using spatial cell location information: cells with a different cell type than all of their neighbors are are reassigned based on the most enriched cluster among their k nearest neighbors. We then recalculate the most variable peaks within these refined clusters and use these features for the next iteration of LSI. This process is repeated until cluster assignments converge. Finally, we annotate peaks with TF motifs using ArchR’s addMotifAnnotations function, which scans peak sequences for TF binding motifs from the CIS-BP database [ 43 ], retaining only motif matches with p < 1 × 10 −5 to ensure statistically significance. Given the cell-by-peak matrix P and the peak-by-TF-motif matrix T , TF activity estimates for each cell are computed as: which gives the number of times each TF motif appears in a peak measured for a given cell. To account for motif prevalence, we normalize TF activity across all cells for each TF to the range [0, 1]. Determining TF influence on metagenes Conceptually, TF influence is modeled as a bipartite graph between TFs and metagenes, with directed edges from each TF to each metagene, and edge weights representing the influence of a TF on metagene expression. To compute these edge weights, we use a regularized linear model, fit separately for each cell using the metagene expression and TF activity of its n spatially proximal neighbors. We choose n to include a sufficient number of cells for the model while avoiding those too distant. Since spatial multiome datasets vary in resolutions (e.g., spot-based vs. single-cell), n is dataset-specific, though we found n = 100 to be effective across datasets analyzed. For each cell, we construct a neighborhood comprising spatially proximal cells – half from the same cell type, and half unrestricted. We train the model to predict the metagene expression matrix M i for the neighborhood of cell i based on the TF activity matrix A i and a learned weight matrix W i where β is a vector of biases (one per metagene). This formulation assumes potential regulatory influence of each TF on all metagenes. Although not all TF interactions are additive, it provides a simple model that is robust to high noise and sparsity inherent to spatial multiome data. Another advantage of this formulation is that it allows for interpretable simulations, as the relationship between gene expression, chromatin accessibility, metagene expression, and TF activity remains linear. To avoid overfitting, we employ L2 regularization via Ridge regression [ 44 ] with regularization strength α = 1, and further enhance the robustness of the model by bagging. Specifically, we train 10 estimators on bootstrap samples, each with 80% of randomly selected features, and combine their output to obtain a posterior distribution over the coefficients, where µ W and σ W are matrices of shape ( k × t )), with each µ k,t and σ k,t representing the mean and standard deviation for the effect of TF t on metagene k . This ensemble approach reduces variance by averaging the coefficients across multiple bootstrapped models, producing more robust estimate of TF influence on metagene expression. Simulating changes in TF activity After learning TF–metagene interactions, we simulate changes to TF activity to predict their impact on metagene expression. For each TF j , we set its activity to zero across all cells – simulating an experimental knockout – effectively removing its regulatory influence across the tissue. The adjusted metagene expression matrix for cell ,is computed as: where represents the change to A from setting TF j ’s activity to zero. By performing this tissue-wide in silico TF knockout, we aim to uncover how gene expression and cell identity might change in the absence of a specific TF. A TF knockout is more straightforward to simulate than upregulation or downregulation, as it is unclear how to realistically increase or decrease TF activity across all cells. While E ykthyr ’s simulated knockouts reliably indicate the direction of change in the metagene embedding space, the magnitude of change is often small due to strong regularization. Nevertheless, the relative ranking of TF effects remains robust for downstream tasks such as identifying developmental regulators. For downstream applications that require a more pronounced signal – such as visualizing cell identity changes – we apply a scaling factor to amplify magnitude shifts, enhancing clarity in visualizations and robustness in clustering analyses. For the visualizations of cell state transition ( Figs . 3g, 3h , 4c , and 5d ), we used a scaling factor of 10. Evaluating E ykthyr ’s performance To evaluate E ykthyr ’s predictions of TF influence, we developed visualization techniques, statistical tests, and scoring metrics to assess shifts in cell embeddings following simulated TF perturbations. Scoring known TFs using batch correction metrics Validating E ykthyr ’s predictions is challenging due to the limited availability of spatially-informed experimental TF knockouts. We leveraged the biological principle that cell types derived from common progenitors exhibit similar metagene expression patterns. Thus, increased similarity among these cell types after simulated TF knockout suggests a regulatory role for that TF in their differentiation. To assess this, we adapted batch correction metrics – traditionally used to evaluate cell mixing across datasets – to evaluate TF perturbation effects. Using metrics from the scib-metrics Python package [ 45 ], including average silhouette width, graph iLISI, and principal component regression, we quantified the extent to which differentiated cells mix in the embedding space following simulated TF knockouts. These metrics yield scores reflecting each TF’s predicted impact on differentiation. We ranked TFs based on these scores and evaluated the statistical significance of their rankings using a t-test, focusing on TFs previously linked to differentiation. This approach provides a statistical framework for validating E ykthyr ’s biologically meaningful predictions. Calculating pseudotime from spatial distance Pseudotime provides an estimated measure of a cell’s progression through a developmental process and is commonly used to study differentiation. Each cell is assigned a value along a pseudotime axis, enabling analysis of gene regulation relative to differentiation trajectories. While pseudotime is typically computed from transcriptomic embeddings, many tissues exhibit spatial stability, where cells remain physically proximal to their progenitors, allowing spatial distance to serve as a proxy for pseudotime. Given a set of progenitor cells P and differentiated cells C , we define the pseudotime t ( c ) of each cell c ∈ C as its minimum distance to any progenitor: where d ( a, b ) is the Euclidean distance between a and b . This distance is normalized across all cells in C to the range [0, 1]. This spatially defined pseudotime serves as a directional vector of differentiation within the tissue. To evaluate TF effects, we examine each cell’s nearest spatial neighbors and identify the neighbor it most resembles after a simulated TF knockout. This yields two directional vectors per cell: one for the differentiation trajectory d and one for the TF knockout effect s . We compute the magnitude of their inner product, |⟨ d, s ⟩|, and rank TFs by their projected impact on the differentiation process. Benchmarking against CellOracle We benchmarked E ykthyr against CellOracle using the MISAR-seq spatial multiome dataset, focusing on the E15.5-S2 slice, as well as the spatial ATAC–RNA-seq mouse embryo dataset. CellOracle was run using its standard preprocessing steps outlined in its tutorial, with spatial information omitted, aligning with its design as a non-spatial method. We note that the dataset likely contains more than one cell per spot, which poses additional challenges for CellOracle, as it is optimized for single-cell data. In preprocessing, we retained all highly variable genes and included all TFs used by E ykthyr for this dataset, excluding TFs without detected transcripts. CellOracle’s clustering was performed using its non-spatial method, producing clusters that were less refined than those generated by P opari but sufficient for this benchmark. For GRN construction, we tested both CellOracle’s default mouse GRN (derived from a large scATAC-seq dataset) and a sample-specific GRN built per its guidelines. The sample-specific GRN underperformed, likely due to sparse chromatin accessibility data and limited cell numbers, so we used the default GRN for our comparisons. To quantitatively compare performance, we applied the batch integration evaluation framework to the MISAR-seq dataset and the spatial pseudotime procedure to the spatial ATAC–RNA-seq dataset, as described above. Visualizations To analyze the downstream effects of TF knockouts, we used visualization techniques inspired by CellOracle and adapted from RNA velocity approaches in the Velocyto package [ 46 ]. CellOracle, designed for non-spatial scRNA-seq data, computes cell-state transition probabilities based on changes in gene expression. To visualize cell identity changes after setting TF activity to zero, CellOracle requires both (i) the predicted gene expression change and (ii) the gene expression–based cell embedding. For E ykthyr , we adapted this framework by replacing gene expression changes with changes in metagene expression. We generated a two-dimensional embedding of the metagenes using partition-based graph abstraction (PAGA) [ 47 ], which preserves the data’s topological structure. Both metagene embeddings and spatial cell locations were used as embedding spaces for downstream analysis. With these adaptations, we applied CellOracle’s transition probability framework to compute how each cell shifts relative to its neighbors in either spatial or metagene space. These probabilities were then converted into directional vectors and combined into a vector field using an averaging approach from Velocyto. To visualize metagene affinities, we used P opari ’s empirical correlation plotting functions. For assessing changes in Leiden clustering, we used P opari ’s function to identify the optimal resolution for a specified number of clusters. A detailed description of dataset-specific preprocessing, E ykthyr parameter selection, and analysis methods is provided in the Supplementary Note . Code Availability The source code for E ykthyr can be accessed at: https://github.com/gkrieg/eykthyr . Data Availability All datasets analyzed in this work are publicly available. The MISAR-seq mouse embryo dataset was obtained from ref. [ 7 ]; the whole mouse embryo dataset was obtained from ref. [ 8 ]; the human metastatic melanoma dataset was obtained from ref. [ 9 ] and the zebrafish scRNA-seq dataset was obtained from ref. [ 14 ]. Author Contributions Conceptualization, S.K., J.M.; Methodology, S.K., J.M; Software, S.K.; Investigation, S.K., E.H., J.M; Writing, S.K., J.M; Funding Acquisition, J.M. Competing Interests The authors declare no competing interests. Acknowledgment This work was supported, in part, by National Institutes of Health Common Fund 4D Nucleome Program grant UM1HG011593 (J.M.); National Institutes of Health Common Fund Cellular Senescence Network Program grant UH3CA268202 (J.M.); and National Institutes of Health grants R01HG007352 (J.M.), R01HG012303 (J.M.), R21DA061481 (J.M.), and U24HG012070 (J.M.). J.M. was additionally supported by the Ray and Stephanie Lane Professorship, a Guggenheim Fellowship from the John Simon Guggenheim Memorial Foundation, a Google Research Award, and a Single-Cell Biology Data Insights award from the Chan Zuckerberg Initiative. S.K. is a Lane Fellow. The funders had no role in study design, data collection and analysis, decision to publish or preparation of the manuscript. Funder Information Declared National Institutes of Health, https://ror.org/01cwqze88 , UM1HG011593 , UH3CA268202 References [1]. ↵ Badia-i Mompel , P. et al. Gene regulatory network inference in the era of single-cell multi-omics . Nature Reviews Genetics 24 , 739 – 754 ( 2023 ). OpenUrl CrossRef PubMed [2]. Larsson , L. , Frisén , J. & Lundeberg , J. Spatially resolved transcriptomics adds a new dimension to genomics . Nature Methods 18 , 15 – 18 ( 2021 ). OpenUrl CrossRef PubMed [3]. ↵ Gilmour , D. , Rembold , M. & Leptin , M. From morphogen to morphogenesis and back . Nature 541 , 311 – 320 ( 2017 ). OpenUrl CrossRef PubMed [4]. ↵ Vickovic , S. et al. SM-Omics is an automated platform for high-throughput spatial multi-omics . Nature Communications 13 , 795 ( 2022 ). OpenUrl CrossRef PubMed [5]. Li , X. et al. Inflammation and aging: signaling pathways and intervention therapies . Signal Transduction and Targeted Therapy 8 , 239 ( 2023 ). OpenUrl CrossRef [6]. ↵ Wang , L. et al. Primary cilia signaling in astrocytes mediates development and regional-specific functional specification . Nature Neuroscience 27 , 1708 – 1720 ( 2024 ). OpenUrl CrossRef PubMed [7]. ↵ Jiang , F. et al. Simultaneous profiling of spatial gene expression and chromatin accessibility during mouse brain development . Nature Methods 20 , 1048 – 1057 ( 2023 ). OpenUrl CrossRef PubMed [8]. ↵ Zhang , D. et al. Spatial epigenome–transcriptome co-profiling of mammalian tissues . Nature 616 , 113 – 122 ( 2023 ). OpenUrl CrossRef PubMed [9]. ↵ Russell , A. J. C. et al. Slide-tags enables single-nucleus barcoding for multimodal spatial genomics . Nature 625 , 101 – 109 ( 2024 ). OpenUrl CrossRef PubMed [10]. ↵ Aibar , S. et al. SCENIC: single-cell regulatory network inference and clustering . Nature Methods 14 , 1083 – 1086 ( 2017 ). OpenUrl CrossRef PubMed [11]. Bravo González-Blas , C. et al. SCENIC+: single-cell multiomic inference of enhancers and gene regulatory networks . Nature Methods 20 , 1355 – 1367 ( 2023 ). OpenUrl CrossRef PubMed [12]. Kartha , V. K. et al. Functional inference of gene regulation using single-cell multi-omics . Cell Genomics 2 , 100166 ( 2022 ). OpenUrl CrossRef PubMed [13]. Fleck , J. S. et al. Inferring and perturbing cell fate regulomes in human brain organoids . Nature 621 , 365 – 372 ( 2023 ). OpenUrl CrossRef PubMed [14]. ↵ Kamimoto , K. et al. Dissecting cell identity via network inference and in silico gene perturbation . Nature 614 , 742 – 751 ( 2023 ). OpenUrl CrossRef PubMed [15]. Osorio , D. et al. scTenifoldKnk: An efficient virtual knockout tool for gene function predictions via single-cell gene regulatory network perturbation . Patterns 3 , 100434 ( 2022 ). OpenUrl CrossRef PubMed [16]. ↵ Zhang , S. et al. Inference of cell type-specific gene regulatory networks on cell lineages from single cell omic datasets . Nature Communications 14 , 3064 ( 2023 ). OpenUrl CrossRef PubMed [17]. ↵ Alam , S. et al. POPARI: Modeling multisample variation in spatial transcriptomics . bioRxiv 2025 – 05 ( 2025 ). [18]. ↵ Lee , D. & Seung , H. S. Algorithms for non-negative matrix factorization . Advances in Neural Information Processing Systems 13 ( 2000 ). [19]. ↵ Chidester , B. , Zhou , T. , Alam , S. & Ma , J. SPICEMIX enables integrative single-cell spatial modeling of cell identity . Nature Genetics 55 , 78 – 88 ( 2023 ). OpenUrl CrossRef PubMed [20]. ↵ Ypsilanti , A. R. & Rubenstein , J. L. Mammalian Cortical Regional Specification , 443 – 475 ( John Wiley & Sons, Ltd , 2023 ). [21]. ↵ Ramos , C. , Fernández-Llebrez , P. , Bach , A. , Robert , B. & Soriano , E. Msx1 disruption leads to diencephalon defects and hydrocephalus . Developmental Dynamics 230 , 446 – 460 ( 2004 ). OpenUrl CrossRef PubMed Web of Science [22]. ↵ Ramos , C. & Robert , B. msh/msx gene family in neural development . Trends in Genetics 21 , 624 – 632 ( 2005 ). OpenUrl CrossRef PubMed Web of Science [23]. ↵ Moore , J. E. et al. Expanded encyclopaedias of dna elements in the human and mouse genomes . Nature 583 , 699 – 710 ( 2020 ). OpenUrl CrossRef PubMed [24]. ↵ Falls , D. L. Neuregulins: functions, forms, and signaling strategies . The EGF Receptor Family 15 – 31 ( 2003 ). [25]. ↵ Song , L. , Chen , W. , Hou , J. , Guo , M. & Yang , J. Spatially resolved mapping of cells associated with human complex traits . Nature 1 – 10 ( 2025 ). [26]. ↵ Meyer , D. et al. Isoform-specific expression and function of neuregulin . Development 124 , 3575 – 3586 ( 1997 ). OpenUrl Abstract [27]. ↵ Liu , A. & Niswander , L. A. Bone morphogenetic protein signalling and vertebrate nervous system development . Nature Reviews Neuroscience 6 , 945 – 954 ( 2005 ). OpenUrl CrossRef PubMed Web of Science [28]. ↵ Maness , P. F. & Schachner , M. Neural recognition molecules of the immunoglobulin superfamily: signaling transducers of axon guidance and neuronal migration . Nature Neuroscience 10 , 19 – 26 ( 2007 ). OpenUrl CrossRef PubMed Web of Science [29]. ↵ Hevner , R. F. From radial glia to pyramidal-projection neuron: transcription factor cascades in cerebral cortex development . Molecular Neurobiology 33 , 33 – 50 ( 2006 ). OpenUrl CrossRef PubMed Web of Science [30]. ↵ Nakamura , Y. et al. The bhlh gene hes1 as a repressor of the neuronal commitment of cns stem cells . Journal of Neuroscience 20 , 283 – 293 ( 2000 ). OpenUrl Abstract / FREE Full Text [31]. ↵ de Almeida Magalhães , T. et al. Notch pathway in ependymoma rela-fused subgroup: upregulation and association with cancer stem cells markers expression . Cancer Gene Therapy 27 , 509 – 512 ( 2020 ). OpenUrl CrossRef PubMed [32]. ↵ Nah , J. & Seong , R. H. Krüppel-like factor 4 regulates the cytolytic effector function of exhausted CD8 T cells . Science Advances 8 , eadc9346 ( 2022 ). OpenUrl CrossRef PubMed [33]. ↵ Hao , Z. et al. K48-linked KLF4 ubiquitination by E3 ligase mule controls T-cell proliferation and cell cycle progression . Nature Communications 8 , 14003 ( 2017 ). OpenUrl CrossRef PubMed [34]. ↵ Hwang , S. S. et al. YY1 inhibits differentiation and function of regulatory t cells by blocking foxp3 expression and activity . Nature Communications 7 , 10789 ( 2016 ). OpenUrl CrossRef PubMed [35]. ↵ Hwang , S. S. et al. Transcription factor yy1 is essential for regulation of the th2 cytokine locus and for th2 cell differentiation . Proceedings of the National Academy of Sciences 110 , 276 – 281 ( 2013 ). OpenUrl Abstract / FREE Full Text [36]. ↵ Kudo , T. et al. Highly multiplexed, image-based pooled screens in primary cells and tissues with perturbview . bioRxiv 2023 – 12 ( 2023 ). [37]. ↵ Binan , L. et al. Simultaneous CRISPR screening and spatial transcriptomics reveal intracellular, intercellular, and functional transcriptional circuits . Cell 188 , 2141 – 2158 ( 2025 ). OpenUrl CrossRef PubMed [38]. ↵ Dhainaut , M. et al. Spatial CRISPR genomics identifies regulators of the tumor microenvironment . Cell 185 , 1223 – 1239 .e20 ( 2022 ). OpenUrl CrossRef PubMed [39]. ↵ Rood , J. E. , Hupalowska , A. & Regev , A. Toward a foundation model of causal cell and tissue biology with a perturbation cell and tissue atlas . Cell 187 , 4520 – 4545 ( 2024 ). OpenUrl CrossRef PubMed [40]. ↵ Wolf , F. A. , Angerer , P. & Theis , F. J. SCANPY: large-scale single-cell gene expression data analysis . Genome Biology 19 , 15 ( 2018 ). OpenUrl CrossRef PubMed [41]. ↵ Granja , J. M. et al. ArchR is a scalable software package for integrative single-cell chromatin accessibility analysis . Nature Genetics 53 , 403 – 411 ( 2021 ). OpenUrl CrossRef PubMed [42]. ↵ Zhang , Y. et al. Model-based analysis of ChIP-Seq (MACS) . Genome Biology 9 , R137 ( 2008 ). OpenUrl CrossRef PubMed [43]. ↵ Weirauch , M. T. et al. Determination and inference of eukaryotic transcription factor sequence specificity . Cell 158 , 1431 – 1443 ( 2014 ). OpenUrl CrossRef PubMed Web of Science [44]. ↵ Hoerl , A. E. & Kennard , R. W. Ridge regression: Biased estimation for nonorthogonal problems . Technometrics 12 , 55 – 67 ( 1970 ). OpenUrl CrossRef Web of Science [45]. ↵ Luecken , M. D. et al. Benchmarking atlas-level data integration in single-cell genomics . Nature Methods 19 , 41 – 50 ( 2022 ). OpenUrl CrossRef PubMed [46]. ↵ La Manno , G. et al. RNA velocity of single cells . Nature 560 , 494 – 498 ( 2018 ). OpenUrl CrossRef PubMed [47]. ↵ Wolf , F. A. et al. PAGA: graph abstraction reconciles clustering with trajectory inference through a topology preserving map of single cells . Genome Biology 20 , 1 – 9 ( 2019 ). OpenUrl CrossRef PubMed View the discussion thread. Back to top Previous Next Posted May 23, 2025. Download PDF Supplementary Material Email Thank you for your interest in spreading the word about bioRxiv. NOTE: Your email address is requested solely to identify you as the sender of this article. Your Email * Your Name * Send To * Enter multiple addresses on separate lines or separate them with commas. You are going to email the following EYKTHYR reveals transcriptional regulators of spatial gene programs Message Subject (Your Name) has forwarded a page to you from bioRxiv Message Body (Your Name) thought you would like to see this page from the bioRxiv website. Your Personal Message CAPTCHA This question is for testing whether or not you are a human visitor and to prevent automated spam submissions. Share EYKTHYR reveals transcriptional regulators of spatial gene programs Spencer Krieger , Ellie Haber , Jian Ma bioRxiv 2025.05.19.654884; doi: https://doi.org/10.1101/2025.05.19.654884 Share This Article: Copy Citation Tools EYKTHYR reveals transcriptional regulators of spatial gene programs Spencer Krieger , Ellie Haber , Jian Ma bioRxiv 2025.05.19.654884; doi: https://doi.org/10.1101/2025.05.19.654884 Citation Manager Formats BibTeX Bookends EasyBib EndNote (tagged) EndNote 8 (xml) Medlars Mendeley Papers RefWorks Tagged Ref Manager RIS Zotero Tweet Widget Facebook Like Google Plus One Subject Area Bioinformatics Subject Areas All Articles Animal Behavior and Cognition (7617) Biochemistry (17633) Bioengineering (13856) Bioinformatics (41841) Biophysics (21399) Cancer Biology (18529) Cell Biology (25422) Clinical Trials (138) Developmental Biology (13352) Ecology (19860) Epidemiology (2067) Evolutionary Biology (24281) Genetics (15582) Genomics (22461) Immunology (17700) Microbiology (40295) Molecular Biology (17140) Neuroscience (88413) Paleontology (666) Pathology (2823) Pharmacology and Toxicology (4813) Physiology (7632) Plant Biology (15107) Scientific Communication and Education (2042) Synthetic Biology (4284) Systems Biology (9808) Zoology (2267)