Contextual AI models for single-cell protein biology.

Understanding protein function and developing molecular therapies require deciphering the cell types in which proteins act as well as the interactions between proteins. However, modeling protein interactions across biological contexts remains challenging for existing algorithms. Here we introduce PINNACLE, a geometric deep learning approach that generates context-aware protein representations. Leveraging a multiorgan single-cell atlas, PINNACLE learns on contextualized protein interaction networks to produce 394,760 protein representations from 156 cell type contexts across 24 tissues. PINNACLE's embedding space reflects cellular and tissue organization, enabling zero-shot retrieval of the tissue hierarchy. Pretrained protein representations can be adapted for downstream tasks: enhancing 3D structure-based representations for resolving immuno-oncological protein interactions, and investigating drugs' effects across cell types. PINNACLE outperforms state-of-the-art models in nominating therapeutic targets for rheumatoid arthritis and inflammatory bowel diseases and pinpoints cell type contexts with higher predictive capability than context-free models. PINNACLE's ability to adjust its outputs on the basis of the context in which it operates paves the way for large-scale context-specific predictions in biology.

CD19-CD28: an affinity-optimized CD28 agonist for combination with glofitamab (CD20-TCB) as off-the-shelf immunotherapy.

ABSTRACT: Effective T-cell responses not only require the engagement of T-cell receptors (TCRs; "signal 1"), but also the availability of costimulatory signals ("signal 2"). T-cell bispecific antibodies (TCBs) deliver a robust signal 1 by engaging the TCR signaling component CD3ε, while simultaneously binding to tumor antigens. The CD20-TCB glofitamab redirects T cells to CD20-expressing malignant B cells. Although glofitamab exhibits strong single-agent efficacy, adding costimulatory signaling may enhance the depth and durability of T-cell-mediated tumor cell killing. We developed a bispecific CD19-targeted CD28 agonist (CD19-CD28), RG6333, to enhance the efficacy of glofitamab and similar TCBs by delivering signal 2 to tumor-infiltrating T cells. CD19-CD28 distinguishes itself from the superagonistic antibody TGN1412, because its activity requires the simultaneous presence of a TCR signal and CD19 target binding. This is achieved through its engineered format incorporating a mutated Fc region with abolished FcγR and C1q binding, CD28 monovalency, and a moderate CD28 binding affinity. In combination with glofitamab, CD19-CD28 strongly increased T-cell effector functions in ex vivo assays using peripheral blood mononuclear cells and spleen samples derived from patients with lymphoma and enhanced glofitamab-mediated regression of aggressive lymphomas in humanized mice. Notably, the triple combination of glofitamab with CD19-CD28 with the costimulatory 4-1BB agonist, CD19-4-1BBL, offered substantially improved long-term tumor control over glofitamab monotherapy and respective duplet combinations. Our findings highlight CD19-CD28 as a safe and highly efficacious off-the-shelf combination partner for glofitamab, similar TCBs, and other costimulatory agonists. CD19-CD28 is currently in a phase 1 clinical trial in combination with glofitamab. This trial was registered at www.clinicaltrials.gov as #NCT05219513.

Assessment of network module identification across complex diseases.

Many bioinformatics methods have been proposed for reducing the complexity of large gene or protein networks into relevant subnetworks or modules. Yet, how such methods compare to each other in terms of their ability to identify disease-relevant modules in different types of network remains poorly understood. We launched the 'Disease Module Identification DREAM Challenge', an open competition to comprehensively assess module identification methods across diverse protein-protein interaction, signaling, gene co-expression, homology and cancer-gene networks. Predicted network modules were tested for association with complex traits and diseases using a unique collection of 180 genome-wide association studies. Our robust assessment of 75 module identification methods reveals top-performing algorithms, which recover complementary trait-associated modules. We find that most of these modules correspond to core disease-relevant pathways, which often comprise therapeutic targets. This community challenge establishes biologically interpretable benchmarks, tools and guidelines for molecular network analysis to study human disease biology.

MONET: a toolbox integrating top-performing methods for network modularization.

SUMMARY: We define a disease module as a partition of a molecular network whose components are jointly associated with one or several diseases or risk factors thereof. Identification of such modules, across different types of networks, has great potential for elucidating disease mechanisms and establishing new powerful biomarkers. To this end, we launched the 'Disease Module Identification (DMI) DREAM Challenge', a community effort to build and evaluate unsupervised molecular network modularization algorithms. Here, we present MONET, a toolbox providing easy and unified access to the three top-performing methods from the DMI DREAM Challenge for the bioinformatics community. AVAILABILITY AND IMPLEMENTATION: MONET is a command line tool for Linux, based on Docker and Singularity containers; the core algorithms were written in R, Python, Ada and C++. It is freely available for download at https://github.com/BergmannLab/MONET.git. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Tissue-specific regulatory circuits reveal variable modular perturbations across complex diseases.

Mapping perturbed molecular circuits that underlie complex diseases remains a great challenge. We developed a comprehensive resource of 394 cell type- and tissue-specific gene regulatory networks for human, each specifying the genome-wide connectivity among transcription factors, enhancers, promoters and genes. Integration with 37 genome-wide association studies (GWASs) showed that disease-associated genetic variants--including variants that do not reach genome-wide significance--often perturb regulatory modules that are highly specific to disease-relevant cell types or tissues. Our resource opens the door to systematic analysis of regulatory programs across hundreds of human cell types and tissues (http://regulatorycircuits.org).

Genome-Wide Association between Transcription Factor Expression and Chromatin Accessibility Reveals Regulators of Chromatin Accessibility.

To better understand genome regulation, it is important to uncover the role of transcription factors in the process of chromatin structure establishment and maintenance. Here we present a data-driven approach to systematically characterise transcription factors that are relevant for this process. Our method uses a linear mixed modelling approach to combine datasets of transcription factor binding motif enrichments in open chromatin and gene expression across the same set of cell lines. Applying this approach to the ENCODE dataset, we confirm already known and imply numerous novel transcription factors that play a role in the establishment or maintenance of open chromatin. In particular, our approach rediscovers many factors that have been annotated as pioneer factors.

Fast and Rigorous Computation of Gene and Pathway Scores from SNP-Based Summary Statistics.

Integrating single nucleotide polymorphism (SNP) p-values from genome-wide association studies (GWAS) across genes and pathways is a strategy to improve statistical power and gain biological insight. Here, we present Pascal (Pathway scoring algorithm), a powerful tool for computing gene and pathway scores from SNP-phenotype association summary statistics. For gene score computation, we implemented analytic and efficient numerical solutions to calculate test statistics. We examined in particular the sum and the maximum of chi-squared statistics, which measure the strongest and the average association signals per gene, respectively. For pathway scoring, we use a modified Fisher method, which offers not only significant power improvement over more traditional enrichment strategies, but also eliminates the problem of arbitrary threshold selection inherent in any binary membership based pathway enrichment approach. We demonstrate the marked increase in power by analyzing summary statistics from dozens of large meta-studies for various traits. Our extensive testing indicates that our method not only excels in rigorous type I error control, but also results in more biologically meaningful discoveries.

Predictive regulatory models in Drosophila melanogaster by integrative inference of transcriptional networks.

Gaining insights on gene regulation from large-scale functional data sets is a grand challenge in systems biology. In this article, we develop and apply methods for transcriptional regulatory network inference from diverse functional genomics data sets and demonstrate their value for gene function and gene expression prediction. We formulate the network inference problem in a machine-learning framework and use both supervised and unsupervised methods to predict regulatory edges by integrating transcription factor (TF) binding, evolutionarily conserved sequence motifs, gene expression, and chromatin modification data sets as input features. Applying these methods to Drosophila melanogaster, we predict ∼300,000 regulatory edges in a network of ∼600 TFs and 12,000 target genes. We validate our predictions using known regulatory interactions, gene functional annotations, tissue-specific expression, protein-protein interactions, and three-dimensional maps of chromosome conformation. We use the inferred network to identify putative functions for hundreds of previously uncharacterized genes, including many in nervous system development, which are independently confirmed based on their tissue-specific expression patterns. Last, we use the regulatory network to predict target gene expression levels as a function of TF expression, and find significantly higher predictive power for integrative networks than for motif or ChIP-based networks. Our work reveals the complementarity between physical evidence of regulatory interactions (TF binding, motif conservation) and functional evidence (coordinated expression or chromatin patterns) and demonstrates the power of data integration for network inference and studies of gene regulation at the systems level.

Wisdom of crowds for robust gene network inference.

Reconstructing gene regulatory networks from high-throughput data is a long-standing challenge. Through the Dialogue on Reverse Engineering Assessment and Methods (DREAM) project, we performed a comprehensive blind assessment of over 30 network inference methods on Escherichia coli, Staphylococcus aureus, Saccharomyces cerevisiae and in silico microarray data. We characterize the performance, data requirements and inherent biases of different inference approaches, and we provide guidelines for algorithm application and development. We observed that no single inference method performs optimally across all data sets. In contrast, integration of predictions from multiple inference methods shows robust and high performance across diverse data sets. We thereby constructed high-confidence networks for E. coli and S. aureus, each comprising ~1,700 transcriptional interactions at a precision of ~50%. We experimentally tested 53 previously unobserved regulatory interactions in E. coli, of which 23 (43%) were supported. Our results establish community-based methods as a powerful and robust tool for the inference of transcriptional gene regulatory networks.

Contextualizing protein representations using deep learning on protein networks and single-cell data.

Understanding protein function and developing molecular therapies require deciphering the cell types in which proteins act as well as the interactions between proteins. However, modeling protein interactions across diverse biological contexts, such as tissues and cell types, remains a significant challenge for existing algorithms. We introduce Pinnacle, a flexible geometric deep learning approach that is trained on contextualized protein interaction networks to generate context-aware protein representations. Leveraging a human multi-organ single-cell transcriptomic atlas, Pinnacle provides 394,760 protein representations split across 156 cell type contexts from 24 tissues and organs. Pinnacle's contextualized representations of proteins reflect cellular and tissue organization and Pinnacle's tissue representations enable zero-shot retrieval of the tissue hierarchy. Pretrained Pinnacle's protein representations can be adapted for downstream tasks: to enhance 3D structure-based protein representations for important protein interactions in immuno-oncology (PD-1/PD-L1 and B7-1/CTLA-4) and to study the effects of drugs across cell type contexts. Pinnacle outperforms state-of-the-art, yet context-free, models in nominating therapeutic targets for rheumatoid arthritis and inflammatory bowel diseases, and can pinpoint cell type contexts that predict therapeutic targets better than context-free models (29 out of 156 cell types in rheumatoid arthritis; 13 out of 152 cell types in inflammatory bowel diseases). Pinnacle is a graph-based contextual AI model that dynamically adjusts its outputs based on biological contexts in which it operates.

Phase 2 study of the antitumour activity and safety of simlukafusp alfa (FAP-IL2v) combined with atezolizumab in patients with recurrent and/or metastatic cervical squamous cell carcinoma.

BACKGROUND: Simlukafusp alfa (FAP-IL2v) is an immune cytokine engineered to selectively promote immune responses in the tumour microenvironment. We evaluated the antitumour activity and safety of FAP-IL2v plus atezolizumab in recurrent and/or metastatic cervical squamous cell carcinoma (SCC) in a phase 2 basket study (NCT03386721). METHODS: Patients with confirmed metastatic, persistent or recurrent cervical SCC who had progressed on ≥1 anti-cancer therapy and had measurable disease were enrolled. FAP-IL2v 10 mg was administered once every 3 weeks (Q3W) or once weekly (QW) for 4 weeks then once every 2 weeks (Q2W) with the corresponding Q3W or Q2W atezolizumab regimens. The primary endpoint was objective response rate by investigator assessment. FINDINGS: Forty-eight patients were enrolled (Q3W: n = 47; QW/Q2W: n = 1). Among 45 response evaluable patients, objective responses occurred in 12 patients (27%; CI 16.0-41.0), including 3 complete and 9 partial responses. Responses occurred in 6/19 PD-L1 positive patients (32%; 95% CI 15.4-54.0) and 5/24 PD-L1 negative patients (21%; 95% CI 9.2-35.6). Median duration of response was 13.3 months (95% CI 7.6-NE). Median progression-free survival was 3.7 months (95% CI 3.3-9.0). Adverse events (AEs) were consistent with the known safety profile of each drug. AEs leading to withdrawal of either agent occurred in 6 patients (13%). Pronounced expansion and activation of natural killer and CD8 T cells in peripheral blood and increased tumour infiltration and inflammation were observed. INTERPRETATION: FAP-IL2v plus atezolizumab is clinically active and has manageable safety in patients with recurrent and/or metastatic cervical SCC. FUNDING: F. Hoffmann-La Roche Ltd.

Phase II Study to Determine the Antitumor Activity and Safety of Simlukafusp Alfa (FAP-IL2v) Combined with Atezolizumab in Esophageal Cancer.

PURPOSE: In this study, we report the results from the esophageal squamous cell carcinoma (SCC) cohort of a phase II, noncomparative, basket study evaluating the antitumor activity and safety of fibroblast activation protein-IL2 variant (FAP-IL2v) plus atezolizumab in patients with advanced/metastatic solid tumors (NCT03386721). PATIENTS AND METHODS: Eligible patients had an Eastern Cooperative Oncology Group performance status of 0 to 1; measurable metastatic, persistent, or recurrent esophageal SCC; progression on ≥1 prior therapy; and were checkpoint inhibitor-naïve. Patients received FAP-IL2v 10 mg plus atezolizumab 1,200 mg intravenously every 3 weeks, or FAP-IL2v weekly for 4 weeks and then every 2 weeks plus atezolizumab 840 mg intravenously every 2 weeks. The primary endpoint was investigator-assessed objective response rate (ORR). RESULTS: In the response-evaluable population (N = 34), the best confirmed ORR was 20.6% [95% confidence interval (CI), 10.4-36.8], with a complete response seen in 1 patient and partial responses in 6 patients. The disease control rate was 44.1% (complete response = 2.9%; partial response = 17.6%; stable disease = 23.5%), and the median duration of response was 10.1 mon/ths (95% CI, 5.6-26.7). The median progression-free survival was 1.9 months (95% CI, 1.8-3.7). Analysis of response by PDL1 expression (Ventana SP263) resulted in an ORR of 26.7% for patients with PDL1-positive tumors (tumor area positivity cutoff ≥1%; n = 15) and 7.1% for patients with PDL1-negative tumors (tumor area positivity cutoff <1%; n = 14). Overall, the treatment combination was tolerable, and adverse events were consistent with the known safety profiles of each drug. CONCLUSIONS: FAP-IL2v plus atezolizumab demonstrated clinical activity and was tolerable in patients with previously treated esophageal SCC.

Revealing strengths and weaknesses of methods for gene network inference.

Numerous methods have been developed for inferring gene regulatory networks from expression data, however, both their absolute and comparative performance remain poorly understood. In this paper, we introduce a framework for critical performance assessment of methods for gene network inference. We present an in silico benchmark suite that we provided as a blinded, community-wide challenge within the context of the DREAM (Dialogue on Reverse Engineering Assessment and Methods) project. We assess the performance of 29 gene-network-inference methods, which have been applied independently by participating teams. Performance profiling reveals that current inference methods are affected, to various degrees, by different types of systematic prediction errors. In particular, all but the best-performing method failed to accurately infer multiple regulatory inputs (combinatorial regulation) of genes. The results of this community-wide experiment show that reliable network inference from gene expression data remains an unsolved problem, and they indicate potential ways of network reconstruction improvements.

GeneNetWeaver: in silico benchmark generation and performance profiling of network inference methods.

MOTIVATION: Over the last decade, numerous methods have been developed for inference of regulatory networks from gene expression data. However, accurate and systematic evaluation of these methods is hampered by the difficulty of constructing adequate benchmarks and the lack of tools for a differentiated analysis of network predictions on such benchmarks. RESULTS: Here, we describe a novel and comprehensive method for in silico benchmark generation and performance profiling of network inference methods available to the community as an open-source software called GeneNetWeaver (GNW). In addition to the generation of detailed dynamical models of gene regulatory networks to be used as benchmarks, GNW provides a network motif analysis that reveals systematic prediction errors, thereby indicating potential ways of improving inference methods. The accuracy of network inference methods is evaluated using standard metrics such as precision-recall and receiver operating characteristic curves. We show how GNW can be used to assess the performance and identify the strengths and weaknesses of six inference methods. Furthermore, we used GNW to provide the international Dialogue for Reverse Engineering Assessments and Methods (DREAM) competition with three network inference challenges (DREAM3, DREAM4 and DREAM5). AVAILABILITY: GNW is available at http://gnw.sourceforge.net along with its Java source code, user manual and supporting data. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online. CONTACT: dario.floreano@epfl.ch.

DUSP4 protects BRAF- and NRAS-mutant melanoma from oncogene overdose through modulation of MITF.

MAPK inhibitors (MAPKi) remain an important component of the standard of care for metastatic melanoma. However, acquired resistance to these drugs limits their therapeutic benefit. Tumor cells can become refractory to MAPKi by reactivation of ERK. When this happens, tumors often become sensitive to drug withdrawal. This drug addiction phenotype results from the hyperactivation of the oncogenic pathway, a phenomenon commonly referred to as oncogene overdose. Several feedback mechanisms are involved in regulating ERK signaling. However, the genes that serve as gatekeepers of oncogene overdose in mutant melanoma remain unknown. Here, we demonstrate that depletion of the ERK phosphatase, DUSP4, leads to toxic levels of MAPK activation in both drug-naive and drug-resistant mutant melanoma cells. Importantly, ERK hyperactivation is associated with down-regulation of lineage-defining genes including MITF Our results offer an alternative therapeutic strategy to treat mutant melanoma patients with acquired MAPKi resistance and those unable to tolerate MAPKi.

BET Inhibition Enhances TNF-Mediated Antitumor Immunity.

Targeting chromatin binding proteins and modifying enzymes can concomitantly affect tumor cell proliferation and survival, as well as enhance antitumor immunity and augment cancer immunotherapies. By screening a small-molecule library of epigenetics-based therapeutics, BET (bromo- and extra-terminal domain) inhibitors (BETi) were identified as agents that sensitize tumor cells to the antitumor activity of CD8+ T cells. BETi modulated tumor cells to be sensitized to the cytotoxic effects of the proinflammatory cytokine TNF. By preventing the recruitment of BRD4 to p65-bound cis-regulatory elements, BETi suppressed the induction of inflammatory gene expression, including the key NF-κB target genes BIRC2 (cIAP1) and BIRC3 (cIAP2). Disruption of prosurvival NF-κB signaling by BETi led to unrestrained TNF-mediated activation of the extrinsic apoptotic cascade and tumor cell death. Administration of BETi in combination with T-cell bispecific antibodies (TCB) or immune-checkpoint blockade increased bystander killing of tumor cells and enhanced tumor growth inhibition in vivo in a TNF-dependent manner. This novel epigenetic mechanism of immunomodulation may guide future use of BETi as adjuvants for immune-oncology agents.

Deciphering molecular and cellular ex vivo responses to bispecific antibodies PD1-TIM3 and PD1-LAG3 in human tumors.

BACKGROUND: Next-generation cancer immunotherapies are designed to broaden the therapeutic repertoire by targeting new immune checkpoints including lymphocyte-activation gene 3 (LAG-3) and T cell immunoglobulin and mucin-domain containing-3 (TIM-3). Yet, the molecular and cellular mechanisms by which either receptor functions to mediate its inhibitory effects are still poorly understood. Similarly, little is known on the differential effects of dual, compared with single, checkpoint inhibition. METHODS: We here performed in-depth characterization, including multicolor flow cytometry, single cell RNA sequencing and multiplex supernatant analysis, using tumor single cell suspensions from patients with cancer treated ex vivo with novel bispecific antibodies targeting programmed cell death protein 1 (PD-1) and TIM-3 (PD1-TIM3), PD-1 and LAG-3 (PD1-LAG3), or with anti-PD-1. RESULTS: We identified patient samples which were responsive to PD1-TIM3, PD1-LAG3 or anti-PD-1 using an in vitro approach, validated by the analysis of 659 soluble proteins and enrichment for an anti-PD-1 responder signature. We found increased abundance of an activated (HLA-DR+CD25+GranzymeB+) CD8+ T cell subset and of proliferating CD8+ T cells, in response to bispecific antibody or anti-PD-1 treatment. Bispecific antibodies, but not anti-PD-1, significantly increased the abundance of a proliferating natural killer cell subset, which exhibited enrichment for a tissue-residency signature. Key phenotypic and transcriptional changes occurred in a PD-1+CXCL13+CD4+ T cell subset, in response to all treatments, including increased interleukin-17 secretion and signaling toward plasma cells. Interestingly, LAG-3 protein upregulation was detected as a unique pharmacodynamic effect mediated by PD1-LAG3, but not by PD1-TIM3 or anti-PD-1. CONCLUSIONS: Our in vitro system reliably assessed responses to bispecific antibodies co-targeting PD-1 together with LAG-3 or TIM-3 using patients' tumor infiltrating immune cells and revealed transcriptional and phenotypic imprinting by bispecific antibody formats currently tested in early clinical trials.

Discoidin Domain Receptors in Melanoma: Potential Therapeutic Targets to Overcome MAPK Inhibitor Resistance.

Melanoma is a highly malignant skin cancer with high propensity to metastasize and develop drug resistance, making it a difficult cancer to treat. Current therapies targeting BRAF (V600) mutations are initially effective, but eventually tumors overcome drug sensitivity and reoccur. This process is accomplished in part by reactivating alternate signaling networks that reinstate melanoma proliferative and survival capacity, mostly through reprogramming of receptor tyrosine kinase (RTK) signaling. Evidence indicates that the discoidin domain receptors (DDRs), a set of RTKs that signal in response to collagen, are part of the kinome network that confer drug resistance. We previously reported that DDR1 is expressed in melanomas, where it can promote tumor malignancy in mouse models of melanoma, and thus, DDR1 could be a promising target to overcome drug resistance. In this review, we summarize the current knowledge on DDRs in melanoma and their implication for therapy, with emphasis in resistance to MAPK inhibitors.

A Community Challenge for Inferring Genetic Predictors of Gene Essentialities through Analysis of a Functional Screen of Cancer Cell Lines.

We report the results of a DREAM challenge designed to predict relative genetic essentialities based on a novel dataset testing 98,000 shRNAs against 149 molecularly characterized cancer cell lines. We analyzed the results of over 3,000 submissions over a period of 4 months. We found that algorithms combining essentiality data across multiple genes demonstrated increased accuracy; gene expression was the most informative molecular data type; the identity of the gene being predicted was far more important than the modeling strategy; well-predicted genes and selected molecular features showed enrichment in functional categories; and frequently selected expression features correlated with survival in primary tumors. This study establishes benchmarks for gene essentiality prediction, presents a community resource for future comparison with this benchmark, and provides insights into factors influencing the ability to predict gene essentiality from functional genetic screens. This study also demonstrates the value of releasing pre-publication data publicly to engage the community in an open research collaboration.

Network deconvolution as a general method to distinguish direct dependencies in networks.

Recognizing direct relationships between variables connected in a network is a pervasive problem in biological, social and information sciences as correlation-based networks contain numerous indirect relationships. Here we present a general method for inferring direct effects from an observed correlation matrix containing both direct and indirect effects. We formulate the problem as the inverse of network convolution, and introduce an algorithm that removes the combined effect of all indirect paths of arbitrary length in a closed-form solution by exploiting eigen-decomposition and infinite-series sums. We demonstrate the effectiveness of our approach in several network applications: distinguishing direct targets in gene expression regulatory networks; recognizing directly interacting amino-acid residues for protein structure prediction from sequence alignments; and distinguishing strong collaborations in co-authorship social networks using connectivity information alone. In addition to its theoretical impact as a foundational graph theoretic tool, our results suggest network deconvolution is widely applicable for computing direct dependencies in network science across diverse disciplines.