Biology subjects

Meimetis, N.

Publications and source records attributed to Meimetis, N..

8 recordsLinked to original sources

bioRxiv2026

Interactome-based deep learning and path-level decomposition reveal TGFβ-IFNα competition in MASH

Dysregulated signaling drives Metabolic Dysfunction-Associated Steatohepatitis (MASH) through a complex crosstalk between pro- and anti-fibrotic cues. We previously observed potential competition between IFN-mediated JAK-STAT signaling and TGF{beta} in modulating MASH fibrosis, but the precise mechanism of that competition, and whether it can act directly on hepatocytes, is yet unknown. We used DT-LEMBAS, an interactome-based deep learning framework, training an ensemble of 50 models on drug-induced transcriptomics from primary human hepatocytes. The ensemble captured transcriptional changes induced by simulated receptor knockouts and aligned predicted profiles with known TGF{beta}- and IFN-induced phenotypes. Systematic simulations demonstrate their competitive modulation of MASH phenotype scores benchmarked against human cohorts. To extract mechanistic insights, we developed a gradient-based approach, where we computed the Jacobian of the used recurrent signaling network and applied chain rule decomposition to identify a four-node module of SMAD3, TP63, STAT2, and STAT3. Mechanistic hypotheses involving these canonical and non-canonical TF mediators were experimentally validated in primary hepatocyte spheroids. This work establishes a generalizable and scalable approach to decode large signaling neural networks into interpretable biological subnetworks.

systems biology↗

bioRxiv2026

scLEMBAS: Context-Aware Modeling of Signaling Pathway Activity at Single-Cell Resolution

Cells sense and integrate extracellular cues through intracellular signaling networks that reshape transcription factor activity to dictate cellular responses. Signaling activity is difficult to decipher: it is non-linear, and it contains extensive feedback and crosstalk. Furthermore, the same perturbation can elicit markedly different responses depending on context (e.g., cell type, disease state, and tissue microenvironment) such that identical stimuli produce diverse responses in multicellular populations. Consequently, there is a vast combinatorial space of complex interactions and context-dependent responses that necessitate computational models. Computational models of single-cell perturbation responses are demonstrated to predict cellular responses, but are often limited in mechanistic insight. Prior knowledge networks offer a route to bridge predictive capability and interpretability. Here we present scLEMBAS, a context-aware, gray-box neural network that models signaling pathway activity at single-cell resolution while preserving mechanistic grounding. scLEMBAS encodes a prior-knowledge network of protein-protein interactions as a recurrent neural network whose learnable edge weights correspond to signaling interaction strengths. It also captures context and individual cell variance through compositional bias terms. An adversarial approach allows the model to answer a single-cell counterfactual - what a given cells TF activity would be under a different perturbation or context - while involving mechanistic rather than simply relational information. Across two scRNA-seq datasets spanning single- and multi-perturbation settings, scLEMBAS accurately predicts out-of-distribution combinations of perturbation and context. Capturing population variance across individual cells enables the model to predict cell subtype specific perturbation responses, despite being agnostic to such labels. Beyond prediction, scLEMBAS learned parameters are biologically interpretable: learned edge weights carry information beyond network topology and "self-prune" spurious interactions, while the categorical bias nominates proteins associated with cell-type-specific perturbation states. Overall, scLEMBAS enables quantitative dissection of how signaling pathway activity is reshaped by perturbation within specific cellular contexts.

bioinformatics↗

bioRxiv2026

FlowTransOP: Distributional Translation of Omics Signatures via Constrained Deep Flow Matching

Observations from pre-clinical models rarely generalize to human patients, leading to many failures in clinical trials. Most existing methods cannot handle domains with non-overlapping features and no paired samples. Here, we developed FlowTransOP to translate biological observations across such domains without requiring 1-to-1 feature mappings and paired data, while providing a guideline for model selection across four translational regimes. We use flow matching to align full domain distributions in a pre-aligned latent space, with a structural regularization term that keeps similar conditions proximate after transformation. FlowTransOP remains competitive with gold-standard approaches requiring paired samples, but outperforms them when pairs become scarce (<35 pairs) or when cross-domain features are only moderately correlated (r<=0.58). Overall, FlowTransOP can translate perturbations between pre-clinical models and patients when direct correspondences are unavailable, enabling reliable therapeutic inference. As a proof-of-concept, we trained a foundational mouse-human transcriptomic map on ARCHS4 and applied it to liver disease predictions.

bioinformatics↗

bioRxiv2025

Joint linear modeling of transcriptomics and proteomics is predictive of cancer metastasis

A central goal of conducting omics measurements is to understand how molecular features inform higher-order cell- and tissue-level phenotypes. In particular, multi-omics offers insights into how information encoded by the genome is coordinated through biological layers, resulting in functional outputs1. Due to myriad post-transcriptional regulatory processes, the coordination between mRNA and protein cannot be simply reduced to gene-wise correlation. Yet, both modalities have been shown to serve as representations of biological state, and multi-omics integration has been used to improve these representations. Multi-omics approaches typically do not focus on how mRNA and protein features coordinate, but rather use the additional information for improved prediction or feature selection. Here, instead, we showed that standard linear machine learning models provide an understanding of transcriptomic and proteomic coordination in the context of a biological phenotype of interest, in this case cancer metastasis. We find that, in the context of metastasis, a select subset of proteomic features--reflecting a more concentrated signal relative to the broadly distributed transcriptomic signal--offers additional information to that encoded by transcriptomics, as demonstrated by improved model performance when integrating the two modalities and the relative feature importance of proteomics. Top features show a depletion of gene-product overlap across modalities, indicating that the model primarily leverages instances in which the two modalities are providing complementary information with respect to phenotype. However, in instances when both modalities are selected for a given gene product, there is high information consistency that synergistically bolsters phenotype prediction. Altogether, by using model fits that relate both modalities to phenotype, we observe a nuanced coordination of protein and mRNA, in which both modalities tend to provide consistent information about phenotype, yet benefits remain to incorporating a combination of both complementary and reinforcing signals across modalities.

systems biology↗

bioRxiv2025

Systems biology framework for rational design of operational conditions for in vitro / in vivo translation of microphysiological systems

Preclinical models are used extensively to study diseases and potential therapeutic treatments. Complex in vitro platforms incorporating human cellular components, known as microphysiological systems (MPS), can model cellular and microenvironmental features of diseased tissues. However, determining experimental conditions -- particularly biomolecular cues such as growth factors, cytokines, and matrix proteins -- providing the most effective translatability of MPS-generated information to in vivo human subject contexts is a major challenge. Here, using metabolic dysfunction-associated fatty liver disease (MAFLD) studied using the CNBio PhysioMimix as a case study, we developed a machine learning framework called Latent In Vitro to In Vivo Translation (LIV2TRANS) to ascertain how MPS data map to in vivo data, first sharpening translation insights and consequently elucidating experimental conditions that can further enhance translation capability. Our findings in this case study highlight TGF{beta} as a crucial cue for MPS translatability and indicate that adding JAK-STAT pathway perturbations via interferon stimuli could increase the predictive performance of this MPS in MAFLD studies. Finally, we developed an optimization approach that identified androgen and EGFR signaling as key for maximizing the capacity of this MPS to capture in vivo human biological information germane to MAFLD. More broadly, this work establishes a mathematically principled approach for identifying experimental conditions most beneficially capturing in vivo human-relevant molecular pathways and processes, generalizable to preclinical studies for a wide range of diseases and potential treatments.

systems biology↗

bioRxiv2023

Inference of drug off-target effects on cellular signaling using Interactome-Based Deep Learning

Many diseases emerge from dysregulated cellular signaling, and drugs are often designed to target specific nodes in cellular networks e.g. signaling proteins, or transcription factors. However, off-target effects are common and may ultimately result in failed clinical trials. Computational modeling of the cells transcriptional response to drugs could improve our understanding of their mechanisms of action. Here we develop such an approach based on ensembles of artificial neural networks, that simultaneously infer drug-target interactions and their downstream effects on intracellular signaling. Applied to gene expression data from different cell lines, it outperforms basic machine learning approaches in predicting transcription factors activity, while recovering most known drug-target interactions and inferring many new, which we validate in an independent dataset. As a case study, we explore the inferred interactions of the drug Lestaurtinib and its effects on downstream signaling. Beyond its intended target FLT3 the model predicts an inhibition of CDK2 that enhances downregulation of the cell cycle-critical transcription factor FOXM1, corroborating literature findings. Our approach can therefore enhance our understanding of drug signaling for therapeutic design.

systems biology↗

bioRxiv2023

Autoencoder Model for Translating Omics Signatures

The development of effective therapeutics and vaccines for human diseases requires a systematic understanding of human biology. While animal and in vitro culture models have successfully elucidated the molecular mechanisms of diseases in many studies, they yet fail to adequately recapitulate human biology as evidenced by the predominant likelihood of failure in clinical trials. To address this broadly important problem, we developed AutoTransOP, a neural network autoencoder framework to map omics profiles from designated species or cellular contexts into a global latent space, from which germane information can be mapped between different contexts. This approach performs as well or better than extant machine learning methods and can identify animal/culture-specific molecular features predictive of other contexts, without requiring homology matching. For an especially challenging test case, we successfully apply our framework to a set of inter-species vaccine serology studies, where no 1-1 mapping between human and non-human primate features exists.

bioinformatics↗

bioRxiv2021

DeepSNEM: Deep Signaling Network Embeddings for compound mechanism of action identification

MotivationThe analysis and comparison of compounds transcriptomic signatures can help elucidate a compounds Mechanism of Action (MoA) in a biological system. In order to take into account the complexity of the biological system, several computational methods have been developed that utilize prior knowledge of molecular interactions to create a signaling network representation that best explains the compounds effect. However, due to their complex structure, large scale datasets of compound-induced signaling networks and methods specifically tailored to their analysis and comparison are very limited. Our goal is to develop graph deep learning models that are optimized to transform compound-induced signaling networks into high-dimensional representations and investigate their relationship with their respective MoAs. ResultsWe created a new dataset of compound-induced signaling networks by applying the CARNIVAL network creation pipeline on the gene expression profiles of the CMap dataset. Furthermore, we developed a novel unsupervised graph deep learning pipeline, called deepSNEM, to encode the information in the compound-induced signaling networks in fixed-length high-dimensional representations. The core of deepSNEM is a graph transformer network, trained to maximize the mutual information between whole-graph and sub-graph representations that belong to similar perturbations. By clustering the deepSNEM embeddings, using the k-means algorithm, we were able to identify distinct clusters that are significantly enriched for mTOR, topoisomerase, HDAC and protein synthesis inhibitors respectively. Additionally, we developed a subgraph importance pipeline and identified important nodes and subgraphs that were found to be directly related to the most prevalent MoA of the assigned cluster. As a use case, deepSNEM was applied on compounds gene expression profiles from various experimental platforms (MicroArrays and RNA sequencing) and the results indicate that correct hypotheses can be generated regarding their MoA. Availability and ImplementationThe source code and pre-trained deepSNEM models are available at https://github.com/BioSysLab/deepSNEM. ContactEmail for correspondence: leo@mail.ntua.gr. Supplementary informationAccompanying supplementary material are available online.

bioinformatics↗