Biology subjects

Saraswat, M.

Publications and source records attributed to Saraswat, M..

8 recordsLinked to original sources

bioRxiv2026

Mapping trans-eQTLs at single-cell resolution using Latent Interaction Variational Inference.

Single-cell expression quantitative trait loci (eQTL) studies hold promise for linking genetic variants to changes in gene expression in individual cells, thereby offering deeper insights into the genetic basis of human traits and diseases. Existing methods for mapping cis effects rely on predefined cell types, pseudobulk aggregation, or single-gene association tests, and are thus limited in their ability to detect complex trans effects that act across gene networks and cell-type subpopulations. Here, we present Latent Interaction Variational Inference (LIVI), an interpretable deep learning framework for efficient mapping of trans genetic effects at cellular resolution. LIVI employs a structured sparse variational autoencoder architecture to decompose observed gene expression profiles into cell-state- and donor-specific variation. The model enables efficient donor-level association testing while retaining single-cell resolution and interpretation. Applied to the OneK1K dataset, LIVI discovered a greater number of trans-eQTLs than alternative latent variable methods, yielded trans-eQTLs missed by conventional single-gene testing strategies and revealed the cell types and genes implicated in polygenic risk for autoimmune diseases. All in all, we demonstrate that LIVI is a powerful approach for linking genetics to gene regulation and through to human traits.

genomics↗

bioRxiv2025

High-throughput single-cell CRISPRi screens stratify neurodevelopmental functions of schizophrenia-associated genes

Schizophrenia is a complex neuropsychiatric disorder with strong genetic underpinnings, yet the molecular mechanisms linking genetic risk to disrupted brain development remain poorly understood. Transcription factors (TFs) and chromatin regulators (CRs) are increasingly implicated in neuropsychiatric disorders, where their dysregulation may disrupt neurodevelopmental programs. Despite this, systematic functional interrogation in human models has been limited. Here, we combine pooled CRISPR interference (CRISPRi) screens with high- throughput single-cell multiomic profiling in hiPSC-derived neural progenitors and neurons to functionally assess 65 schizophrenia-associated genes. Based on public datasets and literature review, we selected 55 TFs and CRs, along with ten additional risk genes whose loss-of-function has been linked to schizophrenia. Our single-cell CRISPRi readouts revealed that perturbations in TFs and CRs converge on disrupting neurodevelopmental timing. CRISPRi of several factors delayed neural differentiation, whereas others, such as the knockdown of MCRS1, drove precocious neural commitment. Validation screens combined with cell cycle and metabolic indicators confirmed the differentiation-restricting or -promoting roles of these TFs and CRs. Multimodal trajectory analysis uncovered discrete transcriptional and epigenomic states representing delayed and accelerated neurodevelopment, enriched for schizophrenia GWAS loci and disease-relevant pathways. Gene regulatory network (GRN) inference identified TCF4 and ZEB1 as critical mediators opposing the neural differentiation trajectory. Functional overexpression of these TFs followed by chromatin profiling demonstrated that TCF4 restrains, while ZEB1 promotes, neural differentiation in a stage-specific and competitive manner. Furthermore, we show that MCRS1 represses ZEB1 expression, positioning MCRS1 as a key brake on premature neurodevelopment. Together, our study establishes a scalable framework that integrates genetic perturbation, single- cell multiomics, and GRN modeling to functionally annotate disease-linked genes. We reveal convergent regulatory axes that underlie altered neurodevelopmental timing in schizophrenia, offering mechanistic insights into how chromatin misregulation contributes to disease pathogenesis.

neuroscience↗

bioRxiv2025

A spatiotemporal cancer cell trajectory underlies glioblastoma heterogeneity

Cancer cells display highly heterogeneous and plastic states in glioblastoma, an incurable brain tumour. However, how these malignant states arise and whether they follow defined cellular trajectories across tumours is poorly understood. Here, we generated a deep single cell and spatial multi-omic atlas of human glioblastoma that pairs transcriptomic, epigenomic and genomic profiling of 12 tumours across multiple regions. We identify that glioblastoma heterogeneity is driven by spatially-patterned transitions of cancer cells from developmental-like states towards those defined by a glial injury response and hypoxia. This cellular trajectory regionalises tumours into distinct tissue niches and manifests in a molecularly conserved manner across tumours as well as genetically distinct tumour subclones. Moreover, using a new deep learning framework to map cancer cell states jointly with clones in situ, we show that tumour subclones are finely spatially intermixed through glioblastoma tissue niches. Finally, we show that this cancer cell trajectory is intimately linked to myeloid heterogeneity and unfolds across regionalised myeloid signalling environments. Our findings define a stereotyped trajectory of cancer cells in glioblastoma and unify glioblastoma tumour heterogeneity into a tractable cellular and tissue framework.

cancer biology↗

bioRxiv2025

Decoding Plasticity Regulators and Transition Trajectories in Glioblastoma with Single-cell Multiomics

Glioblastoma (GB) is one of the most lethal human cancers, marked by profound intratumoral heterogeneity and near-universal treatment resistance. Cellular plasticity, the capacity of cancer cells to transition between phenotypic states, drives GB progression and resistance. However, the regulatory logic that permits or restricts specific state transitions remains poorly understood. Here, we integrated single-nucleus RNA and chromatin accessibility multi-ome profiles from over one million cells across primary IDH-wildtype GBs and developed scDORI, a scalable deep-learning framework to infer enhancer-driven gene regulatory networks (eGRNs) at single-cell resolution. Our analysis revealed a structured hierarchy of GB cell states governed by distinct regulatory programs, with marked variability in epigenetic plasticity that enables or constrains transitions. Neuronal-like tumor cells emerge as a low plasticity state that deploys active repression, in contrast to more permissive progenitor-like and astrocytic states. We identified the neuronal-like state-specific repressor MYT1L as a key regulator that silences master transcription factors of alternative states. MYT1L gain-of-function in patient-derived GB cells reduced chromatin accessibility, induced neuronal-like identity, and restricted proliferation and invasion in vivo, whereas loss-of-function reactivated plasticity and accelerated malignant features. Our findings delineate the epigenetic architecture and associated transcriptional master regulators that shape GB state trajectories, and establish safeguard repressors such as MYT1L as potential therapeutic targets to constrain malignant plasticity.

cancer biology↗

bioRxiv2025

Synergistic RAS-MAPK and AKT Activation in MYC-Driven Tumors via Adjacent PVT1 Rearrangements

MYC-driven (MYC+) cancers are aggressive and often fatal. MYC dysregulation is a key event in these cancers, but overexpression of MYC alone is not always enough to cause cancer. Plasmocytoma Variant Translocation 1 (PVT1), a long non-coding RNA (lncRNA) adjacent to MYC on chromosome 8 is a rearrangement hotspot in many MYC+ cancers. In addition to being co-amplified with MYC, the genomic rearrangement at PVT1 involves translocation, which has had obscure functional consequences. We report that translocation at the PVT1 locus cause asymmetric enrichment of 5-PVT1 and loss of 3-PVT1. Despite being classified as a non-coding RNA, the retained 5 region of PVT1 generates a circular RNA (CircPVT1) that codes for the novel peptide we call Firefox (FFX). FFX augments AKT signaling and synergistically activates MYC and mTORC1 in these cells. Further, the 3 end of PVT1, which is lost during the translocation, codes for a tumor-suppressing micropeptide we named as Honeybadger (HNB). We demonstrate that HNB interacts with KRAS and disrupts the activation of KRAS effectors. Loss of HNB leads to activation of RAS/MAPK signaling pathway, and enhances MYC stability by promoting phosphorylation of MYC at Ser62. These findings identify PVT1 as a critical node that synchronizes MYC, AKT, and RAS-MAPK activities in cancer. Our study thus identifies a key mechanism by which rearrangements at the PVT1 locus activate additional oncogenic pathways that synergize with MYC to exacerbate the aggressiveness of MYC+ cancers. This newfound understanding explains the poor prognosis associated with MYC+ cancers and offers potential therapeutic targets that could be leveraged in treatment strategies for these cancers.

cancer biology↗

bioRxiv2022

ExplaiNN: interpretable and transparent neural networks for genomics

Sequence-based deep learning models, particularly convolutional neural networks (CNNs), have shown superior performance on a wide range of genomic tasks. A key limitation of these models is the lack of interpretability, slowing down their adoption by the genomics community. Current approaches to model interpretation do not readily reveal how a model makes predictions, can be computationally intensive, and depend on the implemented architecture. Here, we introduce ExplaiNN, an adaptation of neural additive models[1] for genomic tasks wherein predictions are computed as a linear combination of multiple independent CNNs, each consisting of a single convolutional filter and fully connected layers. This approach brings together the expressiveness of CNNs with the interpretability of linear models, providing global (cell state level) as well as local (individual sequence level) biological insights into the data. We use ExplaiNN to predict transcription factor (TF) binding and chromatin accessibility states, demonstrating performance levels comparable to state-of-the-art methods, while providing a transparent view of the models predictions in a straightforward manner. Applied to de novo motif discovery, ExplaiNN identifies equivalent motifs to those obtained from specialized algorithms across a range of datasets. Finally, we present ExplaiNN as a plug-and-play platform in which pretrained TF binding models and annotated position weight matrices from reference databases can be easily combined. We expect that ExplaiNN will accelerate the adoption of deep learning by biological domain experts in their daily genomic sequence analyses.

genomics↗

bioRxiv2020

Biologically-relevant transfer learning improves transcription factor binding prediction

BackgroundDeep learning has proven to be a powerful technique for transcription factor (TF) binding prediction, but requires large training datasets. Transfer learning can reduce the amount of data required for deep learning, while improving overall model performance, compared to training a separate model for each new task. ResultsWe assess a transfer learning strategy for TF binding prediction consisting of a pre-training step, wherein we train a multi-task model with multiple TFs, and a fine-tuning step, wherein we initialize single-task models for individual TFs with the weights learned by the multi-task model, after which the single-task models are trained at a lower learning rate. We corroborate that transfer learning improves model performance, especially if in the pre-training step the multi-task model is trained with biologically-relevant TFs. We show the effectiveness of transfer learning for TFs with [~]500 ChIP-seq peak regions. Using model interpretation techniques, we demonstrate that the features learned in the pre-training step are refined in the fine-tuning step to resemble the binding motif of the target TF (i.e. the recipient of transfer learning in the fine-tuning step). Moreover, pre-training with biologically-relevant TFs allows single-task models in the fine-tuning step to learn features other than the motif of the target TF. ConclusionsOur results confirm that transfer learning is a powerful technique for TF binding prediction.

bioinformatics↗

bioRxiv2019

Sequence-level instructions direct transcription at polyT short tandem repeats.

BackgroundUsing the Cap Analysis of Gene Expression technology, the FANTOM5 consortium provided one of the most comprehensive maps of Transcription Start Sites (TSSs) in several species. Strikingly, ~72% of them could not be assigned to a specific gene and initiate at unconventional regions, outside promoters or enhancers.\n\nResultsHere, we probe these unassigned TSSs and show that, in all species studied, a significant fraction of CAGE peaks initiate at short tandem repeats (STRs) corresponding to homopolymers of thymidines (T). Additional analyse confirm that these CAGEs are truly associated with transcriptionally active chromatin marks. Furthermore, we train a sequence-based deep learning model able to predict CAGE signal at T STRs with high accuracy (~81%) Extracting features learned by this model reveals that transcription at T STRs is mostly directed by STR length but also instructions lying in the downstream sequence. Excitingly, our model also predicts that genetic variants linked to human diseases affect this STR-associated transcription.\n\nConclusionsTogether, our results extend the repertoire of non-coding transcription associated with DNA tandem repeats and complexify STR polymorphism. We also provide a new metric that can be considered in future studies of STR-related complex traits.

genomics↗