Biology subjects

Krishnan, L.

Publications and source records attributed to Krishnan, L..

7 recordsLinked to original sources

bioRxiv2026

Species-specific small models for cell type classification approach the performance of large single cell foundation models

Accurate cross-species cell type classification remains a key evaluation task in single-cell transcriptomics. Recent foundation models trained on millions of single-cell profiles demonstrate great in-distribution and out-of-distribution performance on this task, but their large parameter counts and substantial computational costs limit accessibility and interpretability. Here, we introduce CytoType, a simple and interpretable model for cell type classification that leverages pre-trained ESM-2 protein embeddings of protein-coding transcripts. By learning linear, cell-type-specific weights over transcript embeddings, without relying on gene count information, CytoType achieves F1 scores comparable to or exceeding those of large-scale transformer-based models. We further developed ESM-Cell Embedding (ESM-CE), an even simpler variant that only averages ESM-2 embeddings across expressed genes, which also performs competitively against foundation models. Both CytoType and ESM-CE are trained on species-specific data, maintaining high accuracy when classifying cell types with orders of magnitude fewer parameters compared to larger foundation models. For example, for human tissues, the average performance gap between CytoType and the best foundation model was 0.053 F1 points while CytoType uses 10,000x fewer trainable parameters. Additionally, we quantified the contribution of ESM-2 embeddings to cell type classification tasks and demonstrated a three fold reduction in the performance gap between CytoType and the best foundation model for nine species. Finally, we show that CytoTypes learned gene weights are biologically interpretable.

cell biology↗

bioRxiv2025

VariantFormer: A hierarchical transformer integrating DNA sequences with genetic variations and regulatory landscapes for personalized gene expression prediction

1Accurately predicting gene expression from DNA sequence remains a central challenge in human genetics. Current sequence-based models overlook natural genetic variation across individuals, while population-based models are restricted to variants observed within specific cohorts. Here, we present VariantFormer, a 1.2-billion-parameter transformer that predicts gene-level RNA abundance directly from personalized diploid genomes. Trained on 21,004 genome-transcriptome pairs from 2,330 donors, VariantFormer achieves state-of-the-art performance across both sequence- and population-based prediction tasks, while generalizing better to out-of-distribution contexts--including somatic mutation settings in cancer cell lines--and main-taining robustness across ancestries. Beyond expression prediction, VariantFormer improves eQTL effect size estimation compared to prior methods, with notable gains for lower-frequency and ancestry-specific variants. In applications to Alzheimers disease, VariantFormer gene embeddings prioritize likely causal genes and relevant tissue contexts, and in silico mutagenesis of known APOE alleles faithfully recovers known risk modifying effects. Together, these results establish VariantFormer as a scalable, diploid-aware framework for variant interpretation and personalized gene expression modeling across tissues and populations.

genomics↗

bioRxiv2025

A Cross-Species Generative Cell Atlas Across 1.5 Billion Years ofEvolution: The TranscriptFormer Single-cell Model

Single-cell transcriptomics has revolutionized our understanding of cellular diversity, yet our understanding of the transcriptional programs across the tree of life remains limited. Here we present TranscriptFormer, a family of generative foundation models trained on up to 112 million cells spanning 1.53 billion years of evolution across 12 species. By jointly modeling gene identities and expression levels using a novel generative architecture, TranscriptFormer encodes multi-scale biological structure, functioning as a queryable virtual cell atlas. We demonstrate state-of-the-art performance on both in-distribution and out-of-distribution cell type classification, with robust performance even for species separated by over 685 million years of evolution. TranscriptFormer can also perform zero-shot disease state identification in human cells and accurately transfers cell state annotations across species boundaries. As a generative model, TranscriptFormer can be prompted to predict cell type-specific transcription factors and gene-gene interactions that align with independent experimental observations. Developmental trajectories, phylogenetic relationships and cellular hierarchies emerge naturally in TranscriptFormers representations without any explicit training on these annotations. This work establishes a powerful framework for quantitative single-cell analysis, and comparative cellular biology, thus demonstrating that universal principles of cellular organization can be learned and predicted across the tree of life.

systems biology↗

bioRxiv2020

Mechanical Regulation of Microvascular Angiogenesis

Neovascularization is a critical early step toward successful tissue regeneration during wound healing. While vasculature has long been recognized as highly mechanosensitive (to fluid shear, pulsatile luminal pressure, etc.), the effects of extracellular matrix strains on angiogenesis are poorly understood. Previously, we found that dynamic matrix compression in vivo potently regulated neovascular growth during tissue regeneration; however, whether matrix deformations directly regulate00 angiogenesis remained unknown. Here, we tested the effects of load initiation time, strain magnitude, and mode of compressive deformation (uniform compression vs. compressive indentation that also introduced shear stress) on neovascularization and key angiogenic and mechanotransduction signaling pathways by microvascular fragments in vitro. We hypothesized that neovascularization would be enhanced by delayed, moderate compression and inhibited by early, high magnitude compression and by compressive indentation. Consistent with our hypothesis, early, high magnitude loading inhibited vessel growth, while delayed loading enhanced vessel growth. Compressive indentation led to longer, more branched networks than uniform compression - particularly at high strain magnitude. Gene expression was differentially regulated by time of load initiation; genes associated with active angiogenic sprouts were downregulated by early loading but upregulated by delayed loading. Canonical gene targets of the YAP/TAZ mechanotransduction pathway were increased by loading and abrogated by pharmacological YAP inhibition. Together, these data demonstrate that neovascularization is directly responsive to dynamic matrix strain and is particularly sensitive to the timing of load initiation. This work further identifies putative mechanoregulatory angiogenic mechanisms and implicates a critical role for dynamic mechanical cues in vascularized tissue regeneration. Statement of SignificanceMechanical cues influence tissue regeneration, and although vasculature is known to be mechanically sensitive, remarkably little is known about the effects of bulk extracellular matrix deformation on the nascent vessel networks found in healing tissues. Here, we demonstrated that load initiation time, magnitude, and mode all regulate microvascular growth, as well as upstream angiogenic and mechanotransduction signaling pathways. Across all tested magnitudes and modes, microvascular network formation and upstream signaling were powerfully regulated by the timing of load initiation. This work provides a new foundational understanding of how extracellular matrix mechanics regulate angiogenesis and has critical implications for clinical translation of new regenerative medicine therapies and physical rehabilitation strategies designed to enhance revascularization during tissue regeneration.

bioengineering↗

bioRxiv2019

PGC-1α overexpression partially rescues impaired oxidative and contractile pathophysiology following volumetric muscle loss injury

Volumetric muscle loss (VML) injury is characterized by a non-recoverable loss of muscle fibers due to ablative surgery or severe orthopaedic trauma, that results in chronic functional impairments of the soft tissue. Currently, the effects of VML on the oxidative capacity and adaptability of the remaining injured muscle are unclear. A better understanding of this pathophysiology could significantly shape how VML-injured patients and clinicians approach regenerative medicine and rehabilitation following injury. Herein, the data indicated that VML-injured muscle has diminished mitochondrial content and function (i.e. oxidative capacity), loss of mitochondrial network organization, and attenuated oxidative adaptations to exercise. However, forced PGC-1 over-expression rescued the deficits in oxidative capacity and muscle strength. This implicates physiological activation of PGC1- as a limiting factor in VML-injured muscle adaptive capacity and provides a mechanistic target for regenerative rehabilitation approaches to address the skeletal muscle dysfunction.

physiology↗

bioRxiv2018

Effects of BMP-2 dose and delivery of microvascular fragments on healing of bone defects with concomitant volumetric muscle loss

Traumatic composite bone-muscle injuries, such as open fractures, often require multiple surgical interventions and still typically result in long-term disability. Clinically, a critical indicator of composite injury severity is vascular integrity; vascular damage alone is sufficient to assign an open fracture to the most severe category. Challenging bone injuries are often treated with bone morphogenetic protein 2 (BMP-2), an osteoinductive growth factor, delivered on collagen sponge. Previous studies in a composite defect model found that a minimally bridging dose in the segmental defect model was unable to overcome concomitant muscle damage, but the effect of BMP dose on composite injuries has not yet been studied. Here, we test the hypotheses that BMP-2-mediated functional regeneration of composite extremity injuries is dose dependent and can be further enhanced via co-delivery of adipose-derived microvascular fragments (MVF), which have been previously shown to increase tissue vascular volume. Although MVF treatment did not improve healing outcomes, we observed a significant BMP-2 dose-dependent increase in regenerated bone volume and biomechanical properties. While high dose BMP-2 delivery can induce heterotopic ossification (HO) and increased inflammation, the maximum 10 g dose used in this study did not result in HO and was associated with a lower circulating inflammatory cytokine profile than the low dose (2.5 g) group. These data support the potential benefits of an increased, though still moderate, BMP-2 dose for treatment of bone defects with concomitant muscle damage. Future work to improve vascularization may further improve healing.

bioengineering↗

bioRxiv2017

Wireless Implantable Sensor For Non-Invasive, Longitudinal Quantification Of Axial Strain Across Rodent Long Bone Defects

Bone development, maintenance, and regeneration are remarkably sensitive to mechanical cues. Consequently, mechanical stimulation has long been sought as a putative target to promote endogenous healing after fracture. Given the transient nature of bone repair, tissue-level mechanical cues evolve rapidly over time after injury and are challenging to measure non-invasively. The objective of this work was to develop and characterize an implantable strain sensor for non-invasive monitoring of axial strain across a rodent femoral defect during functional activity. Herein, we present the design, characterization, and in vivo demonstration of the devices capabilities for quantitatively interrogating physiological dynamic strains during bone regeneration. Ex vivo experimental characterization of the device showed that it exceeded the technical requirements for sensitivity, signal resolution, and electromechanical stability. The digital telemetry minimized power consumption, enabling long-term intermittent data collection. Devices were implanted in a rat 6 mm femoral segmental defect model and after three days, data were acquired wirelessly during ambulation and synchronized to corresponding radiographic videos, validating the ability of the sensor to non-invasively measure strain in real-time. Lastly, in vivo strain measurements were utilized in a finite element model to estimate the strain distribution within the defect region. Together, these data indicate the sensor is a promising technology to quantify local tissue mechanics in a specimen specific manner, facilitating more detailed investigations into the role of the mechanical environment in dynamic skeletal healing and remodeling.

bioengineering↗