Kevin Lin
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Kevin Lin has a total of 184 co-authors. Here are the top co-authors by number of shared publications:
- Lai-Sang Young
New York University
9 shared publications Cheryl L. Walker
8 shared publicationsChu-Song Chen
8 shared publications- Jeffrey I. Everitt
Duke University School of Medicine
8 shared publications - K. Leigh Greathouse
Human Science and Design, Robbins College of Health and Human Sciences, Baylor University, Waco, TX, USA
8 shared publications - Kurunthachalam Kannan
New York State Department of Health
8 shared publications
Latest Publications
- 29 September 2026
- Published by SAGE Publications in Journal of Computational Biology
- Vol. 33 (11-12) , 1020-1034
Integrating Morphology and Gene Expression of Neural Cells in Unpaired Single-Cell Data Using GeoAdvAE
Cellular morphological transitions are observed across many diseases, yet their functional role remains unclear because few technologies profile form and function in the same cell. Linking single-cell morphology to transcriptomics is difficult: The two modalities share no feature correspondence and are typically measured in different cells. We present GeoAdvAE, a geometry-aware adversarial autoencoder for diagonal (unpaired) integration of single-cell morphology and single-cell RNA sequencing. GeoAdvAE couples modality-specific variational autoencoders with a Gromov–Wasserstein regularizer and an adversarial discriminator to embed unpaired morphologies and transcriptomes into a shared latent space that preserves both reconstruction fidelity and cross-modal geometry. Using patch-seq neurons with joint morphology-RNA measurements as ground truth, GeoAdvAE attains the best cross-modal cell-type matching accuracy among diagonal integration methods, outperforming optimal-transport, latent-alignment, and adversarial baselines. Applied to 98 CAJAL-quantified microglial morphologies and 31,948 single-cell transcriptomes from the 5xFAD Alzheimer’s disease model, GeoAdvAE recovers a one-dimensional axis that aligns the two modalities. Integrated-gradient attribution highlights transcriptomic shifts (DNA repair in ramified microglia; cell killing in amoeboid microglia), nominates gene markers ( Ms4a6b ; Ftl1 / Fth1 ), and reveals disease-associated microglia signatures that are decoupled from morphology. GeoAdvAE provides a scalable, interpretable approach to connecting cellular “form” and “function” when joint profiling of morphology and transcriptomics is impractical. Our method is publicly available at https://github.com/turbodu222/GeoAdVAE .
- 29 September 2026
- Published by SAGE Publications in Journal of Computational Biology
- Vol. 33 (11-12) , 1020-1034
Integrating Morphology and Gene Expression of Neural Cells in Unpaired Single-Cell Data Using GeoAdvAE
Cellular morphological transitions are observed across many diseases, yet their functional role remains unclear because few technologies profile form and function in the same cell. Linking single-cell morphology to transcriptomics is difficult: The two modalities share no feature correspondence and are typically measured in different cells. We present GeoAdvAE, a geometry-aware adversarial autoencoder for diagonal (unpaired) integration of single-cell morphology and single-cell RNA sequencing. GeoAdvAE couples modality-specific variational autoencoders with a Gromov–Wasserstein regularizer and an adversarial discriminator to embed unpaired morphologies and transcriptomes into a shared latent space that preserves both reconstruction fidelity and cross-modal geometry. Using patch-seq neurons with joint morphology-RNA measurements as ground truth, GeoAdvAE attains the best cross-modal cell-type matching accuracy among diagonal integration methods, outperforming optimal-transport, latent-alignment, and adversarial baselines. Applied to 98 CAJAL-quantified microglial morphologies and 31,948 single-cell transcriptomes from the 5xFAD Alzheimer’s disease model, GeoAdvAE recovers a one-dimensional axis that aligns the two modalities. Integrated-gradient attribution highlights transcriptomic shifts (DNA repair in ramified microglia; cell killing in amoeboid microglia), nominates gene markers ( Ms4a6b ; Ftl1 / Fth1 ), and reveals disease-associated microglia signatures that are decoupled from morphology. GeoAdvAE provides a scalable, interpretable approach to connecting cellular “form” and “function” when joint profiling of morphology and transcriptomics is impractical. Our method is publicly available at https://github.com/turbodu222/GeoAdVAE .
- Open Access
Mostly-monocular responses and other visual functions in a multiscale network model of Macaque V1
Visual signals from the two eyes merge gradually as they pass through the primary visual cortex (V1). Here we use a computational model of Macaque V1 to study the first stage of this integration along the magnocellular pathway, in layer 4C𝛼, aiming to infer neuroanatomical origins of binocular response. It is known that neurons in layer 4C𝛼 are predominantly monocular, though some do exhibit varying degrees of binocularity. We find (1) the emergence of narrow binocular strips along borders of ocular dominance columns (ODC), a finding that aligns with experiments; (2) most consistent with data is when 10-30% of interactions near ODC boundaries are cross-columnar; and (3) feedback from layer 6 is largely monocular. These results were obtained through systematic hypothesis testing using a multiscale model that is orders of magnitude faster than its biologically-detailed predecessors. We propose that multiscale modeling can be an effective tool for bridging anatomy and function.
Open AccessMostly-monocular responses and other visual functions in a multiscale network model of Macaque V1
Visual signals from the two eyes merge gradually as they pass through the primary visual cortex (V1). Here we use a computational model of Macaque V1 to study the first stage of this integration along the magnocellular pathway, in layer 4C𝛼, aiming to infer neuroanatomical origins of binocular response. It is known that neurons in layer 4C𝛼 are predominantly monocular, though some do exhibit varying degrees of binocularity. We find (1) the emergence of narrow binocular strips along borders of ocular dominance columns (ODC), a finding that aligns with experiments; (2) most consistent with data is when 10-30% of interactions near ODC boundaries are cross-columnar; and (3) feedback from layer 6 is largely monocular. These results were obtained through systematic hypothesis testing using a multiscale model that is orders of magnitude faster than its biologically-detailed predecessors. We propose that multiscale modeling can be an effective tool for bridging anatomy and function.
- Open Access
SCOPE: Localizing fate-decision states and their regulatory drivers in single-cell differentiation
Identifying the precise transcriptomic states at which cells commit to a lineage (branchpoints) and the temporal lag in which chromatin accessibility foreshadows gene expression (epigenetic priming) remain fundamental challenges in developmental biology. While current methods for single-cell sequencing data effectively capture developmental flow, they often lack a principled mechanism for delineating the discrete boundaries, a crucial aspect required to map the molecular logic of lineage commitment. We present SCOPE (Semi-supervised Conformal Prediction), a framework that transforms high-dimensional single-cell measurements into rigorous, discrete prediction sets of all plausible future fates. By formalizing fate uncertainty via conformal inference, SCOPE localizes the precise biological windows during which multipotent progenitors specify their fate. In multi-omic data, SCOPE uncovers epigenetic priming and identifies its driving transcription factors by detecting regimes where chromatin-derived prediction sets resolve toward terminal fates significantly before their transcriptomic counterparts. We apply SCOPE across simulations, lineage-traced mouse hematopoiesis, multiple human hematopoietic datasets, and human retinogenesis to demonstrate its broad applicability and ability to recapitulate known fate specification drivers. Ultimately, SCOPE provides a statistically grounded foundation for localizing fate decisions across biological replicates and modalities, offering a robust tool for identifying the onset of lineage specification in complex developmental systems.
Open AccessSCOPE: Localizing fate-decision states and their regulatory drivers in single-cell differentiation
Identifying the precise transcriptomic states at which cells commit to a lineage (branchpoints) and the temporal lag in which chromatin accessibility foreshadows gene expression (epigenetic priming) remain fundamental challenges in developmental biology. While current methods for single-cell sequencing data effectively capture developmental flow, they often lack a principled mechanism for delineating the discrete boundaries, a crucial aspect required to map the molecular logic of lineage commitment. We present SCOPE (Semi-supervised Conformal Prediction), a framework that transforms high-dimensional single-cell measurements into rigorous, discrete prediction sets of all plausible future fates. By formalizing fate uncertainty via conformal inference, SCOPE localizes the precise biological windows during which multipotent progenitors specify their fate. In multi-omic data, SCOPE uncovers epigenetic priming and identifies its driving transcription factors by detecting regimes where chromatin-derived prediction sets resolve toward terminal fates significantly before their transcriptomic counterparts. We apply SCOPE across simulations, lineage-traced mouse hematopoiesis, multiple human hematopoietic datasets, and human retinogenesis to demonstrate its broad applicability and ability to recapitulate known fate specification drivers. Ultimately, SCOPE provides a statistically grounded foundation for localizing fate decisions across biological replicates and modalities, offering a robust tool for identifying the onset of lineage specification in complex developmental systems.
- Open Access
- 13 February 2026
- Published by Oxford University Press (OUP) in Biometrika
- Vol. 113 (2)
Dynamic clustering for heterophilic stochastic block models with time-varying node memberships
Summary: We consider a time-ordered sequence of networks stemming from stochastic block models in which nodes gradually change their membership over time, and no network at any single time-point contains sufficient signal strength to recover its community structure. To estimate the time-varying community structure, we develop the kernel-debiased sum of squares method that performs spectral clustering after a debiased sum-of-squared aggregation of adjacency matrices. Our theory demonstrates, via a novel bias-variance decomposition, that our new method achieves consistent community detection in each network, even for heterophilic networks, without requiring smoothness in the time-varying dynamics of between-community connectivities. We also prove the identifiability of aligning community structures across time based on how rapidly nodes change communities, and we develop a data-adaptive bandwidth-tuning procedure for our new method. We demonstrate the utility and advantages of our proposed method through simulations and a novel analysis of time-varying dynamics in gene coordination in the human developing brain.
- 13 February 2026
- Published by Oxford University Press (OUP) in Biometrika
- Vol. 113 (2)
Open AccessDynamic clustering for heterophilic stochastic block models with time-varying node memberships
Summary: We consider a time-ordered sequence of networks stemming from stochastic block models in which nodes gradually change their membership over time, and no network at any single time-point contains sufficient signal strength to recover its community structure. To estimate the time-varying community structure, we develop the kernel-debiased sum of squares method that performs spectral clustering after a debiased sum-of-squared aggregation of adjacency matrices. Our theory demonstrates, via a novel bias-variance decomposition, that our new method achieves consistent community detection in each network, even for heterophilic networks, without requiring smoothness in the time-varying dynamics of between-community connectivities. We also prove the identifiability of aligning community structures across time based on how rapidly nodes change communities, and we develop a data-adaptive bandwidth-tuning procedure for our new method. We demonstrate the utility and advantages of our proposed method through simulations and a novel analysis of time-varying dynamics in gene coordination in the human developing brain.
- Open Access
- 7 January 2026
- Published by Oxford University Press (OUP) in Bioinformatics Advances
- Vol. 6 (1)
Quantifying stability via count splitting to guide model selection in RNA velocity analyses
Motivation: RNA velocity is a computational framework that predicts future cell states from single-cell RNA sequencing data, offering valuable insights into dynamic biological processes. However, there is a lack of general methods to quantify the uncertainty and stability of these predictions from various RNA velocity methods. Results: We present a novel framework for evaluating RNA velocity stability using negative binomial count splitting to generate independent data replicates, a metric we call replicate coherence. Testing five RNA velocity methods across datasets for mouse erythroid, pancreatic, and human brain development, we identify significant performance differences and inconsistencies, such as reversed velocity flows. Our framework remains robust even when intermediary cell states are missing. Furthermore, we introduce a signal-to-random coherence metric to guide model selection. We demonstrate that selecting fits with high replicate coherence uncovers more biologically informative gene pathways. This broadly applicable approach provides a rigorous tool for assessing and comparing RNA velocity methods across diverse biological contexts. Availability and implementation: The code and analyses are available at https://github.com/linnykos/veloUncertainty.
- 7 January 2026
- Published by Oxford University Press (OUP) in Bioinformatics Advances
- Vol. 6 (1)
Open AccessQuantifying stability via count splitting to guide model selection in RNA velocity analyses
Motivation: RNA velocity is a computational framework that predicts future cell states from single-cell RNA sequencing data, offering valuable insights into dynamic biological processes. However, there is a lack of general methods to quantify the uncertainty and stability of these predictions from various RNA velocity methods. Results: We present a novel framework for evaluating RNA velocity stability using negative binomial count splitting to generate independent data replicates, a metric we call replicate coherence. Testing five RNA velocity methods across datasets for mouse erythroid, pancreatic, and human brain development, we identify significant performance differences and inconsistencies, such as reversed velocity flows. Our framework remains robust even when intermediary cell states are missing. Furthermore, we introduce a signal-to-random coherence metric to guide model selection. We demonstrate that selecting fits with high replicate coherence uncovers more biologically informative gene pathways. This broadly applicable approach provides a rigorous tool for assessing and comparing RNA velocity methods across diverse biological contexts. Availability and implementation: The code and analyses are available at https://github.com/linnykos/veloUncertainty.
- Open Access
Integrating morphology and gene expression of neural cells in unpaired single-cell data using GeoAdvAE
Background: Cellular morphological transitions are observed across many diseases, yet their functional role remains unclear because few technologies profile form and function in the same cell. Linking single-cell morphology to transcriptomics is difficult: the two modalities share no feature correspondence and are typically measured in different cells. Methods: We present GeoAdvAE, a geometry-aware adversarial autoencoder for diagonal (unpaired) integration of single-cell morphology and single-cell RNA sequencing. GeoAdvAE couples modality-specific variational autoencoders with a Gromov-Wasserstein regularizer and an adversarial discriminator to embed unpaired morphologies and transcriptomes into a shared latent space that preserves both reconstruction fidelity and cross-modal geometry. Results: Using patch-seq neurons with joint morphology-RNA measurements as ground truth, GeoAdvAE attains the best cross-modal cell-type matching accuracy among diagonal integration methods, outperforming optimal-transport, latent-alignment, and adversarial baselines. Applied to 98 CAJAL-quantified microglial morphologies and 31,948 single-cell transcriptomes from the 5xFAD Alzheimer’s disease model, GeoAdvAE recovers a one-dimensional axis that aligns the two modalities. Integrated-gradient attribution highlights transcriptomic shifts (DNA repair in ramified microglia; cell killing in amoeboid microglia), nominates gene markers ( Ms4a6b ; Ftl1 / Fth1 ), and reveals disease-associated microglia signatures that are decoupled from morphology. Conclusions: GeoAd-vAE provides a scalable and interpretable approach to connecting cellular “form” and “function” when joint profiling of morphology and transcriptomics is impractical. Our method is publicly available at https://github.com/turbodu222/GeoAdVAE .
Open AccessIntegrating morphology and gene expression of neural cells in unpaired single-cell data using GeoAdvAE
Background: Cellular morphological transitions are observed across many diseases, yet their functional role remains unclear because few technologies profile form and function in the same cell. Linking single-cell morphology to transcriptomics is difficult: the two modalities share no feature correspondence and are typically measured in different cells. Methods: We present GeoAdvAE, a geometry-aware adversarial autoencoder for diagonal (unpaired) integration of single-cell morphology and single-cell RNA sequencing. GeoAdvAE couples modality-specific variational autoencoders with a Gromov-Wasserstein regularizer and an adversarial discriminator to embed unpaired morphologies and transcriptomes into a shared latent space that preserves both reconstruction fidelity and cross-modal geometry. Results: Using patch-seq neurons with joint morphology-RNA measurements as ground truth, GeoAdvAE attains the best cross-modal cell-type matching accuracy among diagonal integration methods, outperforming optimal-transport, latent-alignment, and adversarial baselines. Applied to 98 CAJAL-quantified microglial morphologies and 31,948 single-cell transcriptomes from the 5xFAD Alzheimer’s disease model, GeoAdvAE recovers a one-dimensional axis that aligns the two modalities. Integrated-gradient attribution highlights transcriptomic shifts (DNA repair in ramified microglia; cell killing in amoeboid microglia), nominates gene markers ( Ms4a6b ; Ftl1 / Fth1 ), and reveals disease-associated microglia signatures that are decoupled from morphology. Conclusions: GeoAd-vAE provides a scalable and interpretable approach to connecting cellular “form” and “function” when joint profiling of morphology and transcriptomics is impractical. Our method is publicly available at https://github.com/turbodu222/GeoAdVAE .
- Open Access
- 24 September 2025
- Published by Springer Nature in Research Square
Deciphering Cell Fate and Clonal Dynamics via Integrative Single-Cell Lineage Modeling
Through natural or synthetic lineage barcodes, single-cell technologies now enable the joint measurement of molecular states and clonal identities, providing an unprecedented opportunity to study cell fate and dynamics. Yet, most computational methods for inferring cell development and differentiation rely exclusively on transcriptional similarity, overlooking the lineage information encoded by lineage barcodes. This limitation is exemplified by T cells, where subtle transcriptional differences mark divergent fates with distinct biological activity. Single-cell RNA and matched TCR sequencing is now ubiquitous in the analysis of clinical samples, where the TCR sequence provides an endogenous clonal barcode and could reveal clonal T cell responses. We present Clonotrace, a computational framework that jointly models gene expression and clonotype information to infer cell state transitions and fate biases with higher fidelity. While motivated by challenges in analyzing T cell populations, especially in the tumor microenvironment and immunotherapy settings, Clonotrace is broadly applicable to any lineage-barcoded single-cell dataset. Across diverse systems including T cells, hematopoietic differentiation, and cancer therapy resistance models, Clonotrace reveals differentiation hierarchies, distinguishes unipotent from multipotent states, and identifies candidate fate-determining genes driving lineage commitment.
- 24 September 2025
- Published by Springer Nature in Research Square
Open AccessDeciphering Cell Fate and Clonal Dynamics via Integrative Single-Cell Lineage Modeling
Through natural or synthetic lineage barcodes, single-cell technologies now enable the joint measurement of molecular states and clonal identities, providing an unprecedented opportunity to study cell fate and dynamics. Yet, most computational methods for inferring cell development and differentiation rely exclusively on transcriptional similarity, overlooking the lineage information encoded by lineage barcodes. This limitation is exemplified by T cells, where subtle transcriptional differences mark divergent fates with distinct biological activity. Single-cell RNA and matched TCR sequencing is now ubiquitous in the analysis of clinical samples, where the TCR sequence provides an endogenous clonal barcode and could reveal clonal T cell responses. We present Clonotrace, a computational framework that jointly models gene expression and clonotype information to infer cell state transitions and fate biases with higher fidelity. While motivated by challenges in analyzing T cell populations, especially in the tumor microenvironment and immunotherapy settings, Clonotrace is broadly applicable to any lineage-barcoded single-cell dataset. Across diverse systems including T cells, hematopoietic differentiation, and cancer therapy resistance models, Clonotrace reveals differentiation hierarchies, distinguishes unipotent from multipotent states, and identifies candidate fate-determining genes driving lineage commitment.
- Open Access
- 13 February 2025
- Published by Public Library of Science (PLoS) in PLOS Computational Biology
- Vol. 21 (2) , e1012247
Detection of latent brain states from spontaneous neural activity in the amygdala
The amygdala responds to a large variety of socially and emotionally salient environmental and interoceptive stimuli. The context in which these stimuli occur determines their social and emotional significance. In canonical neurophysiological studies, the fast-paced succession of stimuli and events induce phasic changes in neural activity. During inter-trial intervals, neural activity is expected to return to a stable and featureless level of spontaneous activity, often called baseline. In previous studies we found that context, such as the presence of a social partner, induces brain states that can transcend the fast-paced succession of stimuli and can be recovered from the spontaneous, inter-trial firing rate of neurons. Indeed, the spontaneous firing rates of neurons in the amygdala are different during blocks of gentle grooming touches delivered by a trusted social partner, and during blocks of non-social airflow stimuli delivered by a computer-controlled air valve. Here, we examine local field potentials (LFPs) recorded during periods of spontaneous activity to determine whether information about context can be extracted from these signals. We found that information about social vs. non-social context is present in the local field potential during periods of spontaneous activity between the application of grooming and airflow stimuli, as machine learning techniques can reliably decode context from spectrograms of spontaneous LFPs. No significant differences were detected between the nuclei of the amygdala that receive direct or indirect inputs from areas of the prefrontal cortex known to coordinate flexible, context-dependent behaviors. The lack of nuclear specificity suggests that context-related synaptic inputs arise from a shared source, possibly interoceptive inputs, that signal the physiological state of the body during social and non-social blocks of tactile stimulation.
- 13 February 2025
- Published by Public Library of Science (PLoS) in PLOS Computational Biology
- Vol. 21 (2) , e1012247
Open AccessDetection of latent brain states from spontaneous neural activity in the amygdala
The amygdala responds to a large variety of socially and emotionally salient environmental and interoceptive stimuli. The context in which these stimuli occur determines their social and emotional significance. In canonical neurophysiological studies, the fast-paced succession of stimuli and events induce phasic changes in neural activity. During inter-trial intervals, neural activity is expected to return to a stable and featureless level of spontaneous activity, often called baseline. In previous studies we found that context, such as the presence of a social partner, induces brain states that can transcend the fast-paced succession of stimuli and can be recovered from the spontaneous, inter-trial firing rate of neurons. Indeed, the spontaneous firing rates of neurons in the amygdala are different during blocks of gentle grooming touches delivered by a trusted social partner, and during blocks of non-social airflow stimuli delivered by a computer-controlled air valve. Here, we examine local field potentials (LFPs) recorded during periods of spontaneous activity to determine whether information about context can be extracted from these signals. We found that information about social vs. non-social context is present in the local field potential during periods of spontaneous activity between the application of grooming and airflow stimuli, as machine learning techniques can reliably decode context from spectrograms of spontaneous LFPs. No significant differences were detected between the nuclei of the amygdala that receive direct or indirect inputs from areas of the prefrontal cortex known to coordinate flexible, context-dependent behaviors. The lack of nuclear specificity suggests that context-related synaptic inputs arise from a shared source, possibly interoceptive inputs, that signal the physiological state of the body during social and non-social blocks of tactile stimulation.
- Open Access
Measuring Regulatory Network Inheritance in Dividing Yeast Cells Using Ordinary Differential Equations
Quantifying the inheritance of protein regulation during asymmetric cell division remains a challenge due to the complexity of these systems and the lack of a formal mathematical definition. We introduce ODEinherit, a new statistical framework leveraging ordinary differential equations (ODEs) to measure how much a mother cell’s regulatory network is passed on to its daughters, addressing this gap. ODEin-herit first estimates cell-specific regulatory networks through ODE systems, incorporating novel adjustments for non-oscillatory trajectories. Then, inheritance is quantified by evaluating how well a mother’s regulatory network explains its daughter’s trajectories. We demonstrate that precise quantification of this inheritance relies on pruning and adjustment for the network density. We benchmark ODEinherit on simulated data and apply it to live-cell, time-lapse microscopy data, where we track the expression dynamics of six proteins across 85 dividing S. cerevisiae cells over eight hours. Our results reveal substantial heterogeneity in inheritance rates among mother-daughter pairs, paving the way for applications in cellular stress response and cell-fate prediction studies across generations.
Open AccessMeasuring Regulatory Network Inheritance in Dividing Yeast Cells Using Ordinary Differential Equations
Quantifying the inheritance of protein regulation during asymmetric cell division remains a challenge due to the complexity of these systems and the lack of a formal mathematical definition. We introduce ODEinherit, a new statistical framework leveraging ordinary differential equations (ODEs) to measure how much a mother cell’s regulatory network is passed on to its daughters, addressing this gap. ODEin-herit first estimates cell-specific regulatory networks through ODE systems, incorporating novel adjustments for non-oscillatory trajectories. Then, inheritance is quantified by evaluating how well a mother’s regulatory network explains its daughter’s trajectories. We demonstrate that precise quantification of this inheritance relies on pruning and adjustment for the network density. We benchmark ODEinherit on simulated data and apply it to live-cell, time-lapse microscopy data, where we track the expression dynamics of six proteins across 85 dividing S. cerevisiae cells over eight hours. Our results reveal substantial heterogeneity in inheritance rates among mother-daughter pairs, paving the way for applications in cellular stress response and cell-fate prediction studies across generations.
- Open Access
Quantifying stability via count splitting to guide model selection in RNA velocity analyses
Motivation: RNA velocity is a computational framework that predicts future cell states from single-cell RNA sequencing data, offering valuable insights into dynamic biological processes. However, there is a lack of general methods to quantify the uncertainty and stability of these predictions from various RNA velocity methods. Results: We present a novel framework for evaluating RNA velocity stability using negative binomial count splitting to generate independent data replicates, a metric we call replicate coherence . Testing five RNA velocity methods across datasets for mouse erythroid, pancreatic, and human brain development, we identify significant performance differences and inconsistencies, such as reversed velocity flows. Our framework remains robust even when intermediary cell states are missing. Furthermore, we introduce a signal-to-random coherence metric to guide model selection. We demonstrate that selecting fits with high replicate coherence uncovers more biologically informative gene pathways. This broadly applicable approach provides a rigorous tool for assessing and comparing RNA velocity methods across diverse biological contexts. Availability and implementation: The code and analyses are available at https://github.com/linnykos/veloUncertainty .
Open AccessQuantifying stability via count splitting to guide model selection in RNA velocity analyses
Motivation: RNA velocity is a computational framework that predicts future cell states from single-cell RNA sequencing data, offering valuable insights into dynamic biological processes. However, there is a lack of general methods to quantify the uncertainty and stability of these predictions from various RNA velocity methods. Results: We present a novel framework for evaluating RNA velocity stability using negative binomial count splitting to generate independent data replicates, a metric we call replicate coherence . Testing five RNA velocity methods across datasets for mouse erythroid, pancreatic, and human brain development, we identify significant performance differences and inconsistencies, such as reversed velocity flows. Our framework remains robust even when intermediary cell states are missing. Furthermore, we introduce a signal-to-random coherence metric to guide model selection. We demonstrate that selecting fits with high replicate coherence uncovers more biologically informative gene pathways. This broadly applicable approach provides a rigorous tool for assessing and comparing RNA velocity methods across diverse biological contexts. Availability and implementation: The code and analyses are available at https://github.com/linnykos/veloUncertainty .